diff --git a/.gitattributes b/.gitattributes index 436748a6fc0ae077f95409f24fc85bbeeb9d8ac2..5bdb51df863e579972ad96d945c01a52e87cd70e 100644 --- a/.gitattributes +++ b/.gitattributes @@ -170,3 +170,58 @@ furry-enhancer-ideogram4/example_images/03_image.jpeg filter=lfs diff=lfs merge= furry-enhancer-ideogram4/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text furry-enhancer-ideogram4/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text furry-enhancer-ideogram4/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text +pagani-zonda-r-2009/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text +pagani-zonda-r-2009/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text +pagani-zonda-r-2009/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text +pagani-zonda-r-2009/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text +pagani-zonda-r-2009/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text +pagani-zonda-r-2009/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text +mihoyo-style/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text +mihoyo-style/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text +mihoyo-style/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text +mihoyo-style/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text +mihoyo-style/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text +mihoyo-style/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text +disco-elysium/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text +disco-elysium/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text +disco-elysium/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text +disco-elysium/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text +disco-elysium/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text +disco-elysium/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text +gray-screen-bypass-lora-and-workflow/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text +gray-screen-bypass-lora-and-workflow/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text +gray-screen-bypass-lora-and-workflow/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text +gray-screen-bypass-lora-and-workflow/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text +gray-screen-bypass-lora-and-workflow/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text +gray-screen-bypass-lora-and-workflow/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text +taiwan-bento-packed-lunch/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text +taiwan-bento-packed-lunch/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text +taiwan-bento-packed-lunch/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text +taiwan-bento-packed-lunch/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text +pale-skin-redhead-enhancer/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text +pale-skin-redhead-enhancer/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text +pale-skin-redhead-enhancer/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text +pale-skin-redhead-enhancer/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text +pale-skin-redhead-enhancer/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text +pale-skin-redhead-enhancer/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text +super-robot-style/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text +super-robot-style/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text +super-robot-style/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text +super-robot-style/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text +super-robot-style/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text +super-robot-style/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text +salomon-xt-6-sneakers/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text +salomon-xt-6-sneakers/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text +salomon-xt-6-sneakers/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text +dream-loom/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text +dream-loom/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text +dream-loom/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text +dream-loom/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text +dream-loom/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text +dream-loom/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text +art-the-clown-terrifier/example_images/01_image.jpeg filter=lfs diff=lfs merge=lfs -text +art-the-clown-terrifier/example_images/02_image.jpeg filter=lfs diff=lfs merge=lfs -text +art-the-clown-terrifier/example_images/03_image.jpeg filter=lfs diff=lfs merge=lfs -text +art-the-clown-terrifier/example_images/04_image.jpeg filter=lfs diff=lfs merge=lfs -text +art-the-clown-terrifier/example_images/05_image.jpeg filter=lfs diff=lfs merge=lfs -text +art-the-clown-terrifier/example_images/06_image.jpeg filter=lfs diff=lfs merge=lfs -text diff --git a/art-the-clown-terrifier/4rtcl0wn_id4.safetensors b/art-the-clown-terrifier/4rtcl0wn_id4.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ff167b54a552e198ec0207ed6d32535b14126d77 --- /dev/null +++ b/art-the-clown-terrifier/4rtcl0wn_id4.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d6794f4b0c0866d9e6adb5be781c2ebd82ad268d5c4e31b82d5f3710cc0ccb0a +size 1123105152 diff --git a/art-the-clown-terrifier/example_images/01_image.jpeg b/art-the-clown-terrifier/example_images/01_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..2992801161a2606b61d1fe12664eddb961c2b130 --- /dev/null +++ b/art-the-clown-terrifier/example_images/01_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:92ced69d8f46c5544a5ea3b9afa18cfa6ac9acbc65f2facf5e56de0a1c5a6fd0 +size 3970586 diff --git a/art-the-clown-terrifier/example_images/02_image.jpeg b/art-the-clown-terrifier/example_images/02_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..e8d73d976efd4700477c72bb771a238851b559c1 --- /dev/null +++ b/art-the-clown-terrifier/example_images/02_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2c722613d65113c3872714e23c30758a3c12279ab35fe3680b0c044b185cf4a7 +size 4598743 diff --git a/art-the-clown-terrifier/example_images/03_image.jpeg b/art-the-clown-terrifier/example_images/03_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..8212b7059d173182fe253dff7771ea24d6326629 --- /dev/null +++ b/art-the-clown-terrifier/example_images/03_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3e6fbee24c9b16b8fa315ce4a0628968f06d0340eb0a2dd0d826b88545c6e8e8 +size 3750501 diff --git a/art-the-clown-terrifier/example_images/04_image.jpeg b/art-the-clown-terrifier/example_images/04_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..1693d5bc01191feabdd4f5b3c01397e5aeffb1b9 --- /dev/null +++ b/art-the-clown-terrifier/example_images/04_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e753c32d860ccfe330717c7a67a82820f40abefac229a315e2eae989d65fe249 +size 3982126 diff --git a/art-the-clown-terrifier/example_images/05_image.jpeg b/art-the-clown-terrifier/example_images/05_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..7062f8915cd22d5f6ce6cb7dd9f2cab100c594fb --- /dev/null +++ b/art-the-clown-terrifier/example_images/05_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b7932858981b3d2f13189e0c59ab1c7fe1aa61ff5b1f3f9f955b976a944db2e7 +size 3032783 diff --git a/art-the-clown-terrifier/example_images/06_image.jpeg b/art-the-clown-terrifier/example_images/06_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..8ebf8eddb4842460e65f911220570778fac9d0c5 --- /dev/null +++ b/art-the-clown-terrifier/example_images/06_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b74f76929507672f3fa3369dd661b046fd5ce6a9dc0e68d96e717f5b830c9ac1 +size 3779275 diff --git a/art-the-clown-terrifier/info.txt b/art-the-clown-terrifier/info.txt new file mode 100644 index 0000000000000000000000000000000000000000..b18f3651bd6a908cb8c3948740eaa44b5fcaa417 --- /dev/null +++ b/art-the-clown-terrifier/info.txt @@ -0,0 +1,270 @@ +═══════════════════════════════════════════════════════════════════════════════ + ART THE CLOWN - TERRIFIER +═══════════════════════════════════════════════════════════════════════════════ + +─────────────────────────────────────────────────────────────────────────────── + IDENTITY +─────────────────────────────────────────────────────────────────────────────── + Base model Ideogram 4 + Generation mode not declared by the author + Checkpoint not determined — verify before use + Category Character / subject + Author Plaguekind + Source https://civitai.com/models/2763837 + Provenance CivitAI · model 2763837 · version 3116134 + +─────────────────────────────────────────────────────────────────────────────── + WEIGHTS +─────────────────────────────────────────────────────────────────────────────── + File 4rtcl0wn_id4.safetensors + Version Ideogram4 + Size 1,071.1 MB + Format SafeTensor / unknown precision + SHA256 d6794f4b0c0866d9e6adb5be781c2ebd82ad268d5c4e31b82d5f3710cc0ccb0a + Published 2026-07-10 + +─────────────────────────────────────────────────────────────────────────────── + TRIGGER WORDS +─────────────────────────────────────────────────────────────────────────────── + No trigger word. The LoRA applies as soon as it is loaded — + strength is the only way to dial the effect up or down. + +─────────────────────────────────────────────────────────────────────────────── + RECOMMENDED SETTINGS +─────────────────────────────────────────────────────────────────────────────── + The author did not publish specific settings. + No author settings were published. Community baseline: steps 48 is the + reference default: docs/inference.md says "`V4_QUALITY_48` is the default" + and the pipeline's `num_steps` default is 48; diffusers documents the same + 48-step default. 20 is the practical ComfyUI default — the `V4_DEFAULT_20` + preset and the value the shipped template `image_ideogram4_t2i.json` + selects (its preset combo defaults to "Default"). 12 = `V4_TURBO_12`. With + the TurboTime LoRA: 8 in the nomadoor community workflow; the model card + claims usable output down to 2., cfg A guidance SCHEDULE, not one number, + and the negative branch comes from a second model rather than a negative + prompt. Verified from src/ideogram4/sampler_configs.py: V4_QUALITY_48 = 45 + steps @ gw 7 then 3 polish steps @ gw 3; V4_DEFAULT_20 = 18 @ 7 + 2 @ 3; + V4_TURBO_12 = 11 @ 7 + 1 @ 3. Pipeline default `guidance_scale` = 7.0 when + no schedule is given. In ComfyUI the shipped template sets + `DualModelGuider` cfg 7 plus a `CFGOverride` node with widgets [3, 0.7, 1] + (cfg 3 from 0.7→1.0 of the run); kijai recommends starting the override at + 0.7, not 0.9. TurboTime: cfg 1.0 / no CFG., sampler Euler flow-matching. In + ComfyUI: `KSamplerSelect` → euler, driven by `SamplerCustomAdvanced`, with + the model-specific `Ideogram4Scheduler` (logit-normal sigma schedule). Its + params are `mu` and `std`, not the usual `shift`. Reference presets: mu 0.0 + / std 1.5 (48-step), mu 0.0 / std 1.75 (20-step), mu 0.5 / std 1.75 + (12-step). The Ideogram4Scheduler node's own widget defaults are steps 20, + mu 0.5, std 1.75, but the shipped template overrides mu/std/steps from the + preset JSON, so out of the box it runs 20 / mu 0.0 / std 1.75. mu is auto- + shifted by resolution in both implementations: mean = mu + + 0.5*log((w*h)/(512*512))., resolution 1024×1024 default (pipeline + `height`/`width` default 1024). Any multiple of 16 per side from 256 to + 2048, aspect ratios up to 6:1 or 1:6 — one model covers all of it. + 2048×2048 with V4_QUALITY_48 for maximum quality. Buckets documented in + docs/inference.md: 1536×1024 (3:2), 1024×1536 (2:3), 1920×1088 (~16:9), + 2048×768 (~21:9), 1024×1792 (~9:16), 1600×400 (4:1)., lora_strength No + single official number; these are author-stated values. DeverStyle Archer: + 0.6, and explicitly "on the model only, not on the model_negative". + zjourney v2: 0.55 standalone (range 0.35–0.6), or v1 @ 0.35 + v2 @ 0.45 + stacked. Furry Enhancer: the model page says 0.6–0.8, but its own version + notes ask for 0.7–0.9 (v1.94) and 0.8–0.95 (v2.8). TurboTime: 0.6 in the + nomadoor workflow (the author gives no figure). So ~0.5–0.9 is the + realistic band, read the specific card.. + +─────────────────────────────────────────────────────────────────────────────── + ADOPTION & POPULARITY +─────────────────────────────────────────────────────────────────────────────── + Downloads 221 + Rating 100% positive (24/24) + Comments 0 + Published 57 days ago + Download rate 0.8/day since release + + Adoption rank #29 of 35 in this collection (Niche) + Momentum rank #25 of 35 by download rate (Long tail) + + Adoption tier is the percentile of total downloads within this collection; + momentum is the percentile of downloads-per-day since release. Momentum is + ranked rather than measured against a fixed rate, because this base model + is itself new and every LoRA looks fast on an absolute scale. Both numbers + are CivitAI's, read on the scrape date at the foot of this file. + +─────────────────────────────────────────────────────────────────────────────── + CHAINING / STACKING +─────────────────────────────────────────────────────────────────────────────── + Stacks with others not stated + + The author gave no chaining guidance. The rules below are the + general Ideogram 4 ones, not anything specific to this LoRA. + + General Ideogram 4 rules that apply here (see REFERENCE.txt): + + • Chain `LoraLoaderModelOnly` nodes in series off the CONDITIONAL model + loader, ahead of the guider. DeverStyle states its samples were made + "by setting the LORA @ 0.6 strength on the model only, not on the + model_negative". + • Model-only loaders throughout. The text encoder is a frozen Qwen3-VL-8B + with no LoRA-patchable CLIP branch, so the model+clip `LoraLoader` has + nothing useful to do. + • To affect the unconditional/negative branch you must load a second, + physically distinct model file and patch that one. Passing the same + patched model object into both DualModelGuider inputs is a verified no- + op (Comfy-Org issue #14521; kijai confirmed it is expected behaviour). + • Trigger words go INSIDE the JSON caption, in the semantically matching + field, not prepended to a prose string. DeverStyle instructs putting + `dvr_arch_tv` in `style_description.art_style` (and in the content + field); the Kiki LoRA card describes using `art_style: "Studio Ghibli + style"` with `medium: "anime_screencap"`. + • Two versions of the same LoRA can be stacked at reduced strengths: + zjourney's author recommends v1 @ 0.35 + v2 @ 0.45 over either alone at + full strength. + • TurboTime changes the graph, not just the strength: it runs at cfg 1.0 + and the reference community workflow swaps `DualModelGuider` for a + single-model `CFGGuider`, dropping the unconditional model. Re-tune any + stacked style LoRA downward. + • A LoRA can be applied to only part of the run. The Gray Screen bypass + LoRA is used at NEGATIVE weight (about -0.25) and, in its shipped + workflows, only on the first step and then disabled; its author warns + that "LoRA being active through the whole generation could lead to + degradation". + • The author never states which weight family this targets. Silent + degradation, not hard errors, in every case observed. (1) Loading a + conditional-trained LoRA onto the unconditional transformer succeeds + silently because the architectures match; it perturbs the negative + branch instead of the positive one. (2) Verified ComfyUI trap: in `Du + +─────────────────────────────────────────────────────────────────────────────── + PROS +─────────────────────────────────────────────────────────────────────────────── + (none recorded) + +─────────────────────────────────────────────────────────────────────────────── + CONS & CAVEATS +─────────────────────────────────────────────────────────────────────────────── + - No trigger word — the LoRA is always on once loaded, so strength is the + only control. [source: model metadata] + - Author never states which generation mode it was trained for; test + against your own workflow before relying on it. [source: description] + - Large at 1071 MB — slower to load and heavier to stack. [source: file + size] + - Modest adoption (221 downloads) — less community feedback to rely on. + [source: source stats] + +─────────────────────────────────────────────────────────────────────────────── + COMPATIBILITY WARNING +─────────────────────────────────────────────────────────────────────────────── + Silent degradation, not hard errors, in every case observed. (1) Loading a + conditional-trained LoRA onto the unconditional transformer succeeds + silently because the architectures match; it perturbs the negative branch + instead of the positive one. (2) Verified ComfyUI trap: in + `DualModelGuider`, passing the SAME model object as both the conditional + model and the LoRA-patched unconditional model is a no-op. Comfy-Org issue + #14521 tabulates it — case 3 (DualModelGuider, conditional model as + positive, conditional model + LoRA as negative) was "exactly identical" to + plain conditional-only CFG, while case 4 with the real unconditional model + differed. kijai confirmed it is expected: "there's no (working) mechanic to + swap LoRA on each step on the same model, it would only work if the model + was a different one." To patch the negative branch you must load a + genuinely separate model file. (3) A diffusers/PEFT-prefixed LoRA in + ComfyUI is expected to produce only a "lora key not loaded" console warning + and no visible effect — this follows from the absent key mapping (see + checkpoint_families) rather than from a published Ideogram-4 bug report. + (4) Cross-quantization loading (fp8-trained LoRA onto nvfp4/nf4/GGUF) + generally works with some precision loss; int8_convrot had a specific, now- + fixed bug (see gotchas). + +─────────────────────────────────────────────────────────────────────────────── + AUTHOR'S DESCRIPTION +─────────────────────────────────────────────────────────────────────────────── + Lokr 4 of Art the clown from the terrifier series. + Trigger word 4rtcl0wn + Trained on 100 hd images at 1024 exclusively. + Some expressions to try. + Expressionless + Sad + Bored + Sarcastic + Baring teeth + Smiling + Laughing + Covered in blood + + NOT trained on santa clause outfit images. + Use darkbeast aggressive edition for gore. + If you find my work useful and want to support development: Monero (XMR): 8 + 65BrcfWLdwELwuq5faV1uVTbh93zVK6AUYLY2c3mX6sFfAGRfS6axe1kBTYYKuM7ccN7zBZDAZv + nT7E4NKmUazySdbpc7p Thank you for your support. + +─────────────────────────────────────────────────────────────────────────────── + EXAMPLE MEDIA +─────────────────────────────────────────────────────────────────────────────── + 6 preview(s) in ./example_images/ — community results for + this LoRA. Videos keep a matching _poster.jpg still frame. + + No prompts are recorded below: CivitAI's public API flags these posts as + having generation metadata but does not return it, so the prompts cannot be + scraped. Open the model page to read them. + + 01_image.jpeg (image, 1536x2048) + 02_image.jpeg (image, 1344x2368) + 03_image.jpeg (image, 1536x2048) + 04_image.jpeg (image, 1536x2048) + 05_image.jpeg (image, 1248x1664) + 06_image.jpeg (image, 1536x2048) + +─────────────────────────────────────────────────────────────────────────────── + HOW TO USE (COMFYUI) +─────────────────────────────────────────────────────────────────────────────── + Use ComfyUI's NATIVE Ideogram 4 support (Comfy-Org/ComfyUI PR #14259 "feat: + Add Ideogram4 model support", merged 2026-06-03) — not the official + `ComfyUI-Ideogram4` custom-node wrapper, whose nodes (Magic Prompt / + Pipeline Loader / Generate) wrap Ideogram's reference pipeline and contain + no LoRA input at all (verified: no occurrence of "lora" in its nodes.py). + + 1. Update ComfyUI to a build containing `comfy_extras/nodes_ideogram4.py` + (which supplies `Ideogram4Scheduler`); the same PR added the generic + `DualModelGuider` and `CFGOverride` nodes, which live in + `comfy_extras/nodes_custom_sampler.py`. + 2. From `Comfy-Org/Ideogram-4` download into + `ComfyUI/models/diffusion_models/`: `ideogram4_fp8_scaled.safetensors` + (9.28 GB) AND `ideogram4_unconditional_fp8_scaled.safetensors` (9.28 GB). + Lower-VRAM alternatives in the same folder: `*_nvfp4_mixed.safetensors` + (5.49 GB each) or `*_int8_convrot.safetensors` (9.58 GB each). + 3. Download `qwen3vl_8b_fp8_scaled.safetensors` (10.59 GB; or + `qwen3vl_8b_nvfp4.safetensors`, 6.31 GB) into + `ComfyUI/models/text_encoders/`. The shipped template also lists an + optional `gemma4_e4b_it_fp8_scaled.safetensors` from `Comfy-Org/gemma-4` + for local prompt building. + 4. Download `flux2-vae.safetensors` (0.34 GB, from Comfy-Org/Ideogram-4 or + Comfy-Org/flux2-dev) into `ComfyUI/models/vae/`. + 5. Put LoRA .safetensors in `ComfyUI/models/loras/` (no subfolder + convention enforced). + 6. Load the built-in template `image_ideogram4_t2i` (Workflow → Browse + Templates), or the workflow from + docs.comfy.org/tutorials/image/ideogram/ideogram-v4. + 7. Insert a `LoraLoaderModelOnly` between the main `UNETLoader` and the + `DualModelGuider`'s conditional model input. There is no CLIP-side LoRA + here — the text encoder is a frozen Qwen3-VL — so model-only is the right + loader. Strength ~0.5–0.8. + 8. Do NOT reuse one patched model object for both guider inputs; it + silently does nothing. Patching the negative branch requires a second, + physically separate model file with its own loader. + 9. Write the prompt as a JSON caption into `CLIPTextEncode` (the CLIPLoader + type is `ideogram4`); keep `ConditioningZeroOut` feeding the unconditional + branch. + + Non-ComfyUI paths: diffusers — + `Ideogram4Pipeline.from_pretrained("ideogram-ai/ideogram-4-nf4")` (the + docs' own example; `ideogram-ai/ideogram-4-nf4-diffusers` also exists), + then `pipeline.load_lora_weights(...)`. Training — `ostris/ai-toolkit` + (built-in `extensions_built_in/diffusion_models/ideogram4/` plus an + `Ideogram4Captioner` that writes JSON captions) or diffusers + `examples/dreambooth/train_dreambooth_lora_ideogram4.py`. The official HF + weights are GATED (`gated: auto`): accept the license on the model page and + authenticate (`hf auth login` / `HF_TOKEN`) or downloads fail. + +─────────────────────────────────────────────────────────────────────────────── + Scraped by lorakit on 2026-09-05 01:50 UTC. + Stats and description are the author's; pros/cons are derived from the + evidence tagged beside each line. +─────────────────────────────────────────────────────────────────────────────── diff --git a/art-the-clown-terrifier/metadata.json b/art-the-clown-terrifier/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..d688fa0e1ee6c33d3e244a7a3f684d0061291172 --- /dev/null +++ b/art-the-clown-terrifier/metadata.json @@ -0,0 +1,131 @@ +{ + "name": "Art The Clown - Terrifier", + "slug": "art-the-clown-terrifier", + "base_model": "Ideogram 4", + "media_type": "image", + "model_id": 2763837, + "version_id": 3116134, + "version_name": "Ideogram4", + "source": "civitai", + "url": "https://civitai.com/models/2763837", + "repo_id": null, + "file_path": null, + "author": "Plaguekind", + "category": { + "key": "character", + "label": "Character / subject" + }, + "modalities": [], + "primary_modality": "unspecified", + "checkpoint_family": "unknown", + "trigger_words": [], + "recommended_settings": {}, + "chaining": { + "author": { + "stacks": "not stated", + "notes": "", + "avoid": [], + "requires": [], + "order": "", + "strength": "", + "stated": false + }, + "general_rules": [ + "Chain `LoraLoaderModelOnly` nodes in series off the CONDITIONAL model loader, ahead of the guider. DeverStyle states its samples were made \"by setting the LORA @ 0.6 strength on the model only, not on the model_negative\".", + "Model-only loaders throughout. The text encoder is a frozen Qwen3-VL-8B with no LoRA-patchable CLIP branch, so the model+clip `LoraLoader` has nothing useful to do.", + "To affect the unconditional/negative branch you must load a second, physically distinct model file and patch that one. Passing the same patched model object into both DualModelGuider inputs is a verified no-op (Comfy-Org issue #14521; kijai confirmed it is expected behaviour).", + "Trigger words go INSIDE the JSON caption, in the semantically matching field, not prepended to a prose string. DeverStyle instructs putting `dvr_arch_tv` in `style_description.art_style` (and in the content field); the Kiki LoRA card describes using `art_style: \"Studio Ghibli style\"` with `medium: \"anime_screencap\"`.", + "Two versions of the same LoRA can be stacked at reduced strengths: zjourney's author recommends v1 @ 0.35 + v2 @ 0.45 over either alone at full strength.", + "TurboTime changes the graph, not just the strength: it runs at cfg 1.0 and the reference community workflow swaps `DualModelGuider` for a single-model `CFGGuider`, dropping the unconditional model. Re-tune any stacked style LoRA downward.", + "A LoRA can be applied to only part of the run. The Gray Screen bypass LoRA is used at NEGATIVE weight (about -0.25) and, in its shipped workflows, only on the first step and then disabled; its author warns that \"LoRA being active through the whole generation could lead to degradation\".", + "The author never states which weight family this targets. Silent degradation, not hard errors, in every case observed. (1) Loading a conditional-trained LoRA onto the unconditional transformer succeeds silently because the architectures match; it perturbs the negative branch instead of the positive one. (2) Verified ComfyUI trap: in `Du" + ] + }, + "pros": [], + "cons": [ + "No trigger word — the LoRA is always on once loaded, so strength is the only control. [source: model metadata]", + "Author never states which generation mode it was trained for; test against your own workflow before relying on it. [source: description]", + "Large at 1071 MB — slower to load and heavier to stack. [source: file size]", + "Modest adoption (221 downloads) — less community feedback to rely on. [source: source stats]" + ], + "file": { + "name": "4rtcl0wn_id4.safetensors", + "size_bytes": 1123105152, + "size_mb": 1071.08, + "sha256": "d6794f4b0c0866d9e6adb5be781c2ebd82ad268d5c4e31b82d5f3710cc0ccb0a" + }, + "stats": { + "downloads": 221, + "thumbs_up": 24, + "thumbs_down": 0 + }, + "popularity": { + "downloads": 221, + "thumbs_up": 24, + "thumbs_down": 0, + "comments": 0, + "approval": "100% positive (24/24)", + "age_days": 56.6, + "downloads_per_day": 0.8, + "rank_by_downloads": 29, + "rank_by_momentum": 25, + "of_total": 35, + "adoption_tier": "Niche", + "momentum": "Long tail" + }, + "selection_score": 509.0, + "selection_reasons": [ + "221 downloads", + "24 thumbs-up", + "top-ranked for its category" + ], + "example_media": [ + { + "file": "01_image.jpeg", + "type": "image", + "resolution": "1536x2048", + "prompt": "" + }, + { + "file": "02_image.jpeg", + "type": "image", + "resolution": "1344x2368", + "prompt": "" + }, + { + "file": "03_image.jpeg", + "type": "image", + "resolution": "1536x2048", + "prompt": "" + }, + { + "file": "04_image.jpeg", + "type": "image", + "resolution": "1536x2048", + "prompt": "" + }, + { + "file": "05_image.jpeg", + "type": "image", + "resolution": "1248x1664", + "prompt": "" + }, + { + "file": "06_image.jpeg", + "type": "image", + "resolution": "1536x2048", + "prompt": "" + } + ], + "tags": [ + "terrifier", + "character", + "krea2", + "krea", + "villian", + "horror", + "clown" + ], + "scraped_at": "2026-09-05T01:50:16+00:00", + "path": "ideogram-4/art-the-clown-terrifier" +} diff --git a/disco-elysium/MixedElysium_Ideogram.safetensors b/disco-elysium/MixedElysium_Ideogram.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f6f8582207d7e26fa73ddaf1d85853908ad818c3 --- /dev/null +++ b/disco-elysium/MixedElysium_Ideogram.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e762d7696337f6e50f78a7316a37ef2bcea847ab9a712f5ac5f1fa91697c0caa +size 211734080 diff --git a/disco-elysium/example_images/01_image.jpeg b/disco-elysium/example_images/01_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..e36c862cb494709d649cc417f0f35e836c8766a2 --- /dev/null +++ b/disco-elysium/example_images/01_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:987c44f4f5e1d87231bacbf9c6d1ecc0f7679b4a9c82d1c735f075e71271c338 +size 1971028 diff --git a/disco-elysium/example_images/02_image.jpeg b/disco-elysium/example_images/02_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..066461c4e0765b2842d821b45210d41510c00be9 --- /dev/null +++ b/disco-elysium/example_images/02_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e19eabc607ff5503b93af7a3aca5fb1100a4f8abe25314cbc5c7eb43c43982f9 +size 2351161 diff --git a/disco-elysium/example_images/03_image.jpeg b/disco-elysium/example_images/03_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..246703e16129f6787e57a5a90652fa34ee6ffd08 --- /dev/null +++ b/disco-elysium/example_images/03_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:77d607ee184bff4f273c85586d15a1478b0107db424d10f2287ff74f0cf1cbf7 +size 1540713 diff --git a/disco-elysium/example_images/04_image.jpeg b/disco-elysium/example_images/04_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..ab60756d62b2ef5e53d6208aa04fc0e2ab1cb8df --- /dev/null +++ b/disco-elysium/example_images/04_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:45748c80296f954672220c7a95146ca0d7d3fe9cbbc94f1da5ae17d4a4d48a0f +size 1845807 diff --git a/disco-elysium/example_images/05_image.jpeg b/disco-elysium/example_images/05_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..6c3dedb21b7f6628f9e38c3a38b7e2d7d31845e4 --- /dev/null +++ b/disco-elysium/example_images/05_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:838edc3a8fe9245f5c13dc0ba36b5693b5ad7a89df8ba4ba728becaa88ec8222 +size 1802336 diff --git a/disco-elysium/example_images/06_image.jpeg b/disco-elysium/example_images/06_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..c7e84d4dadc9a713034ae8deb8ba38dd6635d30a --- /dev/null +++ b/disco-elysium/example_images/06_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:98a1090395d4b26e734af01fdf320efa534c4d47df27376fef3325d1cf25e7c4 +size 1255244 diff --git a/disco-elysium/info.txt b/disco-elysium/info.txt new file mode 100644 index 0000000000000000000000000000000000000000..40c592bcd74f5de9ad1e3cd4bc0f0623ca2b0078 --- /dev/null +++ b/disco-elysium/info.txt @@ -0,0 +1,265 @@ +═══════════════════════════════════════════════════════════════════════════════ + DISCO ELYSIUM +═══════════════════════════════════════════════════════════════════════════════ + +─────────────────────────────────────────────────────────────────────────────── + IDENTITY +─────────────────────────────────────────────────────────────────────────────── + Base model Ideogram 4 + Generation mode not declared by the author + Checkpoint not determined — verify before use + Category Visual style + Author wystanbelov785 + Source https://civitai.com/models/2487755 + Provenance CivitAI · model 2487755 · version 3129598 + +─────────────────────────────────────────────────────────────────────────────── + WEIGHTS +─────────────────────────────────────────────────────────────────────────────── + File MixedElysium_Ideogram.safetensors + Version Ideogram + Size 201.9 MB + Format SafeTensor / unknown precision + SHA256 e762d7696337f6e50f78a7316a37ef2bcea847ab9a712f5ac5f1fa91697c0caa + Published 2026-07-14 + +─────────────────────────────────────────────────────────────────────────────── + TRIGGER WORDS +─────────────────────────────────────────────────────────────────────────────── + No trigger word. The LoRA applies as soon as it is loaded — + strength is the only way to dial the effect up or down. + +─────────────────────────────────────────────────────────────────────────────── + RECOMMENDED SETTINGS +─────────────────────────────────────────────────────────────────────────────── + The author did not publish specific settings. + No author settings were published. Community baseline: steps 48 is the + reference default: docs/inference.md says "`V4_QUALITY_48` is the default" + and the pipeline's `num_steps` default is 48; diffusers documents the same + 48-step default. 20 is the practical ComfyUI default — the `V4_DEFAULT_20` + preset and the value the shipped template `image_ideogram4_t2i.json` + selects (its preset combo defaults to "Default"). 12 = `V4_TURBO_12`. With + the TurboTime LoRA: 8 in the nomadoor community workflow; the model card + claims usable output down to 2., cfg A guidance SCHEDULE, not one number, + and the negative branch comes from a second model rather than a negative + prompt. Verified from src/ideogram4/sampler_configs.py: V4_QUALITY_48 = 45 + steps @ gw 7 then 3 polish steps @ gw 3; V4_DEFAULT_20 = 18 @ 7 + 2 @ 3; + V4_TURBO_12 = 11 @ 7 + 1 @ 3. Pipeline default `guidance_scale` = 7.0 when + no schedule is given. In ComfyUI the shipped template sets + `DualModelGuider` cfg 7 plus a `CFGOverride` node with widgets [3, 0.7, 1] + (cfg 3 from 0.7→1.0 of the run); kijai recommends starting the override at + 0.7, not 0.9. TurboTime: cfg 1.0 / no CFG., sampler Euler flow-matching. In + ComfyUI: `KSamplerSelect` → euler, driven by `SamplerCustomAdvanced`, with + the model-specific `Ideogram4Scheduler` (logit-normal sigma schedule). Its + params are `mu` and `std`, not the usual `shift`. Reference presets: mu 0.0 + / std 1.5 (48-step), mu 0.0 / std 1.75 (20-step), mu 0.5 / std 1.75 + (12-step). The Ideogram4Scheduler node's own widget defaults are steps 20, + mu 0.5, std 1.75, but the shipped template overrides mu/std/steps from the + preset JSON, so out of the box it runs 20 / mu 0.0 / std 1.75. mu is auto- + shifted by resolution in both implementations: mean = mu + + 0.5*log((w*h)/(512*512))., resolution 1024×1024 default (pipeline + `height`/`width` default 1024). Any multiple of 16 per side from 256 to + 2048, aspect ratios up to 6:1 or 1:6 — one model covers all of it. + 2048×2048 with V4_QUALITY_48 for maximum quality. Buckets documented in + docs/inference.md: 1536×1024 (3:2), 1024×1536 (2:3), 1920×1088 (~16:9), + 2048×768 (~21:9), 1024×1792 (~9:16), 1600×400 (4:1)., lora_strength No + single official number; these are author-stated values. DeverStyle Archer: + 0.6, and explicitly "on the model only, not on the model_negative". + zjourney v2: 0.55 standalone (range 0.35–0.6), or v1 @ 0.35 + v2 @ 0.45 + stacked. Furry Enhancer: the model page says 0.6–0.8, but its own version + notes ask for 0.7–0.9 (v1.94) and 0.8–0.95 (v2.8). TurboTime: 0.6 in the + nomadoor workflow (the author gives no figure). So ~0.5–0.9 is the + realistic band, read the specific card.. + +─────────────────────────────────────────────────────────────────────────────── + ADOPTION & POPULARITY +─────────────────────────────────────────────────────────────────────────────── + Downloads 251 + Rating 100% positive (22/22) + Comments 0 + Published 52 days ago + Download rate 0.7/day since release + + Adoption rank #26 of 35 in this collection (Modest) + Momentum rank #27 of 35 by download rate (Long tail) + + Adoption tier is the percentile of total downloads within this collection; + momentum is the percentile of downloads-per-day since release. Momentum is + ranked rather than measured against a fixed rate, because this base model + is itself new and every LoRA looks fast on an absolute scale. Both numbers + are CivitAI's, read on the scrape date at the foot of this file. + +─────────────────────────────────────────────────────────────────────────────── + CHAINING / STACKING +─────────────────────────────────────────────────────────────────────────────── + Stacks with others not stated + + The author gave no chaining guidance. The rules below are the + general Ideogram 4 ones, not anything specific to this LoRA. + + General Ideogram 4 rules that apply here (see REFERENCE.txt): + + • Chain `LoraLoaderModelOnly` nodes in series off the CONDITIONAL model + loader, ahead of the guider. DeverStyle states its samples were made + "by setting the LORA @ 0.6 strength on the model only, not on the + model_negative". + • Model-only loaders throughout. The text encoder is a frozen Qwen3-VL-8B + with no LoRA-patchable CLIP branch, so the model+clip `LoraLoader` has + nothing useful to do. + • To affect the unconditional/negative branch you must load a second, + physically distinct model file and patch that one. Passing the same + patched model object into both DualModelGuider inputs is a verified no- + op (Comfy-Org issue #14521; kijai confirmed it is expected behaviour). + • Trigger words go INSIDE the JSON caption, in the semantically matching + field, not prepended to a prose string. DeverStyle instructs putting + `dvr_arch_tv` in `style_description.art_style` (and in the content + field); the Kiki LoRA card describes using `art_style: "Studio Ghibli + style"` with `medium: "anime_screencap"`. + • Two versions of the same LoRA can be stacked at reduced strengths: + zjourney's author recommends v1 @ 0.35 + v2 @ 0.45 over either alone at + full strength. + • TurboTime changes the graph, not just the strength: it runs at cfg 1.0 + and the reference community workflow swaps `DualModelGuider` for a + single-model `CFGGuider`, dropping the unconditional model. Re-tune any + stacked style LoRA downward. + • A LoRA can be applied to only part of the run. The Gray Screen bypass + LoRA is used at NEGATIVE weight (about -0.25) and, in its shipped + workflows, only on the first step and then disabled; its author warns + that "LoRA being active through the whole generation could lead to + degradation". + • The author never states which weight family this targets. Silent + degradation, not hard errors, in every case observed. (1) Loading a + conditional-trained LoRA onto the unconditional transformer succeeds + silently because the architectures match; it perturbs the negative + branch instead of the positive one. (2) Verified ComfyUI trap: in `Du + +─────────────────────────────────────────────────────────────────────────────── + PROS +─────────────────────────────────────────────────────────────────────────────── + + Compact at 202 MB — cheap to stack with other LoRAs. [source: file + size] + +─────────────────────────────────────────────────────────────────────────────── + CONS & CAVEATS +─────────────────────────────────────────────────────────────────────────────── + - Needs a reduced strength; the default 1.0 is likely too strong. + [source: description] + - No trigger word — the LoRA is always on once loaded, so strength is the + only control. [source: model metadata] + - Author never states which generation mode it was trained for; test + against your own workflow before relying on it. [source: description] + - Modest adoption (251 downloads) — less community feedback to rely on. + [source: source stats] + +─────────────────────────────────────────────────────────────────────────────── + COMPATIBILITY WARNING +─────────────────────────────────────────────────────────────────────────────── + Silent degradation, not hard errors, in every case observed. (1) Loading a + conditional-trained LoRA onto the unconditional transformer succeeds + silently because the architectures match; it perturbs the negative branch + instead of the positive one. (2) Verified ComfyUI trap: in + `DualModelGuider`, passing the SAME model object as both the conditional + model and the LoRA-patched unconditional model is a no-op. Comfy-Org issue + #14521 tabulates it — case 3 (DualModelGuider, conditional model as + positive, conditional model + LoRA as negative) was "exactly identical" to + plain conditional-only CFG, while case 4 with the real unconditional model + differed. kijai confirmed it is expected: "there's no (working) mechanic to + swap LoRA on each step on the same model, it would only work if the model + was a different one." To patch the negative branch you must load a + genuinely separate model file. (3) A diffusers/PEFT-prefixed LoRA in + ComfyUI is expected to produce only a "lora key not loaded" console warning + and no visible effect — this follows from the absent key mapping (see + checkpoint_families) rather than from a published Ideogram-4 bug report. + (4) Cross-quantization loading (fp8-trained LoRA onto nvfp4/nf4/GGUF) + generally works with some precision loss; int8_convrot had a specific, now- + fixed bug (see gotchas). + +─────────────────────────────────────────────────────────────────────────────── + AUTHOR'S DESCRIPTION +─────────────────────────────────────────────────────────────────────────────── + Trained on Disco Elysium and Disco Elysium Mobile. Not limited to Disco + Elysium or digital art in general and can be used for paintings or + illustrations. + + Unlike most Disco Elysium LoRAs, this one focuses on versatility rather + than portraits. Logos/text is often present on landscapes and I will + address it in future versions. + + Always start your prompt with what kind of picture it is supposed to be + e.g. Digital art/Oil painting/Illustration of... + + Ernie version works better with lower weights. + +─────────────────────────────────────────────────────────────────────────────── + EXAMPLE MEDIA +─────────────────────────────────────────────────────────────────────────────── + 6 preview(s) in ./example_images/ — community results for + this LoRA. Videos keep a matching _poster.jpg still frame. + + No prompts are recorded below: CivitAI's public API flags these posts as + having generation metadata but does not return it, so the prompts cannot be + scraped. Open the model page to read them. + + 01_image.jpeg (image, 768x1376) + 02_image.jpeg (image, 768x1376) + 03_image.jpeg (image, 1024x1024) + 04_image.jpeg (image, 1024x1024) + 05_image.jpeg (image, 1024x1024) + 06_image.jpeg (image, 1024x1024) + +─────────────────────────────────────────────────────────────────────────────── + HOW TO USE (COMFYUI) +─────────────────────────────────────────────────────────────────────────────── + Use ComfyUI's NATIVE Ideogram 4 support (Comfy-Org/ComfyUI PR #14259 "feat: + Add Ideogram4 model support", merged 2026-06-03) — not the official + `ComfyUI-Ideogram4` custom-node wrapper, whose nodes (Magic Prompt / + Pipeline Loader / Generate) wrap Ideogram's reference pipeline and contain + no LoRA input at all (verified: no occurrence of "lora" in its nodes.py). + + 1. Update ComfyUI to a build containing `comfy_extras/nodes_ideogram4.py` + (which supplies `Ideogram4Scheduler`); the same PR added the generic + `DualModelGuider` and `CFGOverride` nodes, which live in + `comfy_extras/nodes_custom_sampler.py`. + 2. From `Comfy-Org/Ideogram-4` download into + `ComfyUI/models/diffusion_models/`: `ideogram4_fp8_scaled.safetensors` + (9.28 GB) AND `ideogram4_unconditional_fp8_scaled.safetensors` (9.28 GB). + Lower-VRAM alternatives in the same folder: `*_nvfp4_mixed.safetensors` + (5.49 GB each) or `*_int8_convrot.safetensors` (9.58 GB each). + 3. Download `qwen3vl_8b_fp8_scaled.safetensors` (10.59 GB; or + `qwen3vl_8b_nvfp4.safetensors`, 6.31 GB) into + `ComfyUI/models/text_encoders/`. The shipped template also lists an + optional `gemma4_e4b_it_fp8_scaled.safetensors` from `Comfy-Org/gemma-4` + for local prompt building. + 4. Download `flux2-vae.safetensors` (0.34 GB, from Comfy-Org/Ideogram-4 or + Comfy-Org/flux2-dev) into `ComfyUI/models/vae/`. + 5. Put LoRA .safetensors in `ComfyUI/models/loras/` (no subfolder + convention enforced). + 6. Load the built-in template `image_ideogram4_t2i` (Workflow → Browse + Templates), or the workflow from + docs.comfy.org/tutorials/image/ideogram/ideogram-v4. + 7. Insert a `LoraLoaderModelOnly` between the main `UNETLoader` and the + `DualModelGuider`'s conditional model input. There is no CLIP-side LoRA + here — the text encoder is a frozen Qwen3-VL — so model-only is the right + loader. Strength ~0.5–0.8. + 8. Do NOT reuse one patched model object for both guider inputs; it + silently does nothing. Patching the negative branch requires a second, + physically separate model file with its own loader. + 9. Write the prompt as a JSON caption into `CLIPTextEncode` (the CLIPLoader + type is `ideogram4`); keep `ConditioningZeroOut` feeding the unconditional + branch. + + Non-ComfyUI paths: diffusers — + `Ideogram4Pipeline.from_pretrained("ideogram-ai/ideogram-4-nf4")` (the + docs' own example; `ideogram-ai/ideogram-4-nf4-diffusers` also exists), + then `pipeline.load_lora_weights(...)`. Training — `ostris/ai-toolkit` + (built-in `extensions_built_in/diffusion_models/ideogram4/` plus an + `Ideogram4Captioner` that writes JSON captions) or diffusers + `examples/dreambooth/train_dreambooth_lora_ideogram4.py`. The official HF + weights are GATED (`gated: auto`): accept the license on the model page and + authenticate (`hf auth login` / `HF_TOKEN`) or downloads fail. + +─────────────────────────────────────────────────────────────────────────────── + Scraped by lorakit on 2026-09-05 01:49 UTC. + Stats and description are the author's; pros/cons are derived from the + evidence tagged beside each line. +─────────────────────────────────────────────────────────────────────────────── diff --git a/disco-elysium/metadata.json b/disco-elysium/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..e3e0478af5588bd8a2722423083628579a1647b2 --- /dev/null +++ b/disco-elysium/metadata.json @@ -0,0 +1,131 @@ +{ + "name": "Disco Elysium", + "slug": "disco-elysium", + "base_model": "Ideogram 4", + "media_type": "image", + "model_id": 2487755, + "version_id": 3129598, + "version_name": "Ideogram", + "source": "civitai", + "url": "https://civitai.com/models/2487755", + "repo_id": null, + "file_path": null, + "author": "wystanbelov785", + "category": { + "key": "style", + "label": "Visual style" + }, + "modalities": [], + "primary_modality": "unspecified", + "checkpoint_family": "unknown", + "trigger_words": [], + "recommended_settings": {}, + "chaining": { + "author": { + "stacks": "not stated", + "notes": "", + "avoid": [], + "requires": [], + "order": "", + "strength": "", + "stated": false + }, + "general_rules": [ + "Chain `LoraLoaderModelOnly` nodes in series off the CONDITIONAL model loader, ahead of the guider. DeverStyle states its samples were made \"by setting the LORA @ 0.6 strength on the model only, not on the model_negative\".", + "Model-only loaders throughout. The text encoder is a frozen Qwen3-VL-8B with no LoRA-patchable CLIP branch, so the model+clip `LoraLoader` has nothing useful to do.", + "To affect the unconditional/negative branch you must load a second, physically distinct model file and patch that one. Passing the same patched model object into both DualModelGuider inputs is a verified no-op (Comfy-Org issue #14521; kijai confirmed it is expected behaviour).", + "Trigger words go INSIDE the JSON caption, in the semantically matching field, not prepended to a prose string. DeverStyle instructs putting `dvr_arch_tv` in `style_description.art_style` (and in the content field); the Kiki LoRA card describes using `art_style: \"Studio Ghibli style\"` with `medium: \"anime_screencap\"`.", + "Two versions of the same LoRA can be stacked at reduced strengths: zjourney's author recommends v1 @ 0.35 + v2 @ 0.45 over either alone at full strength.", + "TurboTime changes the graph, not just the strength: it runs at cfg 1.0 and the reference community workflow swaps `DualModelGuider` for a single-model `CFGGuider`, dropping the unconditional model. Re-tune any stacked style LoRA downward.", + "A LoRA can be applied to only part of the run. The Gray Screen bypass LoRA is used at NEGATIVE weight (about -0.25) and, in its shipped workflows, only on the first step and then disabled; its author warns that \"LoRA being active through the whole generation could lead to degradation\".", + "The author never states which weight family this targets. Silent degradation, not hard errors, in every case observed. (1) Loading a conditional-trained LoRA onto the unconditional transformer succeeds silently because the architectures match; it perturbs the negative branch instead of the positive one. (2) Verified ComfyUI trap: in `Du" + ] + }, + "pros": [ + "Compact at 202 MB — cheap to stack with other LoRAs. [source: file size]" + ], + "cons": [ + "Needs a reduced strength; the default 1.0 is likely too strong. [source: description]", + "No trigger word — the LoRA is always on once loaded, so strength is the only control. [source: model metadata]", + "Author never states which generation mode it was trained for; test against your own workflow before relying on it. [source: description]", + "Modest adoption (251 downloads) — less community feedback to rely on. [source: source stats]" + ], + "file": { + "name": "MixedElysium_Ideogram.safetensors", + "size_bytes": 211734080, + "size_mb": 201.93, + "sha256": "e762d7696337f6e50f78a7316a37ef2bcea847ab9a712f5ac5f1fa91697c0caa" + }, + "stats": { + "downloads": 251, + "thumbs_up": 22, + "thumbs_down": 0 + }, + "popularity": { + "downloads": 251, + "thumbs_up": 22, + "thumbs_down": 0, + "comments": 0, + "approval": "100% positive (22/22)", + "age_days": 52.4, + "downloads_per_day": 0.71, + "rank_by_downloads": 26, + "rank_by_momentum": 27, + "of_total": 35, + "adoption_tier": "Modest", + "momentum": "Long tail" + }, + "selection_score": 515.0, + "selection_reasons": [ + "251 downloads", + "22 thumbs-up", + "top-ranked for its category" + ], + "example_media": [ + { + "file": "01_image.jpeg", + "type": "image", + "resolution": "768x1376", + "prompt": "" + }, + { + "file": "02_image.jpeg", + "type": "image", + "resolution": "768x1376", + "prompt": "" + }, + { + "file": "03_image.jpeg", + "type": "image", + "resolution": "1024x1024", + "prompt": "" + }, + { + "file": "04_image.jpeg", + "type": "image", + "resolution": "1024x1024", + "prompt": "" + }, + { + "file": "05_image.jpeg", + "type": "image", + "resolution": "1024x1024", + "prompt": "" + }, + { + "file": "06_image.jpeg", + "type": "image", + "resolution": "1024x1024", + "prompt": "" + } + ], + "tags": [ + "style", + "cartoon", + "digital art", + "oil painting", + "illustration" + ], + "scraped_at": "2026-09-05T01:49:31+00:00", + "path": "ideogram-4/disco-elysium" +} diff --git a/dream-loom/Sime2_Ideogram.safetensors b/dream-loom/Sime2_Ideogram.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..1a5bdea76a010432486b64408d7c6616527c1303 --- /dev/null +++ b/dream-loom/Sime2_Ideogram.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fdf61f3f7f4bb948c30c5ed01e87e0b14fbee745c3cfac7dd0944833ee98f16e +size 211734048 diff --git a/dream-loom/example_images/01_image.jpeg b/dream-loom/example_images/01_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..3c545c70d9fc3372d6f623d10b8b2012e3ef7c72 --- /dev/null +++ b/dream-loom/example_images/01_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5e6e0482957045c37f5aeb2f15eef85d5e5e3f84a7db691c2ce61535c6f91c86 +size 2510555 diff --git a/dream-loom/example_images/02_image.jpeg b/dream-loom/example_images/02_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..0e3332dfdae80caf9e5d1f7473dbf9c1d323cf62 --- /dev/null +++ b/dream-loom/example_images/02_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bd3b57f8cbbf283ea125cb308698331056775a2f48f409e5d439b48a4fd21e9d +size 1972648 diff --git a/dream-loom/example_images/03_image.jpeg b/dream-loom/example_images/03_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..806e51b2e1ce660f9acddb03a8d177aaaaf94f52 --- /dev/null +++ b/dream-loom/example_images/03_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:548f261f0ef2d373b48defed3db9d23eabeb41c18528cb05a29e892c9506db77 +size 1896350 diff --git a/dream-loom/example_images/04_image.jpeg b/dream-loom/example_images/04_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..e748054ae2aa84265fa50a04a99468b8abb1b3be --- /dev/null +++ b/dream-loom/example_images/04_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e7ce59847160db4c30f4c4156b41fb9f46baa73a0396b321e44019df4753fe74 +size 1777546 diff --git a/dream-loom/example_images/05_image.jpeg b/dream-loom/example_images/05_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..0b19bf669b87fcbe7a9333a39f94a47ed0501922 --- /dev/null +++ b/dream-loom/example_images/05_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f6e46202617511f904ec199b3144e110f097e2d0327411b4b6377b9c13459a93 +size 1837099 diff --git a/dream-loom/example_images/06_image.jpeg b/dream-loom/example_images/06_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..1e88aaaad09287c016b73508b2e65a4ac64f4f24 --- /dev/null +++ b/dream-loom/example_images/06_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f94660fdb75142c21df6b3764ccf74b00f16e4809b9073eec2227aa42da81b96 +size 2000477 diff --git a/dream-loom/info.txt b/dream-loom/info.txt new file mode 100644 index 0000000000000000000000000000000000000000..e0c1b9a4c6249f0482f8ceead4c217eae8e912c7 --- /dev/null +++ b/dream-loom/info.txt @@ -0,0 +1,258 @@ +═══════════════════════════════════════════════════════════════════════════════ + DREAM LOOM +═══════════════════════════════════════════════════════════════════════════════ + +─────────────────────────────────────────────────────────────────────────────── + IDENTITY +─────────────────────────────────────────────────────────────────────────────── + Base model Ideogram 4 + Generation mode not declared by the author + Checkpoint not determined — verify before use + Category Visual style + Author wystanbelov785 + Source https://civitai.com/models/2605989 + Provenance CivitAI · model 2605989 · version 3113138 + +─────────────────────────────────────────────────────────────────────────────── + WEIGHTS +─────────────────────────────────────────────────────────────────────────────── + File Sime2_Ideogram.safetensors + Version Ideogram + Size 201.9 MB + Format SafeTensor / unknown precision + SHA256 fdf61f3f7f4bb948c30c5ed01e87e0b14fbee745c3cfac7dd0944833ee98f16e + Published 2026-07-09 + +─────────────────────────────────────────────────────────────────────────────── + TRIGGER WORDS +─────────────────────────────────────────────────────────────────────────────── + No trigger word. The LoRA applies as soon as it is loaded — + strength is the only way to dial the effect up or down. + +─────────────────────────────────────────────────────────────────────────────── + RECOMMENDED SETTINGS +─────────────────────────────────────────────────────────────────────────────── + The author did not publish specific settings. + No author settings were published. Community baseline: steps 48 is the + reference default: docs/inference.md says "`V4_QUALITY_48` is the default" + and the pipeline's `num_steps` default is 48; diffusers documents the same + 48-step default. 20 is the practical ComfyUI default — the `V4_DEFAULT_20` + preset and the value the shipped template `image_ideogram4_t2i.json` + selects (its preset combo defaults to "Default"). 12 = `V4_TURBO_12`. With + the TurboTime LoRA: 8 in the nomadoor community workflow; the model card + claims usable output down to 2., cfg A guidance SCHEDULE, not one number, + and the negative branch comes from a second model rather than a negative + prompt. Verified from src/ideogram4/sampler_configs.py: V4_QUALITY_48 = 45 + steps @ gw 7 then 3 polish steps @ gw 3; V4_DEFAULT_20 = 18 @ 7 + 2 @ 3; + V4_TURBO_12 = 11 @ 7 + 1 @ 3. Pipeline default `guidance_scale` = 7.0 when + no schedule is given. In ComfyUI the shipped template sets + `DualModelGuider` cfg 7 plus a `CFGOverride` node with widgets [3, 0.7, 1] + (cfg 3 from 0.7→1.0 of the run); kijai recommends starting the override at + 0.7, not 0.9. TurboTime: cfg 1.0 / no CFG., sampler Euler flow-matching. In + ComfyUI: `KSamplerSelect` → euler, driven by `SamplerCustomAdvanced`, with + the model-specific `Ideogram4Scheduler` (logit-normal sigma schedule). Its + params are `mu` and `std`, not the usual `shift`. Reference presets: mu 0.0 + / std 1.5 (48-step), mu 0.0 / std 1.75 (20-step), mu 0.5 / std 1.75 + (12-step). The Ideogram4Scheduler node's own widget defaults are steps 20, + mu 0.5, std 1.75, but the shipped template overrides mu/std/steps from the + preset JSON, so out of the box it runs 20 / mu 0.0 / std 1.75. mu is auto- + shifted by resolution in both implementations: mean = mu + + 0.5*log((w*h)/(512*512))., resolution 1024×1024 default (pipeline + `height`/`width` default 1024). Any multiple of 16 per side from 256 to + 2048, aspect ratios up to 6:1 or 1:6 — one model covers all of it. + 2048×2048 with V4_QUALITY_48 for maximum quality. Buckets documented in + docs/inference.md: 1536×1024 (3:2), 1024×1536 (2:3), 1920×1088 (~16:9), + 2048×768 (~21:9), 1024×1792 (~9:16), 1600×400 (4:1)., lora_strength No + single official number; these are author-stated values. DeverStyle Archer: + 0.6, and explicitly "on the model only, not on the model_negative". + zjourney v2: 0.55 standalone (range 0.35–0.6), or v1 @ 0.35 + v2 @ 0.45 + stacked. Furry Enhancer: the model page says 0.6–0.8, but its own version + notes ask for 0.7–0.9 (v1.94) and 0.8–0.95 (v2.8). TurboTime: 0.6 in the + nomadoor workflow (the author gives no figure). So ~0.5–0.9 is the + realistic band, read the specific card.. + +─────────────────────────────────────────────────────────────────────────────── + ADOPTION & POPULARITY +─────────────────────────────────────────────────────────────────────────────── + Downloads 89 + Rating 100% positive (4/4) + Comments 0 + Published 58 days ago + Download rate 0.6/day since release + + Adoption rank #35 of 35 in this collection (Niche) + Momentum rank #28 of 35 by download rate (Long tail) + + Adoption tier is the percentile of total downloads within this collection; + momentum is the percentile of downloads-per-day since release. Momentum is + ranked rather than measured against a fixed rate, because this base model + is itself new and every LoRA looks fast on an absolute scale. Both numbers + are CivitAI's, read on the scrape date at the foot of this file. + +─────────────────────────────────────────────────────────────────────────────── + CHAINING / STACKING +─────────────────────────────────────────────────────────────────────────────── + Stacks with others not stated + + The author gave no chaining guidance. The rules below are the + general Ideogram 4 ones, not anything specific to this LoRA. + + General Ideogram 4 rules that apply here (see REFERENCE.txt): + + • Chain `LoraLoaderModelOnly` nodes in series off the CONDITIONAL model + loader, ahead of the guider. DeverStyle states its samples were made + "by setting the LORA @ 0.6 strength on the model only, not on the + model_negative". + • Model-only loaders throughout. The text encoder is a frozen Qwen3-VL-8B + with no LoRA-patchable CLIP branch, so the model+clip `LoraLoader` has + nothing useful to do. + • To affect the unconditional/negative branch you must load a second, + physically distinct model file and patch that one. Passing the same + patched model object into both DualModelGuider inputs is a verified no- + op (Comfy-Org issue #14521; kijai confirmed it is expected behaviour). + • Trigger words go INSIDE the JSON caption, in the semantically matching + field, not prepended to a prose string. DeverStyle instructs putting + `dvr_arch_tv` in `style_description.art_style` (and in the content + field); the Kiki LoRA card describes using `art_style: "Studio Ghibli + style"` with `medium: "anime_screencap"`. + • Two versions of the same LoRA can be stacked at reduced strengths: + zjourney's author recommends v1 @ 0.35 + v2 @ 0.45 over either alone at + full strength. + • TurboTime changes the graph, not just the strength: it runs at cfg 1.0 + and the reference community workflow swaps `DualModelGuider` for a + single-model `CFGGuider`, dropping the unconditional model. Re-tune any + stacked style LoRA downward. + • A LoRA can be applied to only part of the run. The Gray Screen bypass + LoRA is used at NEGATIVE weight (about -0.25) and, in its shipped + workflows, only on the first step and then disabled; its author warns + that "LoRA being active through the whole generation could lead to + degradation". + • The author never states which weight family this targets. Silent + degradation, not hard errors, in every case observed. (1) Loading a + conditional-trained LoRA onto the unconditional transformer succeeds + silently because the architectures match; it perturbs the negative + branch instead of the positive one. (2) Verified ComfyUI trap: in `Du + +─────────────────────────────────────────────────────────────────────────────── + PROS +─────────────────────────────────────────────────────────────────────────────── + + Compact at 202 MB — cheap to stack with other LoRAs. [source: file + size] + +─────────────────────────────────────────────────────────────────────────────── + CONS & CAVEATS +─────────────────────────────────────────────────────────────────────────────── + - No trigger word — the LoRA is always on once loaded, so strength is the + only control. [source: model metadata] + - Author never states which generation mode it was trained for; test + against your own workflow before relying on it. [source: description] + - Modest adoption (89 downloads) — less community feedback to rely on. + [source: source stats] + +─────────────────────────────────────────────────────────────────────────────── + COMPATIBILITY WARNING +─────────────────────────────────────────────────────────────────────────────── + Silent degradation, not hard errors, in every case observed. (1) Loading a + conditional-trained LoRA onto the unconditional transformer succeeds + silently because the architectures match; it perturbs the negative branch + instead of the positive one. (2) Verified ComfyUI trap: in + `DualModelGuider`, passing the SAME model object as both the conditional + model and the LoRA-patched unconditional model is a no-op. Comfy-Org issue + #14521 tabulates it — case 3 (DualModelGuider, conditional model as + positive, conditional model + LoRA as negative) was "exactly identical" to + plain conditional-only CFG, while case 4 with the real unconditional model + differed. kijai confirmed it is expected: "there's no (working) mechanic to + swap LoRA on each step on the same model, it would only work if the model + was a different one." To patch the negative branch you must load a + genuinely separate model file. (3) A diffusers/PEFT-prefixed LoRA in + ComfyUI is expected to produce only a "lora key not loaded" console warning + and no visible effect — this follows from the absent key mapping (see + checkpoint_families) rather than from a published Ideogram-4 bug report. + (4) Cross-quantization loading (fp8-trained LoRA onto nvfp4/nf4/GGUF) + generally works with some precision loss; int8_convrot had a specific, now- + fixed bug (see gotchas). + +─────────────────────────────────────────────────────────────────────────────── + AUTHOR'S DESCRIPTION +─────────────────────────────────────────────────────────────────────────────── + Custom style Lora. Trained on works of Sidney Sime in color and my + generations. + Dreamy, atmospheric, mystical, romantic, nostalgic, slightly dark oil + paintings, Illustrations, watercolor. Mostly Fantasy/Fairy tale, but works + with anything. + + Sometimes it "overdoes" the style, slightly lowered weights work well. + +─────────────────────────────────────────────────────────────────────────────── + EXAMPLE MEDIA +─────────────────────────────────────────────────────────────────────────────── + 6 preview(s) in ./example_images/ — community results for + this LoRA. Videos keep a matching _poster.jpg still frame. + + No prompts are recorded below: CivitAI's public API flags these posts as + having generation metadata but does not return it, so the prompts cannot be + scraped. Open the model page to read them. + + 01_image.jpeg (image, 768x1376) + 02_image.jpeg (image, 1024x1024) + 03_image.jpeg (image, 1024x1024) + 04_image.jpeg (image, 1024x1024) + 05_image.jpeg (image, 1024x1024) + 06_image.jpeg (image, 1024x1024) + +─────────────────────────────────────────────────────────────────────────────── + HOW TO USE (COMFYUI) +─────────────────────────────────────────────────────────────────────────────── + Use ComfyUI's NATIVE Ideogram 4 support (Comfy-Org/ComfyUI PR #14259 "feat: + Add Ideogram4 model support", merged 2026-06-03) — not the official + `ComfyUI-Ideogram4` custom-node wrapper, whose nodes (Magic Prompt / + Pipeline Loader / Generate) wrap Ideogram's reference pipeline and contain + no LoRA input at all (verified: no occurrence of "lora" in its nodes.py). + + 1. Update ComfyUI to a build containing `comfy_extras/nodes_ideogram4.py` + (which supplies `Ideogram4Scheduler`); the same PR added the generic + `DualModelGuider` and `CFGOverride` nodes, which live in + `comfy_extras/nodes_custom_sampler.py`. + 2. From `Comfy-Org/Ideogram-4` download into + `ComfyUI/models/diffusion_models/`: `ideogram4_fp8_scaled.safetensors` + (9.28 GB) AND `ideogram4_unconditional_fp8_scaled.safetensors` (9.28 GB). + Lower-VRAM alternatives in the same folder: `*_nvfp4_mixed.safetensors` + (5.49 GB each) or `*_int8_convrot.safetensors` (9.58 GB each). + 3. Download `qwen3vl_8b_fp8_scaled.safetensors` (10.59 GB; or + `qwen3vl_8b_nvfp4.safetensors`, 6.31 GB) into + `ComfyUI/models/text_encoders/`. The shipped template also lists an + optional `gemma4_e4b_it_fp8_scaled.safetensors` from `Comfy-Org/gemma-4` + for local prompt building. + 4. Download `flux2-vae.safetensors` (0.34 GB, from Comfy-Org/Ideogram-4 or + Comfy-Org/flux2-dev) into `ComfyUI/models/vae/`. + 5. Put LoRA .safetensors in `ComfyUI/models/loras/` (no subfolder + convention enforced). + 6. Load the built-in template `image_ideogram4_t2i` (Workflow → Browse + Templates), or the workflow from + docs.comfy.org/tutorials/image/ideogram/ideogram-v4. + 7. Insert a `LoraLoaderModelOnly` between the main `UNETLoader` and the + `DualModelGuider`'s conditional model input. There is no CLIP-side LoRA + here — the text encoder is a frozen Qwen3-VL — so model-only is the right + loader. Strength ~0.5–0.8. + 8. Do NOT reuse one patched model object for both guider inputs; it + silently does nothing. Patching the negative branch requires a second, + physically separate model file with its own loader. + 9. Write the prompt as a JSON caption into `CLIPTextEncode` (the CLIPLoader + type is `ideogram4`); keep `ConditioningZeroOut` feeding the unconditional + branch. + + Non-ComfyUI paths: diffusers — + `Ideogram4Pipeline.from_pretrained("ideogram-ai/ideogram-4-nf4")` (the + docs' own example; `ideogram-ai/ideogram-4-nf4-diffusers` also exists), + then `pipeline.load_lora_weights(...)`. Training — `ostris/ai-toolkit` + (built-in `extensions_built_in/diffusion_models/ideogram4/` plus an + `Ideogram4Captioner` that writes JSON captions) or diffusers + `examples/dreambooth/train_dreambooth_lora_ideogram4.py`. The official HF + weights are GATED (`gated: auto`): accept the license on the model page and + authenticate (`hf auth login` / `HF_TOKEN`) or downloads fail. + +─────────────────────────────────────────────────────────────────────────────── + Scraped by lorakit on 2026-09-05 01:50 UTC. + Stats and description are the author's; pros/cons are derived from the + evidence tagged beside each line. +─────────────────────────────────────────────────────────────────────────────── diff --git a/dream-loom/metadata.json b/dream-loom/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..b66d4176d7c9ef12305fe313b4fca0e4a19b4a39 --- /dev/null +++ b/dream-loom/metadata.json @@ -0,0 +1,129 @@ +{ + "name": "Dream Loom", + "slug": "dream-loom", + "base_model": "Ideogram 4", + "media_type": "image", + "model_id": 2605989, + "version_id": 3113138, + "version_name": "Ideogram", + "source": "civitai", + "url": "https://civitai.com/models/2605989", + "repo_id": null, + "file_path": null, + "author": "wystanbelov785", + "category": { + "key": "style", + "label": "Visual style" + }, + "modalities": [], + "primary_modality": "unspecified", + "checkpoint_family": "unknown", + "trigger_words": [], + "recommended_settings": {}, + "chaining": { + "author": { + "stacks": "not stated", + "notes": "", + "avoid": [], + "requires": [], + "order": "", + "strength": "", + "stated": false + }, + "general_rules": [ + "Chain `LoraLoaderModelOnly` nodes in series off the CONDITIONAL model loader, ahead of the guider. DeverStyle states its samples were made \"by setting the LORA @ 0.6 strength on the model only, not on the model_negative\".", + "Model-only loaders throughout. The text encoder is a frozen Qwen3-VL-8B with no LoRA-patchable CLIP branch, so the model+clip `LoraLoader` has nothing useful to do.", + "To affect the unconditional/negative branch you must load a second, physically distinct model file and patch that one. Passing the same patched model object into both DualModelGuider inputs is a verified no-op (Comfy-Org issue #14521; kijai confirmed it is expected behaviour).", + "Trigger words go INSIDE the JSON caption, in the semantically matching field, not prepended to a prose string. DeverStyle instructs putting `dvr_arch_tv` in `style_description.art_style` (and in the content field); the Kiki LoRA card describes using `art_style: \"Studio Ghibli style\"` with `medium: \"anime_screencap\"`.", + "Two versions of the same LoRA can be stacked at reduced strengths: zjourney's author recommends v1 @ 0.35 + v2 @ 0.45 over either alone at full strength.", + "TurboTime changes the graph, not just the strength: it runs at cfg 1.0 and the reference community workflow swaps `DualModelGuider` for a single-model `CFGGuider`, dropping the unconditional model. Re-tune any stacked style LoRA downward.", + "A LoRA can be applied to only part of the run. The Gray Screen bypass LoRA is used at NEGATIVE weight (about -0.25) and, in its shipped workflows, only on the first step and then disabled; its author warns that \"LoRA being active through the whole generation could lead to degradation\".", + "The author never states which weight family this targets. Silent degradation, not hard errors, in every case observed. (1) Loading a conditional-trained LoRA onto the unconditional transformer succeeds silently because the architectures match; it perturbs the negative branch instead of the positive one. (2) Verified ComfyUI trap: in `Du" + ] + }, + "pros": [ + "Compact at 202 MB — cheap to stack with other LoRAs. [source: file size]" + ], + "cons": [ + "No trigger word — the LoRA is always on once loaded, so strength is the only control. [source: model metadata]", + "Author never states which generation mode it was trained for; test against your own workflow before relying on it. [source: description]", + "Modest adoption (89 downloads) — less community feedback to rely on. [source: source stats]" + ], + "file": { + "name": "Sime2_Ideogram.safetensors", + "size_bytes": 211734048, + "size_mb": 201.93, + "sha256": "fdf61f3f7f4bb948c30c5ed01e87e0b14fbee745c3cfac7dd0944833ee98f16e" + }, + "stats": { + "downloads": 89, + "thumbs_up": 4, + "thumbs_down": 0 + }, + "popularity": { + "downloads": 89, + "thumbs_up": 4, + "thumbs_down": 0, + "comments": 0, + "approval": "100% positive (4/4)", + "age_days": 57.5, + "downloads_per_day": 0.64, + "rank_by_downloads": 35, + "rank_by_momentum": 28, + "of_total": 35, + "adoption_tier": "Niche", + "momentum": "Long tail" + }, + "selection_score": 137.0, + "selection_reasons": [ + "89 downloads", + "4 thumbs-up", + "top-ranked for its category" + ], + "example_media": [ + { + "file": "01_image.jpeg", + "type": "image", + "resolution": "768x1376", + "prompt": "" + }, + { + "file": "02_image.jpeg", + "type": "image", + "resolution": "1024x1024", + "prompt": "" + }, + { + "file": "03_image.jpeg", + "type": "image", + "resolution": "1024x1024", + "prompt": "" + }, + { + "file": "04_image.jpeg", + "type": "image", + "resolution": "1024x1024", + "prompt": "" + }, + { + "file": "05_image.jpeg", + "type": "image", + "resolution": "1024x1024", + "prompt": "" + }, + { + "file": "06_image.jpeg", + "type": "image", + "resolution": "1024x1024", + "prompt": "" + } + ], + "tags": [ + "illustration", + "style", + "watercolor", + "oil painting" + ], + "scraped_at": "2026-09-05T01:50:09+00:00", + "path": "ideogram-4/dream-loom" +} diff --git a/gray-screen-bypass-lora-and-workflow/Gray_000002000.safetensors b/gray-screen-bypass-lora-and-workflow/Gray_000002000.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2589cc98fb877c22284311c83cbbc27290b342fe --- /dev/null +++ b/gray-screen-bypass-lora-and-workflow/Gray_000002000.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0292898e3f3cd3edbb15bcff80e81a81a5f4e5fc221438fc9fd0b5fcab89ae78 +size 211734024 diff --git a/gray-screen-bypass-lora-and-workflow/example_images/01_image.jpeg b/gray-screen-bypass-lora-and-workflow/example_images/01_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..66d22bab00dcd44183829212dafaf87ccea28d48 --- /dev/null +++ b/gray-screen-bypass-lora-and-workflow/example_images/01_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b0017b6788d303cfed3cb39211aadccbe6134d2c9d4e12090e19f3fc7f1af82c +size 438038 diff --git a/gray-screen-bypass-lora-and-workflow/example_images/02_image.jpeg b/gray-screen-bypass-lora-and-workflow/example_images/02_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..c3b75f27e1c0d24c0fdddf07cfd049cd23126a29 --- /dev/null +++ b/gray-screen-bypass-lora-and-workflow/example_images/02_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b042a766f1a04ee206314ce62fc0af855ac3b382f41642b1629572c9eadfa467 +size 473774 diff --git a/gray-screen-bypass-lora-and-workflow/example_images/03_image.jpeg b/gray-screen-bypass-lora-and-workflow/example_images/03_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..dfe8a743347572575a8cacbd078d8887b3ddadae --- /dev/null +++ b/gray-screen-bypass-lora-and-workflow/example_images/03_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f7e4dea7dbbe1e33f80d0776a02ec9ad325ec49835f1226d7a31b135a0bc017d +size 462538 diff --git a/gray-screen-bypass-lora-and-workflow/example_images/04_image.jpeg b/gray-screen-bypass-lora-and-workflow/example_images/04_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..269506ae6d0145a7724dbeab9c6cc71bf3c93933 --- /dev/null +++ b/gray-screen-bypass-lora-and-workflow/example_images/04_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5314227d020aab03209229a701368681056732e3ab9c3c85e97f780d87f5a85d +size 1797959 diff --git a/gray-screen-bypass-lora-and-workflow/example_images/05_image.jpeg b/gray-screen-bypass-lora-and-workflow/example_images/05_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..e3e54a9952effc94aa2d39b6f920c7c20b22685c --- /dev/null +++ b/gray-screen-bypass-lora-and-workflow/example_images/05_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c8b6071cc777941dc0996dcb79517fb8ad3ceba75056be321e2afbfdb2565de0 +size 2772604 diff --git a/gray-screen-bypass-lora-and-workflow/example_images/06_image.jpeg b/gray-screen-bypass-lora-and-workflow/example_images/06_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..56106da7379fa73ac9c06fb161562b50ad05a947 --- /dev/null +++ b/gray-screen-bypass-lora-and-workflow/example_images/06_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:77e82cf50f543bd8c185548eda1cba83a1677efe0cebe8f8e8be5a0d803c1fe0 +size 497162 diff --git a/gray-screen-bypass-lora-and-workflow/info.txt b/gray-screen-bypass-lora-and-workflow/info.txt new file mode 100644 index 0000000000000000000000000000000000000000..8a595314929622c5e4f7693dafa2974caa0615b7 --- /dev/null +++ b/gray-screen-bypass-lora-and-workflow/info.txt @@ -0,0 +1,267 @@ +═══════════════════════════════════════════════════════════════════════════════ + GRAY SCREEN BYPASS (LORA AND WORKFLOW) +═══════════════════════════════════════════════════════════════════════════════ + +─────────────────────────────────────────────────────────────────────────────── + IDENTITY +─────────────────────────────────────────────────────────────────────────────── + Base model Ideogram 4 + Generation mode Text-to-image + Checkpoint Conditional transformer (`transformer/`) + Category Visual style + Author wystanbelov785 + Source https://civitai.com/models/2750357 + Provenance CivitAI · model 2750357 · version 3094072 + +─────────────────────────────────────────────────────────────────────────────── + WEIGHTS +─────────────────────────────────────────────────────────────────────────────── + File Gray_000002000.safetensors + Version v1.0 + Size 201.9 MB + Format SafeTensor / unknown precision + SHA256 0292898e3f3cd3edbb15bcff80e81a81a5f4e5fc221438fc9fd0b5fcab89ae78 + Published 2026-07-03 + +─────────────────────────────────────────────────────────────────────────────── + TRIGGER WORDS +─────────────────────────────────────────────────────────────────────────────── + No trigger word. The LoRA applies as soon as it is loaded — + strength is the only way to dial the effect up or down. + +─────────────────────────────────────────────────────────────────────────────── + RECOMMENDED SETTINGS +─────────────────────────────────────────────────────────────────────────────── + The author did not publish specific settings. + No author settings were published. Community baseline: steps 48 is the + reference default: docs/inference.md says "`V4_QUALITY_48` is the default" + and the pipeline's `num_steps` default is 48; diffusers documents the same + 48-step default. 20 is the practical ComfyUI default — the `V4_DEFAULT_20` + preset and the value the shipped template `image_ideogram4_t2i.json` + selects (its preset combo defaults to "Default"). 12 = `V4_TURBO_12`. With + the TurboTime LoRA: 8 in the nomadoor community workflow; the model card + claims usable output down to 2., cfg A guidance SCHEDULE, not one number, + and the negative branch comes from a second model rather than a negative + prompt. Verified from src/ideogram4/sampler_configs.py: V4_QUALITY_48 = 45 + steps @ gw 7 then 3 polish steps @ gw 3; V4_DEFAULT_20 = 18 @ 7 + 2 @ 3; + V4_TURBO_12 = 11 @ 7 + 1 @ 3. Pipeline default `guidance_scale` = 7.0 when + no schedule is given. In ComfyUI the shipped template sets + `DualModelGuider` cfg 7 plus a `CFGOverride` node with widgets [3, 0.7, 1] + (cfg 3 from 0.7→1.0 of the run); kijai recommends starting the override at + 0.7, not 0.9. TurboTime: cfg 1.0 / no CFG., sampler Euler flow-matching. In + ComfyUI: `KSamplerSelect` → euler, driven by `SamplerCustomAdvanced`, with + the model-specific `Ideogram4Scheduler` (logit-normal sigma schedule). Its + params are `mu` and `std`, not the usual `shift`. Reference presets: mu 0.0 + / std 1.5 (48-step), mu 0.0 / std 1.75 (20-step), mu 0.5 / std 1.75 + (12-step). The Ideogram4Scheduler node's own widget defaults are steps 20, + mu 0.5, std 1.75, but the shipped template overrides mu/std/steps from the + preset JSON, so out of the box it runs 20 / mu 0.0 / std 1.75. mu is auto- + shifted by resolution in both implementations: mean = mu + + 0.5*log((w*h)/(512*512))., resolution 1024×1024 default (pipeline + `height`/`width` default 1024). Any multiple of 16 per side from 256 to + 2048, aspect ratios up to 6:1 or 1:6 — one model covers all of it. + 2048×2048 with V4_QUALITY_48 for maximum quality. Buckets documented in + docs/inference.md: 1536×1024 (3:2), 1024×1536 (2:3), 1920×1088 (~16:9), + 2048×768 (~21:9), 1024×1792 (~9:16), 1600×400 (4:1)., lora_strength No + single official number; these are author-stated values. DeverStyle Archer: + 0.6, and explicitly "on the model only, not on the model_negative". + zjourney v2: 0.55 standalone (range 0.35–0.6), or v1 @ 0.35 + v2 @ 0.45 + stacked. Furry Enhancer: the model page says 0.6–0.8, but its own version + notes ask for 0.7–0.9 (v1.94) and 0.8–0.95 (v2.8). TurboTime: 0.6 in the + nomadoor workflow (the author gives no figure). So ~0.5–0.9 is the + realistic band, read the specific card.. + +─────────────────────────────────────────────────────────────────────────────── + ADOPTION & POPULARITY +─────────────────────────────────────────────────────────────────────────────── + Downloads 241 + Rating 100% positive (22/22) + Comments 0 + Published 64 days ago + Download rate 3.8/day since release + + Adoption rank #27 of 35 in this collection (Niche) + Momentum rank #8 of 35 by download rate (Trending) + + Adoption tier is the percentile of total downloads within this collection; + momentum is the percentile of downloads-per-day since release. Momentum is + ranked rather than measured against a fixed rate, because this base model + is itself new and every LoRA looks fast on an absolute scale. Both numbers + are CivitAI's, read on the scrape date at the foot of this file. + +─────────────────────────────────────────────────────────────────────────────── + CHAINING / STACKING +─────────────────────────────────────────────────────────────────────────────── + Stacks with others not stated + + The author gave no chaining guidance. The rules below are the + general Ideogram 4 ones, not anything specific to this LoRA. + + General Ideogram 4 rules that apply here (see REFERENCE.txt): + + • Chain `LoraLoaderModelOnly` nodes in series off the CONDITIONAL model + loader, ahead of the guider. DeverStyle states its samples were made + "by setting the LORA @ 0.6 strength on the model only, not on the + model_negative". + • Model-only loaders throughout. The text encoder is a frozen Qwen3-VL-8B + with no LoRA-patchable CLIP branch, so the model+clip `LoraLoader` has + nothing useful to do. + • To affect the unconditional/negative branch you must load a second, + physically distinct model file and patch that one. Passing the same + patched model object into both DualModelGuider inputs is a verified no- + op (Comfy-Org issue #14521; kijai confirmed it is expected behaviour). + • Trigger words go INSIDE the JSON caption, in the semantically matching + field, not prepended to a prose string. DeverStyle instructs putting + `dvr_arch_tv` in `style_description.art_style` (and in the content + field); the Kiki LoRA card describes using `art_style: "Studio Ghibli + style"` with `medium: "anime_screencap"`. + • Two versions of the same LoRA can be stacked at reduced strengths: + zjourney's author recommends v1 @ 0.35 + v2 @ 0.45 over either alone at + full strength. + • TurboTime changes the graph, not just the strength: it runs at cfg 1.0 + and the reference community workflow swaps `DualModelGuider` for a + single-model `CFGGuider`, dropping the unconditional model. Re-tune any + stacked style LoRA downward. + • A LoRA can be applied to only part of the run. The Gray Screen bypass + LoRA is used at NEGATIVE weight (about -0.25) and, in its shipped + workflows, only on the first step and then disabled; its author warns + that "LoRA being active through the whole generation could lead to + degradation". + • The main 9.3B DiT and what essentially every Ideogram 4 LoRA targets. + diffusers' README_ideogram4.md is explicit: "This trainer adds LoRA to + the conditional `transformer` only; the unconditional one stays + frozen." In ComfyUI this is `ideogram4_fp8_scaled.safetensors` (9.28 + GB) or its int8_convrot / nvfp4_mixed equivalents. + +─────────────────────────────────────────────────────────────────────────────── + PROS +─────────────────────────────────────────────────────────────────────────────── + + Compact at 202 MB — cheap to stack with other LoRAs. [source: file + size] + +─────────────────────────────────────────────────────────────────────────────── + CONS & CAVEATS +─────────────────────────────────────────────────────────────────────────────── + - No trigger word — the LoRA is always on once loaded, so strength is the + only control. [source: model metadata] + - Modest adoption (241 downloads) — less community feedback to rely on. + [source: source stats] + +─────────────────────────────────────────────────────────────────────────────── + COMPATIBILITY WARNING +─────────────────────────────────────────────────────────────────────────────── + Silent degradation, not hard errors, in every case observed. (1) Loading a + conditional-trained LoRA onto the unconditional transformer succeeds + silently because the architectures match; it perturbs the negative branch + instead of the positive one. (2) Verified ComfyUI trap: in + `DualModelGuider`, passing the SAME model object as both the conditional + model and the LoRA-patched unconditional model is a no-op. Comfy-Org issue + #14521 tabulates it — case 3 (DualModelGuider, conditional model as + positive, conditional model + LoRA as negative) was "exactly identical" to + plain conditional-only CFG, while case 4 with the real unconditional model + differed. kijai confirmed it is expected: "there's no (working) mechanic to + swap LoRA on each step on the same model, it would only work if the model + was a different one." To patch the negative branch you must load a + genuinely separate model file. (3) A diffusers/PEFT-prefixed LoRA in + ComfyUI is expected to produce only a "lora key not loaded" console warning + and no visible effect — this follows from the absent key mapping (see + checkpoint_families) rather than from a published Ideogram-4 bug report. + (4) Cross-quantization loading (fp8-trained LoRA onto nvfp4/nf4/GGUF) + generally works with some precision loss; int8_convrot had a specific, now- + fixed bug (see gotchas). + +─────────────────────────────────────────────────────────────────────────────── + AUTHOR'S DESCRIPTION +─────────────────────────────────────────────────────────────────────────────── + Trained on the iconic gray screen. If applied at a negative weight (for + example -0.25), it reliably prevents it from triggering no matter what is + in the prompt. + + In the attached workflows, the LoRA is only applied at the first step and + then disabled. This way there's no quality or aesthetic loss. LoRA being + active through the whole generation could lead to degradation. The second + workflow (image_ideogram4_t2i_no_gray_screen_lora_example) shows the way + any other LoRA would be applied. + + The prompt for the first preview picture (the cat) is written in natural + language. The second one had a JSON prompt without any bboxes. + + That being said, Ideogram is clearly not intended for natural language. + Prompt adherence is quite subpar and if the prompt is too short, it could + result in a complete mess. So I recommend using this LoRA AND correct JSON + prompts as an extra measure to make sure the gray screen never appears, not + as a way to use natural language prompts with it. + +─────────────────────────────────────────────────────────────────────────────── + EXAMPLE MEDIA +─────────────────────────────────────────────────────────────────────────────── + 6 preview(s) in ./example_images/ — community results for + this LoRA. Videos keep a matching _poster.jpg still frame. + + No prompts are recorded below: CivitAI's public API flags these posts as + having generation metadata but does not return it, so the prompts cannot be + scraped. Open the model page to read them. + + 01_image.jpeg (image, 1824x1248) + 02_image.jpeg (image, 2128x1456) + 03_image.jpeg (image, 2128x1456) + 04_image.jpeg (image, 1024x1024) + 05_image.jpeg (image, 1024x1024) + 06_image.jpeg (image, 2432x1664) + +─────────────────────────────────────────────────────────────────────────────── + HOW TO USE (COMFYUI) +─────────────────────────────────────────────────────────────────────────────── + Use ComfyUI's NATIVE Ideogram 4 support (Comfy-Org/ComfyUI PR #14259 "feat: + Add Ideogram4 model support", merged 2026-06-03) — not the official + `ComfyUI-Ideogram4` custom-node wrapper, whose nodes (Magic Prompt / + Pipeline Loader / Generate) wrap Ideogram's reference pipeline and contain + no LoRA input at all (verified: no occurrence of "lora" in its nodes.py). + + 1. Update ComfyUI to a build containing `comfy_extras/nodes_ideogram4.py` + (which supplies `Ideogram4Scheduler`); the same PR added the generic + `DualModelGuider` and `CFGOverride` nodes, which live in + `comfy_extras/nodes_custom_sampler.py`. + 2. From `Comfy-Org/Ideogram-4` download into + `ComfyUI/models/diffusion_models/`: `ideogram4_fp8_scaled.safetensors` + (9.28 GB) AND `ideogram4_unconditional_fp8_scaled.safetensors` (9.28 GB). + Lower-VRAM alternatives in the same folder: `*_nvfp4_mixed.safetensors` + (5.49 GB each) or `*_int8_convrot.safetensors` (9.58 GB each). + 3. Download `qwen3vl_8b_fp8_scaled.safetensors` (10.59 GB; or + `qwen3vl_8b_nvfp4.safetensors`, 6.31 GB) into + `ComfyUI/models/text_encoders/`. The shipped template also lists an + optional `gemma4_e4b_it_fp8_scaled.safetensors` from `Comfy-Org/gemma-4` + for local prompt building. + 4. Download `flux2-vae.safetensors` (0.34 GB, from Comfy-Org/Ideogram-4 or + Comfy-Org/flux2-dev) into `ComfyUI/models/vae/`. + 5. Put LoRA .safetensors in `ComfyUI/models/loras/` (no subfolder + convention enforced). + 6. Load the built-in template `image_ideogram4_t2i` (Workflow → Browse + Templates), or the workflow from + docs.comfy.org/tutorials/image/ideogram/ideogram-v4. + 7. Insert a `LoraLoaderModelOnly` between the main `UNETLoader` and the + `DualModelGuider`'s conditional model input. There is no CLIP-side LoRA + here — the text encoder is a frozen Qwen3-VL — so model-only is the right + loader. Strength ~0.5–0.8. + 8. Do NOT reuse one patched model object for both guider inputs; it + silently does nothing. Patching the negative branch requires a second, + physically separate model file with its own loader. + 9. Write the prompt as a JSON caption into `CLIPTextEncode` (the CLIPLoader + type is `ideogram4`); keep `ConditioningZeroOut` feeding the unconditional + branch. + + Non-ComfyUI paths: diffusers — + `Ideogram4Pipeline.from_pretrained("ideogram-ai/ideogram-4-nf4")` (the + docs' own example; `ideogram-ai/ideogram-4-nf4-diffusers` also exists), + then `pipeline.load_lora_weights(...)`. Training — `ostris/ai-toolkit` + (built-in `extensions_built_in/diffusion_models/ideogram4/` plus an + `Ideogram4Captioner` that writes JSON captions) or diffusers + `examples/dreambooth/train_dreambooth_lora_ideogram4.py`. The official HF + weights are GATED (`gated: auto`): accept the license on the model page and + authenticate (`hf auth login` / `HF_TOKEN`) or downloads fail. + +─────────────────────────────────────────────────────────────────────────────── + Scraped by lorakit on 2026-09-05 01:49 UTC. + Stats and description are the author's; pros/cons are derived from the + evidence tagged beside each line. +─────────────────────────────────────────────────────────────────────────────── diff --git a/gray-screen-bypass-lora-and-workflow/metadata.json b/gray-screen-bypass-lora-and-workflow/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..12e495a5c44b67a4c40692ea9b40c84b86e677af --- /dev/null +++ b/gray-screen-bypass-lora-and-workflow/metadata.json @@ -0,0 +1,127 @@ +{ + "name": "Gray Screen bypass (LoRA and workflow)", + "slug": "gray-screen-bypass-lora-and-workflow", + "base_model": "Ideogram 4", + "media_type": "image", + "model_id": 2750357, + "version_id": 3094072, + "version_name": "v1.0", + "source": "civitai", + "url": "https://civitai.com/models/2750357", + "repo_id": null, + "file_path": null, + "author": "wystanbelov785", + "category": { + "key": "style", + "label": "Visual style" + }, + "modalities": [ + "t2i" + ], + "primary_modality": "t2i", + "checkpoint_family": "transformer", + "trigger_words": [], + "recommended_settings": {}, + "chaining": { + "author": { + "stacks": "not stated", + "notes": "", + "avoid": [], + "requires": [], + "order": "", + "strength": "", + "stated": false + }, + "general_rules": [ + "Chain `LoraLoaderModelOnly` nodes in series off the CONDITIONAL model loader, ahead of the guider. DeverStyle states its samples were made \"by setting the LORA @ 0.6 strength on the model only, not on the model_negative\".", + "Model-only loaders throughout. The text encoder is a frozen Qwen3-VL-8B with no LoRA-patchable CLIP branch, so the model+clip `LoraLoader` has nothing useful to do.", + "To affect the unconditional/negative branch you must load a second, physically distinct model file and patch that one. Passing the same patched model object into both DualModelGuider inputs is a verified no-op (Comfy-Org issue #14521; kijai confirmed it is expected behaviour).", + "Trigger words go INSIDE the JSON caption, in the semantically matching field, not prepended to a prose string. DeverStyle instructs putting `dvr_arch_tv` in `style_description.art_style` (and in the content field); the Kiki LoRA card describes using `art_style: \"Studio Ghibli style\"` with `medium: \"anime_screencap\"`.", + "Two versions of the same LoRA can be stacked at reduced strengths: zjourney's author recommends v1 @ 0.35 + v2 @ 0.45 over either alone at full strength.", + "TurboTime changes the graph, not just the strength: it runs at cfg 1.0 and the reference community workflow swaps `DualModelGuider` for a single-model `CFGGuider`, dropping the unconditional model. Re-tune any stacked style LoRA downward.", + "A LoRA can be applied to only part of the run. The Gray Screen bypass LoRA is used at NEGATIVE weight (about -0.25) and, in its shipped workflows, only on the first step and then disabled; its author warns that \"LoRA being active through the whole generation could lead to degradation\".", + "The main 9.3B DiT and what essentially every Ideogram 4 LoRA targets. diffusers' README_ideogram4.md is explicit: \"This trainer adds LoRA to the conditional `transformer` only; the unconditional one stays frozen.\" In ComfyUI this is `ideogram4_fp8_scaled.safetensors` (9.28 GB) or its int8_convrot / nvfp4_mixed equivalents." + ] + }, + "pros": [ + "Compact at 202 MB — cheap to stack with other LoRAs. [source: file size]" + ], + "cons": [ + "No trigger word — the LoRA is always on once loaded, so strength is the only control. [source: model metadata]", + "Modest adoption (241 downloads) — less community feedback to rely on. [source: source stats]" + ], + "file": { + "name": "Gray_000002000.safetensors", + "size_bytes": 211734024, + "size_mb": 201.93, + "sha256": "0292898e3f3cd3edbb15bcff80e81a81a5f4e5fc221438fc9fd0b5fcab89ae78" + }, + "stats": { + "downloads": 241, + "thumbs_up": 22, + "thumbs_down": 0 + }, + "popularity": { + "downloads": 241, + "thumbs_up": 22, + "thumbs_down": 0, + "comments": 0, + "approval": "100% positive (22/22)", + "age_days": 63.7, + "downloads_per_day": 3.78, + "rank_by_downloads": 27, + "rank_by_momentum": 8, + "of_total": 35, + "adoption_tier": "Niche", + "momentum": "Trending" + }, + "selection_score": 505.0, + "selection_reasons": [ + "241 downloads", + "22 thumbs-up", + "top-ranked for its category" + ], + "example_media": [ + { + "file": "01_image.jpeg", + "type": "image", + "resolution": "1824x1248", + "prompt": "" + }, + { + "file": "02_image.jpeg", + "type": "image", + "resolution": "2128x1456", + "prompt": "" + }, + { + "file": "03_image.jpeg", + "type": "image", + "resolution": "2128x1456", + "prompt": "" + }, + { + "file": "04_image.jpeg", + "type": "image", + "resolution": "1024x1024", + "prompt": "" + }, + { + "file": "05_image.jpeg", + "type": "image", + "resolution": "1024x1024", + "prompt": "" + }, + { + "file": "06_image.jpeg", + "type": "image", + "resolution": "2432x1664", + "prompt": "" + } + ], + "tags": [ + "tool" + ], + "scraped_at": "2026-09-05T01:49:40+00:00", + "path": "ideogram-4/gray-screen-bypass-lora-and-workflow" +} diff --git a/mihoyo-style/example_images/01_image.jpeg b/mihoyo-style/example_images/01_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..0dfe745a4e2682450b77de109e29f1e1009674d1 --- /dev/null +++ b/mihoyo-style/example_images/01_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2ce51733f5e5266f9b2679e3d619b7254119402a2fd4aef298582bb36eb50c22 +size 3071124 diff --git a/mihoyo-style/example_images/02_image.jpeg b/mihoyo-style/example_images/02_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..fd8cce243f3fb1fe636095320ae28c66fc9fca60 --- /dev/null +++ b/mihoyo-style/example_images/02_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b47742dd954d1d7e60706a56a562265855c64191eab54005b9446668ef3a37c7 +size 1294644 diff --git a/mihoyo-style/example_images/03_image.jpeg b/mihoyo-style/example_images/03_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..37790669f6018ede651a736d198a77db20d29ab7 --- /dev/null +++ b/mihoyo-style/example_images/03_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bc3aef07a43c0f78e5289ea34e3aa141d4ec4e8a7ef48c9484e7f55ff02cc757 +size 1293659 diff --git a/mihoyo-style/example_images/04_image.jpeg b/mihoyo-style/example_images/04_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..47bacbb119cf7aeb154b491a9bf0f1aff7bdd5f3 --- /dev/null +++ b/mihoyo-style/example_images/04_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8cc5eb5e59aec09a969ced1c5f70cf18ac4c8e8616efc742e8c52ee645594e8a +size 1319578 diff --git a/mihoyo-style/example_images/05_image.jpeg b/mihoyo-style/example_images/05_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..51bec0abba89adf2e4e09376cf6ead8428cc67c4 --- /dev/null +++ b/mihoyo-style/example_images/05_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3ae027ba19df02aec57d965d63727e5281f77837410be8c9a092bdf78edc85c9 +size 3525653 diff --git a/mihoyo-style/example_images/06_image.jpeg b/mihoyo-style/example_images/06_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..2a9c70acd4f3f7709e49080b5aa3e5ac7d33f164 --- /dev/null +++ b/mihoyo-style/example_images/06_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0ee8332085b254bd42c64975dfc69d8a22ab1081f9c0060cd08ad7af2805b60c +size 3312892 diff --git a/mihoyo-style/idegoram4_miha_re3_000003000.safetensors b/mihoyo-style/idegoram4_miha_re3_000003000.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c293d4e472823701d7225ae3dc87f6539f8e2b62 --- /dev/null +++ b/mihoyo-style/idegoram4_miha_re3_000003000.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8d6b75860935f6af02b2f89d5ae6a38e46d1312b60d12b3433db03da5c93ed86 +size 211734008 diff --git a/mihoyo-style/info.txt b/mihoyo-style/info.txt new file mode 100644 index 0000000000000000000000000000000000000000..4d1ceb7fddb4ce99ee17bd13d66d957ec4172380 --- /dev/null +++ b/mihoyo-style/info.txt @@ -0,0 +1,260 @@ +═══════════════════════════════════════════════════════════════════════════════ + MIHOYO STYLE +═══════════════════════════════════════════════════════════════════════════════ + +─────────────────────────────────────────────────────────────────────────────── + IDENTITY +─────────────────────────────────────────────────────────────────────────────── + Base model Ideogram 4 + Generation mode not declared by the author + Checkpoint not determined — verify before use + Category Visual style + Author allpleoleo439 + Source https://civitai.com/models/995865 + Provenance CivitAI · model 995865 · version 3061630 + +─────────────────────────────────────────────────────────────────────────────── + WEIGHTS +─────────────────────────────────────────────────────────────────────────────── + File idegoram4_miha_re3_000003000.safetensors + Version v1.0 | ideogram4 + Size 201.9 MB + Format SafeTensor / unknown precision + SHA256 8d6b75860935f6af02b2f89d5ae6a38e46d1312b60d12b3433db03da5c93ed86 + Published 2026-06-23 + +─────────────────────────────────────────────────────────────────────────────── + TRIGGER WORDS +─────────────────────────────────────────────────────────────────────────────── + No trigger word. The LoRA applies as soon as it is loaded — + strength is the only way to dial the effect up or down. + +─────────────────────────────────────────────────────────────────────────────── + RECOMMENDED SETTINGS +─────────────────────────────────────────────────────────────────────────────── + The author did not publish specific settings. + No author settings were published. Community baseline: steps 48 is the + reference default: docs/inference.md says "`V4_QUALITY_48` is the default" + and the pipeline's `num_steps` default is 48; diffusers documents the same + 48-step default. 20 is the practical ComfyUI default — the `V4_DEFAULT_20` + preset and the value the shipped template `image_ideogram4_t2i.json` + selects (its preset combo defaults to "Default"). 12 = `V4_TURBO_12`. With + the TurboTime LoRA: 8 in the nomadoor community workflow; the model card + claims usable output down to 2., cfg A guidance SCHEDULE, not one number, + and the negative branch comes from a second model rather than a negative + prompt. Verified from src/ideogram4/sampler_configs.py: V4_QUALITY_48 = 45 + steps @ gw 7 then 3 polish steps @ gw 3; V4_DEFAULT_20 = 18 @ 7 + 2 @ 3; + V4_TURBO_12 = 11 @ 7 + 1 @ 3. Pipeline default `guidance_scale` = 7.0 when + no schedule is given. In ComfyUI the shipped template sets + `DualModelGuider` cfg 7 plus a `CFGOverride` node with widgets [3, 0.7, 1] + (cfg 3 from 0.7→1.0 of the run); kijai recommends starting the override at + 0.7, not 0.9. TurboTime: cfg 1.0 / no CFG., sampler Euler flow-matching. In + ComfyUI: `KSamplerSelect` → euler, driven by `SamplerCustomAdvanced`, with + the model-specific `Ideogram4Scheduler` (logit-normal sigma schedule). Its + params are `mu` and `std`, not the usual `shift`. Reference presets: mu 0.0 + / std 1.5 (48-step), mu 0.0 / std 1.75 (20-step), mu 0.5 / std 1.75 + (12-step). The Ideogram4Scheduler node's own widget defaults are steps 20, + mu 0.5, std 1.75, but the shipped template overrides mu/std/steps from the + preset JSON, so out of the box it runs 20 / mu 0.0 / std 1.75. mu is auto- + shifted by resolution in both implementations: mean = mu + + 0.5*log((w*h)/(512*512))., resolution 1024×1024 default (pipeline + `height`/`width` default 1024). Any multiple of 16 per side from 256 to + 2048, aspect ratios up to 6:1 or 1:6 — one model covers all of it. + 2048×2048 with V4_QUALITY_48 for maximum quality. Buckets documented in + docs/inference.md: 1536×1024 (3:2), 1024×1536 (2:3), 1920×1088 (~16:9), + 2048×768 (~21:9), 1024×1792 (~9:16), 1600×400 (4:1)., lora_strength No + single official number; these are author-stated values. DeverStyle Archer: + 0.6, and explicitly "on the model only, not on the model_negative". + zjourney v2: 0.55 standalone (range 0.35–0.6), or v1 @ 0.35 + v2 @ 0.45 + stacked. Furry Enhancer: the model page says 0.6–0.8, but its own version + notes ask for 0.7–0.9 (v1.94) and 0.8–0.95 (v2.8). TurboTime: 0.6 in the + nomadoor workflow (the author gives no figure). So ~0.5–0.9 is the + realistic band, read the specific card.. + +─────────────────────────────────────────────────────────────────────────────── + ADOPTION & POPULARITY +─────────────────────────────────────────────────────────────────────────────── + Downloads 309 + Rating 100% positive (26/26) + Comments 0 + Published 74 days ago + Download rate 0.7/day since release + + Adoption rank #25 of 35 in this collection (Modest) + Momentum rank #26 of 35 by download rate (Long tail) + + Adoption tier is the percentile of total downloads within this collection; + momentum is the percentile of downloads-per-day since release. Momentum is + ranked rather than measured against a fixed rate, because this base model + is itself new and every LoRA looks fast on an absolute scale. Both numbers + are CivitAI's, read on the scrape date at the foot of this file. + +─────────────────────────────────────────────────────────────────────────────── + CHAINING / STACKING +─────────────────────────────────────────────────────────────────────────────── + Stacks with others not stated + + The author gave no chaining guidance. The rules below are the + general Ideogram 4 ones, not anything specific to this LoRA. + + General Ideogram 4 rules that apply here (see REFERENCE.txt): + + • Chain `LoraLoaderModelOnly` nodes in series off the CONDITIONAL model + loader, ahead of the guider. DeverStyle states its samples were made + "by setting the LORA @ 0.6 strength on the model only, not on the + model_negative". + • Model-only loaders throughout. The text encoder is a frozen Qwen3-VL-8B + with no LoRA-patchable CLIP branch, so the model+clip `LoraLoader` has + nothing useful to do. + • To affect the unconditional/negative branch you must load a second, + physically distinct model file and patch that one. Passing the same + patched model object into both DualModelGuider inputs is a verified no- + op (Comfy-Org issue #14521; kijai confirmed it is expected behaviour). + • Trigger words go INSIDE the JSON caption, in the semantically matching + field, not prepended to a prose string. DeverStyle instructs putting + `dvr_arch_tv` in `style_description.art_style` (and in the content + field); the Kiki LoRA card describes using `art_style: "Studio Ghibli + style"` with `medium: "anime_screencap"`. + • Two versions of the same LoRA can be stacked at reduced strengths: + zjourney's author recommends v1 @ 0.35 + v2 @ 0.45 over either alone at + full strength. + • TurboTime changes the graph, not just the strength: it runs at cfg 1.0 + and the reference community workflow swaps `DualModelGuider` for a + single-model `CFGGuider`, dropping the unconditional model. Re-tune any + stacked style LoRA downward. + • A LoRA can be applied to only part of the run. The Gray Screen bypass + LoRA is used at NEGATIVE weight (about -0.25) and, in its shipped + workflows, only on the first step and then disabled; its author warns + that "LoRA being active through the whole generation could lead to + degradation". + • The author never states which weight family this targets. Silent + degradation, not hard errors, in every case observed. (1) Loading a + conditional-trained LoRA onto the unconditional transformer succeeds + silently because the architectures match; it perturbs the negative + branch instead of the positive one. (2) Verified ComfyUI trap: in `Du + +─────────────────────────────────────────────────────────────────────────────── + PROS +─────────────────────────────────────────────────────────────────────────────── + + Compact at 202 MB — cheap to stack with other LoRAs. [source: file + size] + +─────────────────────────────────────────────────────────────────────────────── + CONS & CAVEATS +─────────────────────────────────────────────────────────────────────────────── + - Author labels this experimental — results vary shot to shot. [source: + description] + - No trigger word — the LoRA is always on once loaded, so strength is the + only control. [source: model metadata] + - Author never states which generation mode it was trained for; test + against your own workflow before relying on it. [source: description] + - Modest adoption (309 downloads) — less community feedback to rely on. + [source: source stats] + +─────────────────────────────────────────────────────────────────────────────── + COMPATIBILITY WARNING +─────────────────────────────────────────────────────────────────────────────── + Silent degradation, not hard errors, in every case observed. (1) Loading a + conditional-trained LoRA onto the unconditional transformer succeeds + silently because the architectures match; it perturbs the negative branch + instead of the positive one. (2) Verified ComfyUI trap: in + `DualModelGuider`, passing the SAME model object as both the conditional + model and the LoRA-patched unconditional model is a no-op. Comfy-Org issue + #14521 tabulates it — case 3 (DualModelGuider, conditional model as + positive, conditional model + LoRA as negative) was "exactly identical" to + plain conditional-only CFG, while case 4 with the real unconditional model + differed. kijai confirmed it is expected: "there's no (working) mechanic to + swap LoRA on each step on the same model, it would only work if the model + was a different one." To patch the negative branch you must load a + genuinely separate model file. (3) A diffusers/PEFT-prefixed LoRA in + ComfyUI is expected to produce only a "lora key not loaded" console warning + and no visible effect — this follows from the absent key mapping (see + checkpoint_families) rather than from a published Ideogram-4 bug report. + (4) Cross-quantization loading (fp8-trained LoRA onto nvfp4/nf4/GGUF) + generally works with some precision loss; int8_convrot had a specific, now- + fixed bug (see gotchas). + +─────────────────────────────────────────────────────────────────────────────── + AUTHOR'S DESCRIPTION +─────────────────────────────────────────────────────────────────────────────── + このLORAを使用すると、有名なゲーム会社miHoYoのゲームスタイルに似た画像を生成することができます。最近、FluxのRedux機能を試しているた + め、このLORAを制作しました。ただし、現在の第一版はまだ完全に安定していません。いくつか試行を重ねた結果、比較的安定したバージョンを公開しています。 + Using this LORA, you can generate images in a style similar to the games + produced by the well-known gaming company miHoYo. Recently, I’ve been + experimenting with Flux’s Redux functionality, which led to the creation of + this LORA. However, the first version is still somewhat unstable. After + several attempts, I have released a relatively stable version. + +─────────────────────────────────────────────────────────────────────────────── + EXAMPLE MEDIA +─────────────────────────────────────────────────────────────────────────────── + 6 preview(s) in ./example_images/ — community results for + this LoRA. Videos keep a matching _poster.jpg still frame. + + No prompts are recorded below: CivitAI's public API flags these posts as + having generation metadata but does not return it, so the prompts cannot be + scraped. Open the model page to read them. + + 01_image.jpeg (image, 1088x1936) + 02_image.jpeg (image, 768x1376) + 03_image.jpeg (image, 768x1376) + 04_image.jpeg (image, 768x1376) + 05_image.jpeg (image, 1088x1936) + 06_image.jpeg (image, 1680x1264) + +─────────────────────────────────────────────────────────────────────────────── + HOW TO USE (COMFYUI) +─────────────────────────────────────────────────────────────────────────────── + Use ComfyUI's NATIVE Ideogram 4 support (Comfy-Org/ComfyUI PR #14259 "feat: + Add Ideogram4 model support", merged 2026-06-03) — not the official + `ComfyUI-Ideogram4` custom-node wrapper, whose nodes (Magic Prompt / + Pipeline Loader / Generate) wrap Ideogram's reference pipeline and contain + no LoRA input at all (verified: no occurrence of "lora" in its nodes.py). + + 1. Update ComfyUI to a build containing `comfy_extras/nodes_ideogram4.py` + (which supplies `Ideogram4Scheduler`); the same PR added the generic + `DualModelGuider` and `CFGOverride` nodes, which live in + `comfy_extras/nodes_custom_sampler.py`. + 2. From `Comfy-Org/Ideogram-4` download into + `ComfyUI/models/diffusion_models/`: `ideogram4_fp8_scaled.safetensors` + (9.28 GB) AND `ideogram4_unconditional_fp8_scaled.safetensors` (9.28 GB). + Lower-VRAM alternatives in the same folder: `*_nvfp4_mixed.safetensors` + (5.49 GB each) or `*_int8_convrot.safetensors` (9.58 GB each). + 3. Download `qwen3vl_8b_fp8_scaled.safetensors` (10.59 GB; or + `qwen3vl_8b_nvfp4.safetensors`, 6.31 GB) into + `ComfyUI/models/text_encoders/`. The shipped template also lists an + optional `gemma4_e4b_it_fp8_scaled.safetensors` from `Comfy-Org/gemma-4` + for local prompt building. + 4. Download `flux2-vae.safetensors` (0.34 GB, from Comfy-Org/Ideogram-4 or + Comfy-Org/flux2-dev) into `ComfyUI/models/vae/`. + 5. Put LoRA .safetensors in `ComfyUI/models/loras/` (no subfolder + convention enforced). + 6. Load the built-in template `image_ideogram4_t2i` (Workflow → Browse + Templates), or the workflow from + docs.comfy.org/tutorials/image/ideogram/ideogram-v4. + 7. Insert a `LoraLoaderModelOnly` between the main `UNETLoader` and the + `DualModelGuider`'s conditional model input. There is no CLIP-side LoRA + here — the text encoder is a frozen Qwen3-VL — so model-only is the right + loader. Strength ~0.5–0.8. + 8. Do NOT reuse one patched model object for both guider inputs; it + silently does nothing. Patching the negative branch requires a second, + physically separate model file with its own loader. + 9. Write the prompt as a JSON caption into `CLIPTextEncode` (the CLIPLoader + type is `ideogram4`); keep `ConditioningZeroOut` feeding the unconditional + branch. + + Non-ComfyUI paths: diffusers — + `Ideogram4Pipeline.from_pretrained("ideogram-ai/ideogram-4-nf4")` (the + docs' own example; `ideogram-ai/ideogram-4-nf4-diffusers` also exists), + then `pipeline.load_lora_weights(...)`. Training — `ostris/ai-toolkit` + (built-in `extensions_built_in/diffusion_models/ideogram4/` plus an + `Ideogram4Captioner` that writes JSON captions) or diffusers + `examples/dreambooth/train_dreambooth_lora_ideogram4.py`. The official HF + weights are GATED (`gated: auto`): accept the license on the model page and + authenticate (`hf auth login` / `HF_TOKEN`) or downloads fail. + +─────────────────────────────────────────────────────────────────────────────── + Scraped by lorakit on 2026-09-05 01:49 UTC. + Stats and description are the author's; pros/cons are derived from the + evidence tagged beside each line. +─────────────────────────────────────────────────────────────────────────────── diff --git a/mihoyo-style/metadata.json b/mihoyo-style/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..0a52eec46e3f7166cd761c18e4bf248fb9cb07bc --- /dev/null +++ b/mihoyo-style/metadata.json @@ -0,0 +1,131 @@ +{ + "name": "MiHoYo Style", + "slug": "mihoyo-style", + "base_model": "Ideogram 4", + "media_type": "image", + "model_id": 995865, + "version_id": 3061630, + "version_name": "v1.0 | ideogram4", + "source": "civitai", + "url": "https://civitai.com/models/995865", + "repo_id": null, + "file_path": null, + "author": "allpleoleo439", + "category": { + "key": "style", + "label": "Visual style" + }, + "modalities": [], + "primary_modality": "unspecified", + "checkpoint_family": "unknown", + "trigger_words": [], + "recommended_settings": {}, + "chaining": { + "author": { + "stacks": "not stated", + "notes": "", + "avoid": [], + "requires": [], + "order": "", + "strength": "", + "stated": false + }, + "general_rules": [ + "Chain `LoraLoaderModelOnly` nodes in series off the CONDITIONAL model loader, ahead of the guider. DeverStyle states its samples were made \"by setting the LORA @ 0.6 strength on the model only, not on the model_negative\".", + "Model-only loaders throughout. The text encoder is a frozen Qwen3-VL-8B with no LoRA-patchable CLIP branch, so the model+clip `LoraLoader` has nothing useful to do.", + "To affect the unconditional/negative branch you must load a second, physically distinct model file and patch that one. Passing the same patched model object into both DualModelGuider inputs is a verified no-op (Comfy-Org issue #14521; kijai confirmed it is expected behaviour).", + "Trigger words go INSIDE the JSON caption, in the semantically matching field, not prepended to a prose string. DeverStyle instructs putting `dvr_arch_tv` in `style_description.art_style` (and in the content field); the Kiki LoRA card describes using `art_style: \"Studio Ghibli style\"` with `medium: \"anime_screencap\"`.", + "Two versions of the same LoRA can be stacked at reduced strengths: zjourney's author recommends v1 @ 0.35 + v2 @ 0.45 over either alone at full strength.", + "TurboTime changes the graph, not just the strength: it runs at cfg 1.0 and the reference community workflow swaps `DualModelGuider` for a single-model `CFGGuider`, dropping the unconditional model. Re-tune any stacked style LoRA downward.", + "A LoRA can be applied to only part of the run. The Gray Screen bypass LoRA is used at NEGATIVE weight (about -0.25) and, in its shipped workflows, only on the first step and then disabled; its author warns that \"LoRA being active through the whole generation could lead to degradation\".", + "The author never states which weight family this targets. Silent degradation, not hard errors, in every case observed. (1) Loading a conditional-trained LoRA onto the unconditional transformer succeeds silently because the architectures match; it perturbs the negative branch instead of the positive one. (2) Verified ComfyUI trap: in `Du" + ] + }, + "pros": [ + "Compact at 202 MB — cheap to stack with other LoRAs. [source: file size]" + ], + "cons": [ + "Author labels this experimental — results vary shot to shot. [source: description]", + "No trigger word — the LoRA is always on once loaded, so strength is the only control. [source: model metadata]", + "Author never states which generation mode it was trained for; test against your own workflow before relying on it. [source: description]", + "Modest adoption (309 downloads) — less community feedback to rely on. [source: source stats]" + ], + "file": { + "name": "idegoram4_miha_re3_000003000.safetensors", + "size_bytes": 211734008, + "size_mb": 201.93, + "sha256": "8d6b75860935f6af02b2f89d5ae6a38e46d1312b60d12b3433db03da5c93ed86" + }, + "stats": { + "downloads": 309, + "thumbs_up": 26, + "thumbs_down": 0 + }, + "popularity": { + "downloads": 309, + "thumbs_up": 26, + "thumbs_down": 0, + "comments": 0, + "approval": "100% positive (26/26)", + "age_days": 73.8, + "downloads_per_day": 0.75, + "rank_by_downloads": 25, + "rank_by_momentum": 26, + "of_total": 35, + "adoption_tier": "Modest", + "momentum": "Long tail" + }, + "selection_score": 621.0, + "selection_reasons": [ + "309 downloads", + "26 thumbs-up", + "top-ranked for its category" + ], + "example_media": [ + { + "file": "01_image.jpeg", + "type": "image", + "resolution": "1088x1936", + "prompt": "" + }, + { + "file": "02_image.jpeg", + "type": "image", + "resolution": "768x1376", + "prompt": "" + }, + { + "file": "03_image.jpeg", + "type": "image", + "resolution": "768x1376", + "prompt": "" + }, + { + "file": "04_image.jpeg", + "type": "image", + "resolution": "768x1376", + "prompt": "" + }, + { + "file": "05_image.jpeg", + "type": "image", + "resolution": "1088x1936", + "prompt": "" + }, + { + "file": "06_image.jpeg", + "type": "image", + "resolution": "1680x1264", + "prompt": "" + } + ], + "tags": [ + "china", + "style", + "anime", + "game", + "mmorpg" + ], + "scraped_at": "2026-09-05T01:49:29+00:00", + "path": "ideogram-4/mihoyo-style" +} diff --git a/pagani-zonda-r-2009/example_images/01_image.jpeg b/pagani-zonda-r-2009/example_images/01_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..be66579632b128ab6c7f0919ba754ff5b3bf9a90 --- /dev/null +++ b/pagani-zonda-r-2009/example_images/01_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:29bd55bbe2d519f1d7ee136bdbb8acb73faf8b0a24b2689a3583a4b84c5e7ea2 +size 2865861 diff --git a/pagani-zonda-r-2009/example_images/02_image.jpeg b/pagani-zonda-r-2009/example_images/02_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..248dbc1fdb22dba5edbdec45966b847e54b2ce9c --- /dev/null +++ b/pagani-zonda-r-2009/example_images/02_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9c721bd78b30b90b00d3a2072b9004eb26adbb23f422016deb637f0534b686a8 +size 2643595 diff --git a/pagani-zonda-r-2009/example_images/03_image.jpeg b/pagani-zonda-r-2009/example_images/03_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..5338183c358aa1738de31bbb7a3709bd9d0389e2 --- /dev/null +++ b/pagani-zonda-r-2009/example_images/03_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:172980329748de7f76ada280689fae25133d77bafab7a33be9a935c5d3a129df +size 2389603 diff --git a/pagani-zonda-r-2009/example_images/04_image.jpeg b/pagani-zonda-r-2009/example_images/04_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..1327d5cdbc7392eda74ddc4120e1fcbfd34cfcc2 --- /dev/null +++ b/pagani-zonda-r-2009/example_images/04_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:37b86dca4124a7f177ac9ffb1e188c49c3ca8f30779e95cebd6c75b75a83d029 +size 2641022 diff --git a/pagani-zonda-r-2009/example_images/05_image.jpeg b/pagani-zonda-r-2009/example_images/05_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..028780c8ec4b232d5f1028bf1eb92d9614024d5e --- /dev/null +++ b/pagani-zonda-r-2009/example_images/05_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eaabe50fa7091b1a0ad447543f1e3df44654ac472595399982724a119a30f649 +size 2525407 diff --git a/pagani-zonda-r-2009/example_images/06_image.jpeg b/pagani-zonda-r-2009/example_images/06_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..afb9c01b6705536db099b49bb3dc334299c2fabe --- /dev/null +++ b/pagani-zonda-r-2009/example_images/06_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c4ec6b66d3555b215154510cd7946621cc4f51efad240c62334d65fe3b1d8da5 +size 2366787 diff --git a/pagani-zonda-r-2009/info.txt b/pagani-zonda-r-2009/info.txt new file mode 100644 index 0000000000000000000000000000000000000000..ac447c4c80f6a7dfaa470c682d867e9c8407adf7 --- /dev/null +++ b/pagani-zonda-r-2009/info.txt @@ -0,0 +1,341 @@ +═══════════════════════════════════════════════════════════════════════════════ + PAGANI ZONDA R - 2009 +═══════════════════════════════════════════════════════════════════════════════ + +─────────────────────────────────────────────────────────────────────────────── + IDENTITY +─────────────────────────────────────────────────────────────────────────────── + Base model Ideogram 4 + Generation mode not declared by the author + Checkpoint not determined — verify before use + Category Speed / step-reduction + Author mastermerlin + Source https://civitai.com/models/2710899 + Provenance CivitAI · model 2710899 · version 3053389 + +─────────────────────────────────────────────────────────────────────────────── + WEIGHTS +─────────────────────────────────────────────────────────────────────────────── + File mastermerlin_ideogram4-fp8_Pagani_Zonda_R_2009_final.safetensors + Version v1.0 IDEOGRAM4 + Size 162.6 MB + Format SafeTensor / unknown precision + SHA256 e93e9b2bbf582dbe0e8b744eec81e5fd282f9538eaa2d5f72334cff703a2e6d5 + Published 2026-06-21 + +─────────────────────────────────────────────────────────────────────────────── + TRIGGER WORDS +─────────────────────────────────────────────────────────────────────────────── + › "P4ganiZondaR2009" + +─────────────────────────────────────────────────────────────────────────────── + RECOMMENDED SETTINGS +─────────────────────────────────────────────────────────────────────────────── + The author did not publish specific settings. + No author settings were published. Community baseline: steps 48 is the + reference default: docs/inference.md says "`V4_QUALITY_48` is the default" + and the pipeline's `num_steps` default is 48; diffusers documents the same + 48-step default. 20 is the practical ComfyUI default — the `V4_DEFAULT_20` + preset and the value the shipped template `image_ideogram4_t2i.json` + selects (its preset combo defaults to "Default"). 12 = `V4_TURBO_12`. With + the TurboTime LoRA: 8 in the nomadoor community workflow; the model card + claims usable output down to 2., cfg A guidance SCHEDULE, not one number, + and the negative branch comes from a second model rather than a negative + prompt. Verified from src/ideogram4/sampler_configs.py: V4_QUALITY_48 = 45 + steps @ gw 7 then 3 polish steps @ gw 3; V4_DEFAULT_20 = 18 @ 7 + 2 @ 3; + V4_TURBO_12 = 11 @ 7 + 1 @ 3. Pipeline default `guidance_scale` = 7.0 when + no schedule is given. In ComfyUI the shipped template sets + `DualModelGuider` cfg 7 plus a `CFGOverride` node with widgets [3, 0.7, 1] + (cfg 3 from 0.7→1.0 of the run); kijai recommends starting the override at + 0.7, not 0.9. TurboTime: cfg 1.0 / no CFG., sampler Euler flow-matching. In + ComfyUI: `KSamplerSelect` → euler, driven by `SamplerCustomAdvanced`, with + the model-specific `Ideogram4Scheduler` (logit-normal sigma schedule). Its + params are `mu` and `std`, not the usual `shift`. Reference presets: mu 0.0 + / std 1.5 (48-step), mu 0.0 / std 1.75 (20-step), mu 0.5 / std 1.75 + (12-step). The Ideogram4Scheduler node's own widget defaults are steps 20, + mu 0.5, std 1.75, but the shipped template overrides mu/std/steps from the + preset JSON, so out of the box it runs 20 / mu 0.0 / std 1.75. mu is auto- + shifted by resolution in both implementations: mean = mu + + 0.5*log((w*h)/(512*512))., resolution 1024×1024 default (pipeline + `height`/`width` default 1024). Any multiple of 16 per side from 256 to + 2048, aspect ratios up to 6:1 or 1:6 — one model covers all of it. + 2048×2048 with V4_QUALITY_48 for maximum quality. Buckets documented in + docs/inference.md: 1536×1024 (3:2), 1024×1536 (2:3), 1920×1088 (~16:9), + 2048×768 (~21:9), 1024×1792 (~9:16), 1600×400 (4:1)., lora_strength No + single official number; these are author-stated values. DeverStyle Archer: + 0.6, and explicitly "on the model only, not on the model_negative". + zjourney v2: 0.55 standalone (range 0.35–0.6), or v1 @ 0.35 + v2 @ 0.45 + stacked. Furry Enhancer: the model page says 0.6–0.8, but its own version + notes ask for 0.7–0.9 (v1.94) and 0.8–0.95 (v2.8). TurboTime: 0.6 in the + nomadoor workflow (the author gives no figure). So ~0.5–0.9 is the + realistic band, read the specific card.. + +─────────────────────────────────────────────────────────────────────────────── + ADOPTION & POPULARITY +─────────────────────────────────────────────────────────────────────────────── + Downloads 144 + Rating 100% positive (7/7) + Comments 0 + Published 76 days ago + Download rate 0.4/day since release + + Adoption rank #32 of 35 in this collection (Niche) + Momentum rank #34 of 35 by download rate (Long tail) + + Adoption tier is the percentile of total downloads within this collection; + momentum is the percentile of downloads-per-day since release. Momentum is + ranked rather than measured against a fixed rate, because this base model + is itself new and every LoRA looks fast on an absolute scale. Both numbers + are CivitAI's, read on the scrape date at the foot of this file. + +─────────────────────────────────────────────────────────────────────────────── + CHAINING / STACKING +─────────────────────────────────────────────────────────────────────────────── + Stacks with others not stated + + The author gave no chaining guidance. The rules below are the + general Ideogram 4 ones, not anything specific to this LoRA. + + General Ideogram 4 rules that apply here (see REFERENCE.txt): + + • Chain `LoraLoaderModelOnly` nodes in series off the CONDITIONAL model + loader, ahead of the guider. DeverStyle states its samples were made + "by setting the LORA @ 0.6 strength on the model only, not on the + model_negative". + • Model-only loaders throughout. The text encoder is a frozen Qwen3-VL-8B + with no LoRA-patchable CLIP branch, so the model+clip `LoraLoader` has + nothing useful to do. + • To affect the unconditional/negative branch you must load a second, + physically distinct model file and patch that one. Passing the same + patched model object into both DualModelGuider inputs is a verified no- + op (Comfy-Org issue #14521; kijai confirmed it is expected behaviour). + • Trigger words go INSIDE the JSON caption, in the semantically matching + field, not prepended to a prose string. DeverStyle instructs putting + `dvr_arch_tv` in `style_description.art_style` (and in the content + field); the Kiki LoRA card describes using `art_style: "Studio Ghibli + style"` with `medium: "anime_screencap"`. + • Two versions of the same LoRA can be stacked at reduced strengths: + zjourney's author recommends v1 @ 0.35 + v2 @ 0.45 over either alone at + full strength. + • TurboTime changes the graph, not just the strength: it runs at cfg 1.0 + and the reference community workflow swaps `DualModelGuider` for a + single-model `CFGGuider`, dropping the unconditional model. Re-tune any + stacked style LoRA downward. + • A LoRA can be applied to only part of the run. The Gray Screen bypass + LoRA is used at NEGATIVE weight (about -0.25) and, in its shipped + workflows, only on the first step and then disabled; its author warns + that "LoRA being active through the whole generation could lead to + degradation". + • One real step-reduction LoRA exists: **TurboTime** by ostris + (`ostris/ideogram_4_turbotime_lora`, file + `ideogram_4_turbotime_v1.safetensors`, 846,778,576 bytes ≈ 808 MiB; + also mirrored on CivitAI under the same filename). The model card: it + "is capable of generating good quality images in as little as 2 steps + with no CFG and no unconditional model", made with an in-progress + "continuious turbo training method", with updated versions likely. The + card's own sample grid is at 8 steps. The nomadoor community workflow + describes it as a "2 to 8 step LoRA" and runs it at 8 steps, cfg 1.0, + strength 0.6 + • The author never states which weight family this targets. Silent + degradation, not hard errors, in every case observed. (1) Loading a + conditional-trained LoRA onto the unconditional transformer succeeds + silently because the architectures match; it perturbs the negative + branch instead of the positive one. (2) Verified ComfyUI trap: in `Du + +─────────────────────────────────────────────────────────────────────────────── + PROS +─────────────────────────────────────────────────────────────────────────────── + + Explicit trigger word(s) — P4ganiZondaR2009 — so the effect can be + turned on and off from the prompt. [source: model metadata] + + Compact at 163 MB — cheap to stack with other LoRAs. [source: file + size] + +─────────────────────────────────────────────────────────────────────────────── + CONS & CAVEATS +─────────────────────────────────────────────────────────────────────────────── + - Author labels this experimental — results vary shot to shot. [source: + description] + - Author never states which generation mode it was trained for; test + against your own workflow before relying on it. [source: description] + - Modest adoption (144 downloads) — less community feedback to rely on. + [source: source stats] + +─────────────────────────────────────────────────────────────────────────────── + COMPATIBILITY WARNING +─────────────────────────────────────────────────────────────────────────────── + Silent degradation, not hard errors, in every case observed. (1) Loading a + conditional-trained LoRA onto the unconditional transformer succeeds + silently because the architectures match; it perturbs the negative branch + instead of the positive one. (2) Verified ComfyUI trap: in + `DualModelGuider`, passing the SAME model object as both the conditional + model and the LoRA-patched unconditional model is a no-op. Comfy-Org issue + #14521 tabulates it — case 3 (DualModelGuider, conditional model as + positive, conditional model + LoRA as negative) was "exactly identical" to + plain conditional-only CFG, while case 4 with the real unconditional model + differed. kijai confirmed it is expected: "there's no (working) mechanic to + swap LoRA on each step on the same model, it would only work if the model + was a different one." To patch the negative branch you must load a + genuinely separate model file. (3) A diffusers/PEFT-prefixed LoRA in + ComfyUI is expected to produce only a "lora key not loaded" console warning + and no visible effect — this follows from the absent key mapping (see + checkpoint_families) rather than from a published Ideogram-4 bug report. + (4) Cross-quantization loading (fp8-trained LoRA onto nvfp4/nf4/GGUF) + generally works with some precision loss; int8_convrot had a specific, now- + fixed bug (see gotchas). + +─────────────────────────────────────────────────────────────────────────────── + AUTHOR'S DESCRIPTION +─────────────────────────────────────────────────────────────────────────────── + Symphony of Pure Fury: Inside the Unconstrained Madness of the Pagani Zonda + R The Genesis of a Rulebreaker + In the mid-2000s, the hypercar landscape shifted as manufacturers began + building experimental, track-only client vehicles, exemplified by Ferrari's + Enzo-based FXX. Inspired by these extreme machines, a prominent collector + approached Horacio Pagani in early 2006 with a "money-no-object" brief: + build the ultimate, uncompromised track version of the Zonda. + The resulting 2009 Pagani Zonda R was not built to conform to any road- + homologation standards or specific motorsport racing regulations. This + absolute freedom allowed Pagani to design a vehicle from the ground up, + sharing less than 10% of its components with the road-going Zonda F. + The Alchemy of Carbotanium + At the core of the Zonda R’s design is its groundbreaking central monocoque + tub, which pioneered a patented composite material known as Carbotanium. + Created by weaving carbon fiber with high-tensile titanium wire, this + material offers incredible torsional rigidity while preventing the chassis + from shattering under severe kinetic impacts. Surrounding this tub are + brand-new front and rear chrome-molybdenum (CrMo) spaceframes, engineered + to support a completely redesigned suspension geometry. + To minimize weight, Pagani utilized AvionAl (a copper-aluminum duralumin + alloy) for the suspension arms and structural ErgAl for the engine mount + brackets, all held together by custom aerospace-grade Poggipolini titanium + screws. The result was an astonishingly featherlight dry weight of just + 1,070 kg. + Unrestricted Power: The GT 112 V12 + The crown jewel of the Zonda R is its naturally aspirated 6.0-liter + Mercedes-AMG V12 engine, internally designated as the GT 112. Derived from + the power unit of the legendary Mercedes-Benz CLK GTR race car, this V12 + was completely liberated from the air restrictor plates required in GT1 + racing. In its unrestricted form, the engine produces 750 horsepower and + 710 Nm of torque. + Pagani chose a mechanical, cable-operated throttle linkage instead of + modern drive-by-wire systems, ensuring instantaneous throttle response. The + V12 screams through an un-silenced, hydroformed Inconel 625 exhaust system + coated in ceramic. This setup bypasses mufflers to produce one of the most + violent, high-pitched exhaust notes in automotive history, easily exceeding + standard track noise limits. Power is handled by a longitudinally mounted + Xtrac sequential six-speed gearbox that executes lightning-fast shifts in + just 20 milliseconds. + Telemetry, Downforce, and the Green Hell + Aerodynamic efficiency was refined extensively in the wind tunnel to + generate massive downforce. The Zonda R features a closed flat underbody, + an aggressive front splitter with dive canards, a huge rear diffuser, and + an adjustable rear carbon wing, producing up to 1,500 kg of downforce. A + highly advanced telemetry system monitors, logs, and displays real-time + downforce at each wheel, allowing support crews to adjust the aerodynamic + balance on the fly. + With a power-to-weight ratio of 701 hp-per-ton and custom Pirelli P Zero + slick tires, the Zonda R rocketed from 0 to 100 km/h in less than 2.7 + seconds, on its way to a top speed of 375 km/h. Its handling prowess was + cemented on June 30, 2010, when professional driver Marc Basseng lapped the + legendary 20.8-kilometer Nürburgring Nordschleife in a mind-bending 6 + minutes and 47.5 seconds—shattering the existing production-derived record + by more than 11 seconds. + Evolution and Street-Legal Rebirth + The Zonda R's production was limited to 15 customer units plus one + prototype. However, its success birthed the 2013 Zonda Revolución, which + pushed output to 800 hp, added F1-style active Drag Reduction Systems (DRS) + to the rear wing, and introduced advanced CCMR carbon-ceramic brakes. + Though initially designed strictly for track use, British engineering firm + Lanzante Limited made automotive history by successfully converting select + Zonda R and Revolución models for road-legal use. Achieving this required + complex modifications to suspension heights, the integration of nose-lift + systems, catalytic converters, cooling fans to prevent overheating in + traffic, handbrakes, and public-road-compliant lighting and exhaust + silencers. + A Lasting Technological Legacy + Ultimately, Horacio Pagani envisioned the Zonda R as a mobile "laboratory + car". The innovations pioneered on this extreme platform directly shaped + Pagani’s subsequent hypercars. The carbon-titanium tub, the mechanical + stressed-member V12 mount, the sequential gearbox, and the active + aerodynamics were all transferred to the Huayra, the track-only Huayra R, + and eventually the Utopia. The Zonda R remains a legendary monument to what + happens when engineering is entirely unchained from the rules of the road. + +─────────────────────────────────────────────────────────────────────────────── + VERSION NOTES +─────────────────────────────────────────────────────────────────────────────── + Initial release of the LoRA. + +─────────────────────────────────────────────────────────────────────────────── + EXAMPLE MEDIA +─────────────────────────────────────────────────────────────────────────────── + 6 preview(s) in ./example_images/ — community results for + this LoRA. Videos keep a matching _poster.jpg still frame. + + No prompts are recorded below: CivitAI's public API flags these posts as + having generation metadata but does not return it, so the prompts cannot be + scraped. Open the model page to read them. + + 01_image.jpeg (image, 1600x896) + 02_image.jpeg (image, 1600x896) + 03_image.jpeg (image, 1600x896) + 04_image.jpeg (image, 1600x896) + 05_image.jpeg (image, 1600x896) + 06_image.jpeg (image, 1600x896) + +─────────────────────────────────────────────────────────────────────────────── + HOW TO USE (COMFYUI) +─────────────────────────────────────────────────────────────────────────────── + Use ComfyUI's NATIVE Ideogram 4 support (Comfy-Org/ComfyUI PR #14259 "feat: + Add Ideogram4 model support", merged 2026-06-03) — not the official + `ComfyUI-Ideogram4` custom-node wrapper, whose nodes (Magic Prompt / + Pipeline Loader / Generate) wrap Ideogram's reference pipeline and contain + no LoRA input at all (verified: no occurrence of "lora" in its nodes.py). + + 1. Update ComfyUI to a build containing `comfy_extras/nodes_ideogram4.py` + (which supplies `Ideogram4Scheduler`); the same PR added the generic + `DualModelGuider` and `CFGOverride` nodes, which live in + `comfy_extras/nodes_custom_sampler.py`. + 2. From `Comfy-Org/Ideogram-4` download into + `ComfyUI/models/diffusion_models/`: `ideogram4_fp8_scaled.safetensors` + (9.28 GB) AND `ideogram4_unconditional_fp8_scaled.safetensors` (9.28 GB). + Lower-VRAM alternatives in the same folder: `*_nvfp4_mixed.safetensors` + (5.49 GB each) or `*_int8_convrot.safetensors` (9.58 GB each). + 3. Download `qwen3vl_8b_fp8_scaled.safetensors` (10.59 GB; or + `qwen3vl_8b_nvfp4.safetensors`, 6.31 GB) into + `ComfyUI/models/text_encoders/`. The shipped template also lists an + optional `gemma4_e4b_it_fp8_scaled.safetensors` from `Comfy-Org/gemma-4` + for local prompt building. + 4. Download `flux2-vae.safetensors` (0.34 GB, from Comfy-Org/Ideogram-4 or + Comfy-Org/flux2-dev) into `ComfyUI/models/vae/`. + 5. Put LoRA .safetensors in `ComfyUI/models/loras/` (no subfolder + convention enforced). + 6. Load the built-in template `image_ideogram4_t2i` (Workflow → Browse + Templates), or the workflow from + docs.comfy.org/tutorials/image/ideogram/ideogram-v4. + 7. Insert a `LoraLoaderModelOnly` between the main `UNETLoader` and the + `DualModelGuider`'s conditional model input. There is no CLIP-side LoRA + here — the text encoder is a frozen Qwen3-VL — so model-only is the right + loader. Strength ~0.5–0.8. + 8. Do NOT reuse one patched model object for both guider inputs; it + silently does nothing. Patching the negative branch requires a second, + physically separate model file with its own loader. + 9. Write the prompt as a JSON caption into `CLIPTextEncode` (the CLIPLoader + type is `ideogram4`); keep `ConditioningZeroOut` feeding the unconditional + branch. + + Non-ComfyUI paths: diffusers — + `Ideogram4Pipeline.from_pretrained("ideogram-ai/ideogram-4-nf4")` (the + docs' own example; `ideogram-ai/ideogram-4-nf4-diffusers` also exists), + then `pipeline.load_lora_weights(...)`. Training — `ostris/ai-toolkit` + (built-in `extensions_built_in/diffusion_models/ideogram4/` plus an + `Ideogram4Captioner` that writes JSON captions) or diffusers + `examples/dreambooth/train_dreambooth_lora_ideogram4.py`. The official HF + weights are GATED (`gated: auto`): accept the license on the model page and + authenticate (`hf auth login` / `HF_TOKEN`) or downloads fail. + +─────────────────────────────────────────────────────────────────────────────── + Scraped by lorakit on 2026-09-05 01:49 UTC. + Stats and description are the author's; pros/cons are derived from the + evidence tagged beside each line. +─────────────────────────────────────────────────────────────────────────────── diff --git a/pagani-zonda-r-2009/mastermerlin_ideogram4-fp8_Pagani_Zonda_R_2009_final.safetensors b/pagani-zonda-r-2009/mastermerlin_ideogram4-fp8_Pagani_Zonda_R_2009_final.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..8520d415d182ae4da6c14c90e522fd3ff8cbb129 --- /dev/null +++ b/pagani-zonda-r-2009/mastermerlin_ideogram4-fp8_Pagani_Zonda_R_2009_final.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e93e9b2bbf582dbe0e8b744eec81e5fd282f9538eaa2d5f72334cff703a2e6d5 +size 170503408 diff --git a/pagani-zonda-r-2009/metadata.json b/pagani-zonda-r-2009/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..654737ca961e0506bd35ca4919cbd974c4b0e06b --- /dev/null +++ b/pagani-zonda-r-2009/metadata.json @@ -0,0 +1,133 @@ +{ + "name": "Pagani Zonda R - 2009", + "slug": "pagani-zonda-r-2009", + "base_model": "Ideogram 4", + "media_type": "image", + "model_id": 2710899, + "version_id": 3053389, + "version_name": "v1.0 IDEOGRAM4", + "source": "civitai", + "url": "https://civitai.com/models/2710899", + "repo_id": null, + "file_path": null, + "author": "mastermerlin", + "category": { + "key": "accelerator", + "label": "Speed / step-reduction" + }, + "modalities": [], + "primary_modality": "unspecified", + "checkpoint_family": "unknown", + "trigger_words": [ + "P4ganiZondaR2009" + ], + "recommended_settings": {}, + "chaining": { + "author": { + "stacks": "not stated", + "notes": "", + "avoid": [], + "requires": [], + "order": "", + "strength": "", + "stated": false + }, + "general_rules": [ + "Chain `LoraLoaderModelOnly` nodes in series off the CONDITIONAL model loader, ahead of the guider. DeverStyle states its samples were made \"by setting the LORA @ 0.6 strength on the model only, not on the model_negative\".", + "Model-only loaders throughout. The text encoder is a frozen Qwen3-VL-8B with no LoRA-patchable CLIP branch, so the model+clip `LoraLoader` has nothing useful to do.", + "To affect the unconditional/negative branch you must load a second, physically distinct model file and patch that one. Passing the same patched model object into both DualModelGuider inputs is a verified no-op (Comfy-Org issue #14521; kijai confirmed it is expected behaviour).", + "Trigger words go INSIDE the JSON caption, in the semantically matching field, not prepended to a prose string. DeverStyle instructs putting `dvr_arch_tv` in `style_description.art_style` (and in the content field); the Kiki LoRA card describes using `art_style: \"Studio Ghibli style\"` with `medium: \"anime_screencap\"`.", + "Two versions of the same LoRA can be stacked at reduced strengths: zjourney's author recommends v1 @ 0.35 + v2 @ 0.45 over either alone at full strength.", + "TurboTime changes the graph, not just the strength: it runs at cfg 1.0 and the reference community workflow swaps `DualModelGuider` for a single-model `CFGGuider`, dropping the unconditional model. Re-tune any stacked style LoRA downward.", + "A LoRA can be applied to only part of the run. The Gray Screen bypass LoRA is used at NEGATIVE weight (about -0.25) and, in its shipped workflows, only on the first step and then disabled; its author warns that \"LoRA being active through the whole generation could lead to degradation\".", + "One real step-reduction LoRA exists: **TurboTime** by ostris (`ostris/ideogram_4_turbotime_lora`, file `ideogram_4_turbotime_v1.safetensors`, 846,778,576 bytes ≈ 808 MiB; also mirrored on CivitAI under the same filename). The model card: it \"is capable of generating good quality images in as little as 2 steps with no CFG and no unconditional model\", made with an in-progress \"continuious turbo training method\", with updated versions likely. The card's own sample grid is at 8 steps. The nomadoor community workflow describes it as a \"2 to 8 step LoRA\" and runs it at 8 steps, cfg 1.0, strength 0.6", + "The author never states which weight family this targets. Silent degradation, not hard errors, in every case observed. (1) Loading a conditional-trained LoRA onto the unconditional transformer succeeds silently because the architectures match; it perturbs the negative branch instead of the positive one. (2) Verified ComfyUI trap: in `Du" + ] + }, + "pros": [ + "Explicit trigger word(s) — P4ganiZondaR2009 — so the effect can be turned on and off from the prompt. [source: model metadata]", + "Compact at 163 MB — cheap to stack with other LoRAs. [source: file size]" + ], + "cons": [ + "Author labels this experimental — results vary shot to shot. [source: description]", + "Author never states which generation mode it was trained for; test against your own workflow before relying on it. [source: description]", + "Modest adoption (144 downloads) — less community feedback to rely on. [source: source stats]" + ], + "file": { + "name": "mastermerlin_ideogram4-fp8_Pagani_Zonda_R_2009_final.safetensors", + "size_bytes": 170503408, + "size_mb": 162.6, + "sha256": "e93e9b2bbf582dbe0e8b744eec81e5fd282f9538eaa2d5f72334cff703a2e6d5" + }, + "stats": { + "downloads": 144, + "thumbs_up": 7, + "thumbs_down": 0 + }, + "popularity": { + "downloads": 144, + "thumbs_up": 7, + "thumbs_down": 0, + "comments": 0, + "approval": "100% positive (7/7)", + "age_days": 75.7, + "downloads_per_day": 0.41, + "rank_by_downloads": 32, + "rank_by_momentum": 34, + "of_total": 35, + "adoption_tier": "Niche", + "momentum": "Long tail" + }, + "selection_score": 628.0, + "selection_reasons": [ + "144 downloads", + "7 thumbs-up", + "documented triggers + usage notes", + "top-ranked for its category" + ], + "example_media": [ + { + "file": "01_image.jpeg", + "type": "image", + "resolution": "1600x896", + "prompt": "" + }, + { + "file": "02_image.jpeg", + "type": "image", + "resolution": "1600x896", + "prompt": "" + }, + { + "file": "03_image.jpeg", + "type": "image", + "resolution": "1600x896", + "prompt": "" + }, + { + "file": "04_image.jpeg", + "type": "image", + "resolution": "1600x896", + "prompt": "" + }, + { + "file": "05_image.jpeg", + "type": "image", + "resolution": "1600x896", + "prompt": "" + }, + { + "file": "06_image.jpeg", + "type": "image", + "resolution": "1600x896", + "prompt": "" + } + ], + "tags": [ + "pagani", + "vehicle", + "zonda" + ], + "scraped_at": "2026-09-05T01:49:21+00:00", + "path": "ideogram-4/pagani-zonda-r-2009" +} diff --git a/pale-skin-redhead-enhancer/example_images/01_image.jpeg b/pale-skin-redhead-enhancer/example_images/01_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..3ddf8a4cef8f3195d2ef7d1956ee31ccfa90908e --- /dev/null +++ b/pale-skin-redhead-enhancer/example_images/01_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4e34fcc8d51a6216568a7da159baf933fe131ad3c5abf9775aa815b45f3242ae +size 2801646 diff --git a/pale-skin-redhead-enhancer/example_images/02_image.jpeg b/pale-skin-redhead-enhancer/example_images/02_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..4347dde67da03fd3e55cfb51d752139bd894e122 --- /dev/null +++ b/pale-skin-redhead-enhancer/example_images/02_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d23cfab32e9205173bd0ee89be7e7d98ba71033a1fcfb1fb9f45b575ef85ef92 +size 5815194 diff --git a/pale-skin-redhead-enhancer/example_images/03_image.jpeg b/pale-skin-redhead-enhancer/example_images/03_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..2163b8c97edd8cec6bba07881be0b6dd5304b747 --- /dev/null +++ b/pale-skin-redhead-enhancer/example_images/03_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:77506ae7d72c0f6310c7e602e48bd8d683611aaeaebd4983e69a8483af0c6363 +size 3274490 diff --git a/pale-skin-redhead-enhancer/example_images/04_image.jpeg b/pale-skin-redhead-enhancer/example_images/04_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..1fee748ffc748425194aa5661c8714e2289a2bd5 --- /dev/null +++ b/pale-skin-redhead-enhancer/example_images/04_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:beb112767f3e6450ca8ddda92e60e2decd6a8abb2a6a8850735aee94b1334ed0 +size 2789014 diff --git a/pale-skin-redhead-enhancer/example_images/05_image.jpeg b/pale-skin-redhead-enhancer/example_images/05_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..15b835cb2e6f88a00e39396d25938ae3fb705976 --- /dev/null +++ b/pale-skin-redhead-enhancer/example_images/05_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:151eab29deba982fb37c56139540d2ba08cecd783933bcbcc3860bc96922b19f +size 1575424 diff --git a/pale-skin-redhead-enhancer/example_images/06_image.jpeg b/pale-skin-redhead-enhancer/example_images/06_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..ebf2dca8f71fbc04cd79b33831cdb1f0d820da29 --- /dev/null +++ b/pale-skin-redhead-enhancer/example_images/06_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c0fb9455d207482127914ff4cb46193d4930325fdf7cc479c81b69e22ad99cc0 +size 1362980 diff --git a/pale-skin-redhead-enhancer/info.txt b/pale-skin-redhead-enhancer/info.txt new file mode 100644 index 0000000000000000000000000000000000000000..87d4cd8dfa8e520443bea76d41f2af7c22e5ae00 --- /dev/null +++ b/pale-skin-redhead-enhancer/info.txt @@ -0,0 +1,261 @@ +═══════════════════════════════════════════════════════════════════════════════ + PALE SKIN / REDHEAD ENHANCER +═══════════════════════════════════════════════════════════════════════════════ + +─────────────────────────────────────────────────────────────────────────────── + IDENTITY +─────────────────────────────────────────────────────────────────────────────── + Base model Ideogram 4 + Generation mode not declared by the author + Checkpoint not determined — verify before use + Category Detail / anatomy fixer + Author CallMeMaybe + Source https://civitai.com/models/2738391 + Provenance CivitAI · model 2738391 · version 3079368 + +─────────────────────────────────────────────────────────────────────────────── + WEIGHTS +─────────────────────────────────────────────────────────────────────────────── + File pale-id4-v2_000000500.safetensors + Version v1.0 + Size 101.0 MB + Format SafeTensor / unknown precision + SHA256 138d813cb072bc3a5317072507cb112cbf326caf55c846ccabb99dd3e8d22643 + Published 2026-06-28 + +─────────────────────────────────────────────────────────────────────────────── + TRIGGER WORDS +─────────────────────────────────────────────────────────────────────────────── + › "pale" + +─────────────────────────────────────────────────────────────────────────────── + RECOMMENDED SETTINGS +─────────────────────────────────────────────────────────────────────────────── + The author did not publish specific settings. + No author settings were published. Community baseline: steps 48 is the + reference default: docs/inference.md says "`V4_QUALITY_48` is the default" + and the pipeline's `num_steps` default is 48; diffusers documents the same + 48-step default. 20 is the practical ComfyUI default — the `V4_DEFAULT_20` + preset and the value the shipped template `image_ideogram4_t2i.json` + selects (its preset combo defaults to "Default"). 12 = `V4_TURBO_12`. With + the TurboTime LoRA: 8 in the nomadoor community workflow; the model card + claims usable output down to 2., cfg A guidance SCHEDULE, not one number, + and the negative branch comes from a second model rather than a negative + prompt. Verified from src/ideogram4/sampler_configs.py: V4_QUALITY_48 = 45 + steps @ gw 7 then 3 polish steps @ gw 3; V4_DEFAULT_20 = 18 @ 7 + 2 @ 3; + V4_TURBO_12 = 11 @ 7 + 1 @ 3. Pipeline default `guidance_scale` = 7.0 when + no schedule is given. In ComfyUI the shipped template sets + `DualModelGuider` cfg 7 plus a `CFGOverride` node with widgets [3, 0.7, 1] + (cfg 3 from 0.7→1.0 of the run); kijai recommends starting the override at + 0.7, not 0.9. TurboTime: cfg 1.0 / no CFG., sampler Euler flow-matching. In + ComfyUI: `KSamplerSelect` → euler, driven by `SamplerCustomAdvanced`, with + the model-specific `Ideogram4Scheduler` (logit-normal sigma schedule). Its + params are `mu` and `std`, not the usual `shift`. Reference presets: mu 0.0 + / std 1.5 (48-step), mu 0.0 / std 1.75 (20-step), mu 0.5 / std 1.75 + (12-step). The Ideogram4Scheduler node's own widget defaults are steps 20, + mu 0.5, std 1.75, but the shipped template overrides mu/std/steps from the + preset JSON, so out of the box it runs 20 / mu 0.0 / std 1.75. mu is auto- + shifted by resolution in both implementations: mean = mu + + 0.5*log((w*h)/(512*512))., resolution 1024×1024 default (pipeline + `height`/`width` default 1024). Any multiple of 16 per side from 256 to + 2048, aspect ratios up to 6:1 or 1:6 — one model covers all of it. + 2048×2048 with V4_QUALITY_48 for maximum quality. Buckets documented in + docs/inference.md: 1536×1024 (3:2), 1024×1536 (2:3), 1920×1088 (~16:9), + 2048×768 (~21:9), 1024×1792 (~9:16), 1600×400 (4:1)., lora_strength No + single official number; these are author-stated values. DeverStyle Archer: + 0.6, and explicitly "on the model only, not on the model_negative". + zjourney v2: 0.55 standalone (range 0.35–0.6), or v1 @ 0.35 + v2 @ 0.45 + stacked. Furry Enhancer: the model page says 0.6–0.8, but its own version + notes ask for 0.7–0.9 (v1.94) and 0.8–0.95 (v2.8). TurboTime: 0.6 in the + nomadoor workflow (the author gives no figure). So ~0.5–0.9 is the + realistic band, read the specific card.. + +─────────────────────────────────────────────────────────────────────────────── + ADOPTION & POPULARITY +─────────────────────────────────────────────────────────────────────────────── + Downloads 194 + Rating 100% positive (11/11) + Comments 0 + Published 68 days ago + Download rate 2.8/day since release + + Adoption rank #31 of 35 in this collection (Niche) + Momentum rank #12 of 35 by download rate (Steady) + + Adoption tier is the percentile of total downloads within this collection; + momentum is the percentile of downloads-per-day since release. Momentum is + ranked rather than measured against a fixed rate, because this base model + is itself new and every LoRA looks fast on an absolute scale. Both numbers + are CivitAI's, read on the scrape date at the foot of this file. + +─────────────────────────────────────────────────────────────────────────────── + CHAINING / STACKING +─────────────────────────────────────────────────────────────────────────────── + Stacks with others not stated + + The author gave no chaining guidance. The rules below are the + general Ideogram 4 ones, not anything specific to this LoRA. + + General Ideogram 4 rules that apply here (see REFERENCE.txt): + + • Chain `LoraLoaderModelOnly` nodes in series off the CONDITIONAL model + loader, ahead of the guider. DeverStyle states its samples were made + "by setting the LORA @ 0.6 strength on the model only, not on the + model_negative". + • Model-only loaders throughout. The text encoder is a frozen Qwen3-VL-8B + with no LoRA-patchable CLIP branch, so the model+clip `LoraLoader` has + nothing useful to do. + • To affect the unconditional/negative branch you must load a second, + physically distinct model file and patch that one. Passing the same + patched model object into both DualModelGuider inputs is a verified no- + op (Comfy-Org issue #14521; kijai confirmed it is expected behaviour). + • Trigger words go INSIDE the JSON caption, in the semantically matching + field, not prepended to a prose string. DeverStyle instructs putting + `dvr_arch_tv` in `style_description.art_style` (and in the content + field); the Kiki LoRA card describes using `art_style: "Studio Ghibli + style"` with `medium: "anime_screencap"`. + • Two versions of the same LoRA can be stacked at reduced strengths: + zjourney's author recommends v1 @ 0.35 + v2 @ 0.45 over either alone at + full strength. + • TurboTime changes the graph, not just the strength: it runs at cfg 1.0 + and the reference community workflow swaps `DualModelGuider` for a + single-model `CFGGuider`, dropping the unconditional model. Re-tune any + stacked style LoRA downward. + • A LoRA can be applied to only part of the run. The Gray Screen bypass + LoRA is used at NEGATIVE weight (about -0.25) and, in its shipped + workflows, only on the first step and then disabled; its author warns + that "LoRA being active through the whole generation could lead to + degradation". + • The author never states which weight family this targets. Silent + degradation, not hard errors, in every case observed. (1) Loading a + conditional-trained LoRA onto the unconditional transformer succeeds + silently because the architectures match; it perturbs the negative + branch instead of the positive one. (2) Verified ComfyUI trap: in `Du + +─────────────────────────────────────────────────────────────────────────────── + PROS +─────────────────────────────────────────────────────────────────────────────── + + Explicit trigger word(s) — pale — so the effect can be turned on and + off from the prompt. [source: model metadata] + + Compact at 101 MB — cheap to stack with other LoRAs. [source: file + size] + +─────────────────────────────────────────────────────────────────────────────── + CONS & CAVEATS +─────────────────────────────────────────────────────────────────────────────── + - Author says the LoRA may be overtrained — start at low strength. + [source: description] + - Author never states which generation mode it was trained for; test + against your own workflow before relying on it. [source: description] + - Modest adoption (194 downloads) — less community feedback to rely on. + [source: source stats] + +─────────────────────────────────────────────────────────────────────────────── + COMPATIBILITY WARNING +─────────────────────────────────────────────────────────────────────────────── + Silent degradation, not hard errors, in every case observed. (1) Loading a + conditional-trained LoRA onto the unconditional transformer succeeds + silently because the architectures match; it perturbs the negative branch + instead of the positive one. (2) Verified ComfyUI trap: in + `DualModelGuider`, passing the SAME model object as both the conditional + model and the LoRA-patched unconditional model is a no-op. Comfy-Org issue + #14521 tabulates it — case 3 (DualModelGuider, conditional model as + positive, conditional model + LoRA as negative) was "exactly identical" to + plain conditional-only CFG, while case 4 with the real unconditional model + differed. kijai confirmed it is expected: "there's no (working) mechanic to + swap LoRA on each step on the same model, it would only work if the model + was a different one." To patch the negative branch you must load a + genuinely separate model file. (3) A diffusers/PEFT-prefixed LoRA in + ComfyUI is expected to produce only a "lora key not loaded" console warning + and no visible effect — this follows from the absent key mapping (see + checkpoint_families) rather than from a published Ideogram-4 bug report. + (4) Cross-quantization loading (fp8-trained LoRA onto nvfp4/nf4/GGUF) + generally works with some precision loss; int8_convrot had a specific, now- + fixed bug (see gotchas). + +─────────────────────────────────────────────────────────────────────────────── + AUTHOR'S DESCRIPTION +─────────────────────────────────────────────────────────────────────────────── + Trying out making a lora for Ideogram 4 - it is very difficult and I have + not figured out the right way to train this model, but this is usable. + Somehow this trained in only 500 steps and was way overtrained at anything + above. + +─────────────────────────────────────────────────────────────────────────────── + VERSION NOTES +─────────────────────────────────────────────────────────────────────────────── + v1.0 + +─────────────────────────────────────────────────────────────────────────────── + EXAMPLE MEDIA +─────────────────────────────────────────────────────────────────────────────── + 6 preview(s) in ./example_images/ — community results for + this LoRA. Videos keep a matching _poster.jpg still frame. + + No prompts are recorded below: CivitAI's public API flags these posts as + having generation metadata but does not return it, so the prompts cannot be + scraped. Open the model page to read them. + + 01_image.jpeg (image, 1264x1584) + 02_image.jpeg (image, 2528x1584) + 03_image.jpeg (image, 1264x1584) + 04_image.jpeg (image, 1264x1584) + 05_image.jpeg (image, 752x1328) + 06_image.jpeg (image, 752x1328) + +─────────────────────────────────────────────────────────────────────────────── + HOW TO USE (COMFYUI) +─────────────────────────────────────────────────────────────────────────────── + Use ComfyUI's NATIVE Ideogram 4 support (Comfy-Org/ComfyUI PR #14259 "feat: + Add Ideogram4 model support", merged 2026-06-03) — not the official + `ComfyUI-Ideogram4` custom-node wrapper, whose nodes (Magic Prompt / + Pipeline Loader / Generate) wrap Ideogram's reference pipeline and contain + no LoRA input at all (verified: no occurrence of "lora" in its nodes.py). + + 1. Update ComfyUI to a build containing `comfy_extras/nodes_ideogram4.py` + (which supplies `Ideogram4Scheduler`); the same PR added the generic + `DualModelGuider` and `CFGOverride` nodes, which live in + `comfy_extras/nodes_custom_sampler.py`. + 2. From `Comfy-Org/Ideogram-4` download into + `ComfyUI/models/diffusion_models/`: `ideogram4_fp8_scaled.safetensors` + (9.28 GB) AND `ideogram4_unconditional_fp8_scaled.safetensors` (9.28 GB). + Lower-VRAM alternatives in the same folder: `*_nvfp4_mixed.safetensors` + (5.49 GB each) or `*_int8_convrot.safetensors` (9.58 GB each). + 3. Download `qwen3vl_8b_fp8_scaled.safetensors` (10.59 GB; or + `qwen3vl_8b_nvfp4.safetensors`, 6.31 GB) into + `ComfyUI/models/text_encoders/`. The shipped template also lists an + optional `gemma4_e4b_it_fp8_scaled.safetensors` from `Comfy-Org/gemma-4` + for local prompt building. + 4. Download `flux2-vae.safetensors` (0.34 GB, from Comfy-Org/Ideogram-4 or + Comfy-Org/flux2-dev) into `ComfyUI/models/vae/`. + 5. Put LoRA .safetensors in `ComfyUI/models/loras/` (no subfolder + convention enforced). + 6. Load the built-in template `image_ideogram4_t2i` (Workflow → Browse + Templates), or the workflow from + docs.comfy.org/tutorials/image/ideogram/ideogram-v4. + 7. Insert a `LoraLoaderModelOnly` between the main `UNETLoader` and the + `DualModelGuider`'s conditional model input. There is no CLIP-side LoRA + here — the text encoder is a frozen Qwen3-VL — so model-only is the right + loader. Strength ~0.5–0.8. + 8. Do NOT reuse one patched model object for both guider inputs; it + silently does nothing. Patching the negative branch requires a second, + physically separate model file with its own loader. + 9. Write the prompt as a JSON caption into `CLIPTextEncode` (the CLIPLoader + type is `ideogram4`); keep `ConditioningZeroOut` feeding the unconditional + branch. + + Non-ComfyUI paths: diffusers — + `Ideogram4Pipeline.from_pretrained("ideogram-ai/ideogram-4-nf4")` (the + docs' own example; `ideogram-ai/ideogram-4-nf4-diffusers` also exists), + then `pipeline.load_lora_weights(...)`. Training — `ostris/ai-toolkit` + (built-in `extensions_built_in/diffusion_models/ideogram4/` plus an + `Ideogram4Captioner` that writes JSON captions) or diffusers + `examples/dreambooth/train_dreambooth_lora_ideogram4.py`. The official HF + weights are GATED (`gated: auto`): accept the license on the model page and + authenticate (`hf auth login` / `HF_TOKEN`) or downloads fail. + +─────────────────────────────────────────────────────────────────────────────── + Scraped by lorakit on 2026-09-05 01:49 UTC. + Stats and description are the author's; pros/cons are derived from the + evidence tagged beside each line. +─────────────────────────────────────────────────────────────────────────────── diff --git a/pale-skin-redhead-enhancer/metadata.json b/pale-skin-redhead-enhancer/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..ed06580402bc42e365fca9590fda95bde616a20e --- /dev/null +++ b/pale-skin-redhead-enhancer/metadata.json @@ -0,0 +1,133 @@ +{ + "name": "Pale Skin / Redhead Enhancer", + "slug": "pale-skin-redhead-enhancer", + "base_model": "Ideogram 4", + "media_type": "image", + "model_id": 2738391, + "version_id": 3079368, + "version_name": "v1.0", + "source": "civitai", + "url": "https://civitai.com/models/2738391", + "repo_id": null, + "file_path": null, + "author": "CallMeMaybe", + "category": { + "key": "detail", + "label": "Detail / anatomy fixer" + }, + "modalities": [], + "primary_modality": "unspecified", + "checkpoint_family": "unknown", + "trigger_words": [ + "pale" + ], + "recommended_settings": {}, + "chaining": { + "author": { + "stacks": "not stated", + "notes": "", + "avoid": [], + "requires": [], + "order": "", + "strength": "", + "stated": false + }, + "general_rules": [ + "Chain `LoraLoaderModelOnly` nodes in series off the CONDITIONAL model loader, ahead of the guider. DeverStyle states its samples were made \"by setting the LORA @ 0.6 strength on the model only, not on the model_negative\".", + "Model-only loaders throughout. The text encoder is a frozen Qwen3-VL-8B with no LoRA-patchable CLIP branch, so the model+clip `LoraLoader` has nothing useful to do.", + "To affect the unconditional/negative branch you must load a second, physically distinct model file and patch that one. Passing the same patched model object into both DualModelGuider inputs is a verified no-op (Comfy-Org issue #14521; kijai confirmed it is expected behaviour).", + "Trigger words go INSIDE the JSON caption, in the semantically matching field, not prepended to a prose string. DeverStyle instructs putting `dvr_arch_tv` in `style_description.art_style` (and in the content field); the Kiki LoRA card describes using `art_style: \"Studio Ghibli style\"` with `medium: \"anime_screencap\"`.", + "Two versions of the same LoRA can be stacked at reduced strengths: zjourney's author recommends v1 @ 0.35 + v2 @ 0.45 over either alone at full strength.", + "TurboTime changes the graph, not just the strength: it runs at cfg 1.0 and the reference community workflow swaps `DualModelGuider` for a single-model `CFGGuider`, dropping the unconditional model. Re-tune any stacked style LoRA downward.", + "A LoRA can be applied to only part of the run. The Gray Screen bypass LoRA is used at NEGATIVE weight (about -0.25) and, in its shipped workflows, only on the first step and then disabled; its author warns that \"LoRA being active through the whole generation could lead to degradation\".", + "The author never states which weight family this targets. Silent degradation, not hard errors, in every case observed. (1) Loading a conditional-trained LoRA onto the unconditional transformer succeeds silently because the architectures match; it perturbs the negative branch instead of the positive one. (2) Verified ComfyUI trap: in `Du" + ] + }, + "pros": [ + "Explicit trigger word(s) — pale — so the effect can be turned on and off from the prompt. [source: model metadata]", + "Compact at 101 MB — cheap to stack with other LoRAs. [source: file size]" + ], + "cons": [ + "Author says the LoRA may be overtrained — start at low strength. [source: description]", + "Author never states which generation mode it was trained for; test against your own workflow before relying on it. [source: description]", + "Modest adoption (194 downloads) — less community feedback to rely on. [source: source stats]" + ], + "file": { + "name": "pale-id4-v2_000000500.safetensors", + "size_bytes": 105892920, + "size_mb": 100.99, + "sha256": "138d813cb072bc3a5317072507cb112cbf326caf55c846ccabb99dd3e8d22643" + }, + "stats": { + "downloads": 194, + "thumbs_up": 11, + "thumbs_down": 0 + }, + "popularity": { + "downloads": 194, + "thumbs_up": 11, + "thumbs_down": 0, + "comments": 0, + "approval": "100% positive (11/11)", + "age_days": 68.2, + "downloads_per_day": 2.84, + "rank_by_downloads": 31, + "rank_by_momentum": 12, + "of_total": 35, + "adoption_tier": "Niche", + "momentum": "Steady" + }, + "selection_score": 326.0, + "selection_reasons": [ + "194 downloads", + "11 thumbs-up", + "top-ranked for its category" + ], + "example_media": [ + { + "file": "01_image.jpeg", + "type": "image", + "resolution": "1264x1584", + "prompt": "" + }, + { + "file": "02_image.jpeg", + "type": "image", + "resolution": "2528x1584", + "prompt": "" + }, + { + "file": "03_image.jpeg", + "type": "image", + "resolution": "1264x1584", + "prompt": "" + }, + { + "file": "04_image.jpeg", + "type": "image", + "resolution": "1264x1584", + "prompt": "" + }, + { + "file": "05_image.jpeg", + "type": "image", + "resolution": "752x1328", + "prompt": "" + }, + { + "file": "06_image.jpeg", + "type": "image", + "resolution": "752x1328", + "prompt": "" + } + ], + "tags": [ + "pale skin", + "realistic", + "skin", + "pale", + "concept" + ], + "scraped_at": "2026-09-05T01:49:53+00:00", + "path": "ideogram-4/pale-skin-redhead-enhancer" +} diff --git a/pale-skin-redhead-enhancer/pale-id4-v2_000000500.safetensors b/pale-skin-redhead-enhancer/pale-id4-v2_000000500.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..689e40d76465c35618558081b703eedcdd126a58 --- /dev/null +++ b/pale-skin-redhead-enhancer/pale-id4-v2_000000500.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:138d813cb072bc3a5317072507cb112cbf326caf55c846ccabb99dd3e8d22643 +size 105892920 diff --git a/salomon-xt-6-sneakers/example_images/01_image.jpeg b/salomon-xt-6-sneakers/example_images/01_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..78aa2e8591be2d355652796e304e4736bd352d45 --- /dev/null +++ b/salomon-xt-6-sneakers/example_images/01_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a78fea1b731fde19cd16606c5b575a4b94e256065a3c7a6d7181e905e4539a21 +size 2649011 diff --git a/salomon-xt-6-sneakers/example_images/02_image.jpeg b/salomon-xt-6-sneakers/example_images/02_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..20c5ee437ce06bb0c2779d3794233ff4ea2b56f0 --- /dev/null +++ b/salomon-xt-6-sneakers/example_images/02_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7bee6b8f0c9e6db00d95fe61c28fafe9caa3df941a2ac2b1404d1c870501bc52 +size 2046771 diff --git a/salomon-xt-6-sneakers/example_images/03_image.jpeg b/salomon-xt-6-sneakers/example_images/03_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..e745e0e2086ee520eab9d9fae0d76ff685ab9177 --- /dev/null +++ b/salomon-xt-6-sneakers/example_images/03_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:62d84a4f0c6fe6de40901905aa2271f1fa8b43845c699fe58fba3e2659af3741 +size 3114320 diff --git a/salomon-xt-6-sneakers/info.txt b/salomon-xt-6-sneakers/info.txt new file mode 100644 index 0000000000000000000000000000000000000000..97eaca0185b7df2348ba95c0212ffcad723995dc --- /dev/null +++ b/salomon-xt-6-sneakers/info.txt @@ -0,0 +1,251 @@ +═══════════════════════════════════════════════════════════════════════════════ + SALOMON XT-6 SNEAKERS +═══════════════════════════════════════════════════════════════════════════════ + +─────────────────────────────────────────────────────────────────────────────── + IDENTITY +─────────────────────────────────────────────────────────────────────────────── + Base model Ideogram 4 + Generation mode not declared by the author + Checkpoint not determined — verify before use + Category Clothing / wardrobe + Author feloxxx75 + Source https://civitai.com/models/1827315 + Provenance CivitAI · model 1827315 · version 3153436 + +─────────────────────────────────────────────────────────────────────────────── + WEIGHTS +─────────────────────────────────────────────────────────────────────────────── + File salomon_id4.safetensors + Version Ideogram V2.0 + Size 201.9 MB + Format SafeTensor / unknown precision + SHA256 bcee988a63010113f9470d6d38f5213c829db5152ff01a79caffe576bd4b86b4 + Published 2026-07-22 + +─────────────────────────────────────────────────────────────────────────────── + TRIGGER WORDS +─────────────────────────────────────────────────────────────────────────────── + No trigger word. The LoRA applies as soon as it is loaded — + strength is the only way to dial the effect up or down. + +─────────────────────────────────────────────────────────────────────────────── + RECOMMENDED SETTINGS +─────────────────────────────────────────────────────────────────────────────── + The author did not publish specific settings. + No author settings were published. Community baseline: steps 48 is the + reference default: docs/inference.md says "`V4_QUALITY_48` is the default" + and the pipeline's `num_steps` default is 48; diffusers documents the same + 48-step default. 20 is the practical ComfyUI default — the `V4_DEFAULT_20` + preset and the value the shipped template `image_ideogram4_t2i.json` + selects (its preset combo defaults to "Default"). 12 = `V4_TURBO_12`. With + the TurboTime LoRA: 8 in the nomadoor community workflow; the model card + claims usable output down to 2., cfg A guidance SCHEDULE, not one number, + and the negative branch comes from a second model rather than a negative + prompt. Verified from src/ideogram4/sampler_configs.py: V4_QUALITY_48 = 45 + steps @ gw 7 then 3 polish steps @ gw 3; V4_DEFAULT_20 = 18 @ 7 + 2 @ 3; + V4_TURBO_12 = 11 @ 7 + 1 @ 3. Pipeline default `guidance_scale` = 7.0 when + no schedule is given. In ComfyUI the shipped template sets + `DualModelGuider` cfg 7 plus a `CFGOverride` node with widgets [3, 0.7, 1] + (cfg 3 from 0.7→1.0 of the run); kijai recommends starting the override at + 0.7, not 0.9. TurboTime: cfg 1.0 / no CFG., sampler Euler flow-matching. In + ComfyUI: `KSamplerSelect` → euler, driven by `SamplerCustomAdvanced`, with + the model-specific `Ideogram4Scheduler` (logit-normal sigma schedule). Its + params are `mu` and `std`, not the usual `shift`. Reference presets: mu 0.0 + / std 1.5 (48-step), mu 0.0 / std 1.75 (20-step), mu 0.5 / std 1.75 + (12-step). The Ideogram4Scheduler node's own widget defaults are steps 20, + mu 0.5, std 1.75, but the shipped template overrides mu/std/steps from the + preset JSON, so out of the box it runs 20 / mu 0.0 / std 1.75. mu is auto- + shifted by resolution in both implementations: mean = mu + + 0.5*log((w*h)/(512*512))., resolution 1024×1024 default (pipeline + `height`/`width` default 1024). Any multiple of 16 per side from 256 to + 2048, aspect ratios up to 6:1 or 1:6 — one model covers all of it. + 2048×2048 with V4_QUALITY_48 for maximum quality. Buckets documented in + docs/inference.md: 1536×1024 (3:2), 1024×1536 (2:3), 1920×1088 (~16:9), + 2048×768 (~21:9), 1024×1792 (~9:16), 1600×400 (4:1)., lora_strength No + single official number; these are author-stated values. DeverStyle Archer: + 0.6, and explicitly "on the model only, not on the model_negative". + zjourney v2: 0.55 standalone (range 0.35–0.6), or v1 @ 0.35 + v2 @ 0.45 + stacked. Furry Enhancer: the model page says 0.6–0.8, but its own version + notes ask for 0.7–0.9 (v1.94) and 0.8–0.95 (v2.8). TurboTime: 0.6 in the + nomadoor workflow (the author gives no figure). So ~0.5–0.9 is the + realistic band, read the specific card.. + +─────────────────────────────────────────────────────────────────────────────── + ADOPTION & POPULARITY +─────────────────────────────────────────────────────────────────────────────── + Downloads 117 + Rating 100% positive (6/6) + Comments 0 + Published 45 days ago + Download rate 0.9/day since release + + Adoption rank #34 of 35 in this collection (Niche) + Momentum rank #24 of 35 by download rate (Long tail · fresh release) + + Adoption tier is the percentile of total downloads within this collection; + momentum is the percentile of downloads-per-day since release. Momentum is + ranked rather than measured against a fixed rate, because this base model + is itself new and every LoRA looks fast on an absolute scale. Both numbers + are CivitAI's, read on the scrape date at the foot of this file. + +─────────────────────────────────────────────────────────────────────────────── + CHAINING / STACKING +─────────────────────────────────────────────────────────────────────────────── + Stacks with others not stated + + The author gave no chaining guidance. The rules below are the + general Ideogram 4 ones, not anything specific to this LoRA. + + General Ideogram 4 rules that apply here (see REFERENCE.txt): + + • Chain `LoraLoaderModelOnly` nodes in series off the CONDITIONAL model + loader, ahead of the guider. DeverStyle states its samples were made + "by setting the LORA @ 0.6 strength on the model only, not on the + model_negative". + • Model-only loaders throughout. The text encoder is a frozen Qwen3-VL-8B + with no LoRA-patchable CLIP branch, so the model+clip `LoraLoader` has + nothing useful to do. + • To affect the unconditional/negative branch you must load a second, + physically distinct model file and patch that one. Passing the same + patched model object into both DualModelGuider inputs is a verified no- + op (Comfy-Org issue #14521; kijai confirmed it is expected behaviour). + • Trigger words go INSIDE the JSON caption, in the semantically matching + field, not prepended to a prose string. DeverStyle instructs putting + `dvr_arch_tv` in `style_description.art_style` (and in the content + field); the Kiki LoRA card describes using `art_style: "Studio Ghibli + style"` with `medium: "anime_screencap"`. + • Two versions of the same LoRA can be stacked at reduced strengths: + zjourney's author recommends v1 @ 0.35 + v2 @ 0.45 over either alone at + full strength. + • TurboTime changes the graph, not just the strength: it runs at cfg 1.0 + and the reference community workflow swaps `DualModelGuider` for a + single-model `CFGGuider`, dropping the unconditional model. Re-tune any + stacked style LoRA downward. + • A LoRA can be applied to only part of the run. The Gray Screen bypass + LoRA is used at NEGATIVE weight (about -0.25) and, in its shipped + workflows, only on the first step and then disabled; its author warns + that "LoRA being active through the whole generation could lead to + degradation". + • The author never states which weight family this targets. Silent + degradation, not hard errors, in every case observed. (1) Loading a + conditional-trained LoRA onto the unconditional transformer succeeds + silently because the architectures match; it perturbs the negative + branch instead of the positive one. (2) Verified ComfyUI trap: in `Du + +─────────────────────────────────────────────────────────────────────────────── + PROS +─────────────────────────────────────────────────────────────────────────────── + + Compact at 202 MB — cheap to stack with other LoRAs. [source: file + size] + +─────────────────────────────────────────────────────────────────────────────── + CONS & CAVEATS +─────────────────────────────────────────────────────────────────────────────── + - No trigger word — the LoRA is always on once loaded, so strength is the + only control. [source: model metadata] + - Author never states which generation mode it was trained for; test + against your own workflow before relying on it. [source: description] + - Thin documentation — expect trial and error. [source: description + length] + - Modest adoption (117 downloads) — less community feedback to rely on. + [source: source stats] + +─────────────────────────────────────────────────────────────────────────────── + COMPATIBILITY WARNING +─────────────────────────────────────────────────────────────────────────────── + Silent degradation, not hard errors, in every case observed. (1) Loading a + conditional-trained LoRA onto the unconditional transformer succeeds + silently because the architectures match; it perturbs the negative branch + instead of the positive one. (2) Verified ComfyUI trap: in + `DualModelGuider`, passing the SAME model object as both the conditional + model and the LoRA-patched unconditional model is a no-op. Comfy-Org issue + #14521 tabulates it — case 3 (DualModelGuider, conditional model as + positive, conditional model + LoRA as negative) was "exactly identical" to + plain conditional-only CFG, while case 4 with the real unconditional model + differed. kijai confirmed it is expected: "there's no (working) mechanic to + swap LoRA on each step on the same model, it would only work if the model + was a different one." To patch the negative branch you must load a + genuinely separate model file. (3) A diffusers/PEFT-prefixed LoRA in + ComfyUI is expected to produce only a "lora key not loaded" console warning + and no visible effect — this follows from the absent key mapping (see + checkpoint_families) rather than from a published Ideogram-4 bug report. + (4) Cross-quantization loading (fp8-trained LoRA onto nvfp4/nf4/GGUF) + generally works with some precision loss; int8_convrot had a specific, now- + fixed bug (see gotchas). + +─────────────────────────────────────────────────────────────────────────────── + AUTHOR'S DESCRIPTION +─────────────────────────────────────────────────────────────────────────────── + Training of XT6 Sneakers + +─────────────────────────────────────────────────────────────────────────────── + EXAMPLE MEDIA +─────────────────────────────────────────────────────────────────────────────── + 3 preview(s) in ./example_images/ — community results for + this LoRA. Videos keep a matching _poster.jpg still frame. + + No prompts are recorded below: CivitAI's public API flags these posts as + having generation metadata but does not return it, so the prompts cannot be + scraped. Open the model page to read them. + + 01_image.jpeg (image, 1088x1936) + 02_image.jpeg (image, 1088x1936) + 03_image.jpeg (image, 1088x1936) + +─────────────────────────────────────────────────────────────────────────────── + HOW TO USE (COMFYUI) +─────────────────────────────────────────────────────────────────────────────── + Use ComfyUI's NATIVE Ideogram 4 support (Comfy-Org/ComfyUI PR #14259 "feat: + Add Ideogram4 model support", merged 2026-06-03) — not the official + `ComfyUI-Ideogram4` custom-node wrapper, whose nodes (Magic Prompt / + Pipeline Loader / Generate) wrap Ideogram's reference pipeline and contain + no LoRA input at all (verified: no occurrence of "lora" in its nodes.py). + + 1. Update ComfyUI to a build containing `comfy_extras/nodes_ideogram4.py` + (which supplies `Ideogram4Scheduler`); the same PR added the generic + `DualModelGuider` and `CFGOverride` nodes, which live in + `comfy_extras/nodes_custom_sampler.py`. + 2. From `Comfy-Org/Ideogram-4` download into + `ComfyUI/models/diffusion_models/`: `ideogram4_fp8_scaled.safetensors` + (9.28 GB) AND `ideogram4_unconditional_fp8_scaled.safetensors` (9.28 GB). + Lower-VRAM alternatives in the same folder: `*_nvfp4_mixed.safetensors` + (5.49 GB each) or `*_int8_convrot.safetensors` (9.58 GB each). + 3. Download `qwen3vl_8b_fp8_scaled.safetensors` (10.59 GB; or + `qwen3vl_8b_nvfp4.safetensors`, 6.31 GB) into + `ComfyUI/models/text_encoders/`. The shipped template also lists an + optional `gemma4_e4b_it_fp8_scaled.safetensors` from `Comfy-Org/gemma-4` + for local prompt building. + 4. Download `flux2-vae.safetensors` (0.34 GB, from Comfy-Org/Ideogram-4 or + Comfy-Org/flux2-dev) into `ComfyUI/models/vae/`. + 5. Put LoRA .safetensors in `ComfyUI/models/loras/` (no subfolder + convention enforced). + 6. Load the built-in template `image_ideogram4_t2i` (Workflow → Browse + Templates), or the workflow from + docs.comfy.org/tutorials/image/ideogram/ideogram-v4. + 7. Insert a `LoraLoaderModelOnly` between the main `UNETLoader` and the + `DualModelGuider`'s conditional model input. There is no CLIP-side LoRA + here — the text encoder is a frozen Qwen3-VL — so model-only is the right + loader. Strength ~0.5–0.8. + 8. Do NOT reuse one patched model object for both guider inputs; it + silently does nothing. Patching the negative branch requires a second, + physically separate model file with its own loader. + 9. Write the prompt as a JSON caption into `CLIPTextEncode` (the CLIPLoader + type is `ideogram4`); keep `ConditioningZeroOut` feeding the unconditional + branch. + + Non-ComfyUI paths: diffusers — + `Ideogram4Pipeline.from_pretrained("ideogram-ai/ideogram-4-nf4")` (the + docs' own example; `ideogram-ai/ideogram-4-nf4-diffusers` also exists), + then `pipeline.load_lora_weights(...)`. Training — `ostris/ai-toolkit` + (built-in `extensions_built_in/diffusion_models/ideogram4/` plus an + `Ideogram4Captioner` that writes JSON captions) or diffusers + `examples/dreambooth/train_dreambooth_lora_ideogram4.py`. The official HF + weights are GATED (`gated: auto`): accept the license on the model page and + authenticate (`hf auth login` / `HF_TOKEN`) or downloads fail. + +─────────────────────────────────────────────────────────────────────────────── + Scraped by lorakit on 2026-09-05 01:50 UTC. + Stats and description are the author's; pros/cons are derived from the + evidence tagged beside each line. +─────────────────────────────────────────────────────────────────────────────── diff --git a/salomon-xt-6-sneakers/metadata.json b/salomon-xt-6-sneakers/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..f2b4a4b86d176d098e97e1d6a1202d5ab03a7c09 --- /dev/null +++ b/salomon-xt-6-sneakers/metadata.json @@ -0,0 +1,112 @@ +{ + "name": "Salomon XT-6 Sneakers", + "slug": "salomon-xt-6-sneakers", + "base_model": "Ideogram 4", + "media_type": "image", + "model_id": 1827315, + "version_id": 3153436, + "version_name": "Ideogram V2.0", + "source": "civitai", + "url": "https://civitai.com/models/1827315", + "repo_id": null, + "file_path": null, + "author": "feloxxx75", + "category": { + "key": "clothing", + "label": "Clothing / wardrobe" + }, + "modalities": [], + "primary_modality": "unspecified", + "checkpoint_family": "unknown", + "trigger_words": [], + "recommended_settings": {}, + "chaining": { + "author": { + "stacks": "not stated", + "notes": "", + "avoid": [], + "requires": [], + "order": "", + "strength": "", + "stated": false + }, + "general_rules": [ + "Chain `LoraLoaderModelOnly` nodes in series off the CONDITIONAL model loader, ahead of the guider. DeverStyle states its samples were made \"by setting the LORA @ 0.6 strength on the model only, not on the model_negative\".", + "Model-only loaders throughout. The text encoder is a frozen Qwen3-VL-8B with no LoRA-patchable CLIP branch, so the model+clip `LoraLoader` has nothing useful to do.", + "To affect the unconditional/negative branch you must load a second, physically distinct model file and patch that one. Passing the same patched model object into both DualModelGuider inputs is a verified no-op (Comfy-Org issue #14521; kijai confirmed it is expected behaviour).", + "Trigger words go INSIDE the JSON caption, in the semantically matching field, not prepended to a prose string. DeverStyle instructs putting `dvr_arch_tv` in `style_description.art_style` (and in the content field); the Kiki LoRA card describes using `art_style: \"Studio Ghibli style\"` with `medium: \"anime_screencap\"`.", + "Two versions of the same LoRA can be stacked at reduced strengths: zjourney's author recommends v1 @ 0.35 + v2 @ 0.45 over either alone at full strength.", + "TurboTime changes the graph, not just the strength: it runs at cfg 1.0 and the reference community workflow swaps `DualModelGuider` for a single-model `CFGGuider`, dropping the unconditional model. Re-tune any stacked style LoRA downward.", + "A LoRA can be applied to only part of the run. The Gray Screen bypass LoRA is used at NEGATIVE weight (about -0.25) and, in its shipped workflows, only on the first step and then disabled; its author warns that \"LoRA being active through the whole generation could lead to degradation\".", + "The author never states which weight family this targets. Silent degradation, not hard errors, in every case observed. (1) Loading a conditional-trained LoRA onto the unconditional transformer succeeds silently because the architectures match; it perturbs the negative branch instead of the positive one. (2) Verified ComfyUI trap: in `Du" + ] + }, + "pros": [ + "Compact at 202 MB — cheap to stack with other LoRAs. [source: file size]" + ], + "cons": [ + "No trigger word — the LoRA is always on once loaded, so strength is the only control. [source: model metadata]", + "Author never states which generation mode it was trained for; test against your own workflow before relying on it. [source: description]", + "Thin documentation — expect trial and error. [source: description length]", + "Modest adoption (117 downloads) — less community feedback to rely on. [source: source stats]" + ], + "file": { + "name": "salomon_id4.safetensors", + "size_bytes": 211734000, + "size_mb": 201.93, + "sha256": "bcee988a63010113f9470d6d38f5213c829db5152ff01a79caffe576bd4b86b4" + }, + "stats": { + "downloads": 117, + "thumbs_up": 6, + "thumbs_down": 0 + }, + "popularity": { + "downloads": 117, + "thumbs_up": 6, + "thumbs_down": 0, + "comments": 0, + "approval": "100% positive (6/6)", + "age_days": 44.6, + "downloads_per_day": 0.85, + "rank_by_downloads": 34, + "rank_by_momentum": 24, + "of_total": 35, + "adoption_tier": "Niche", + "momentum": "Long tail · fresh release" + }, + "selection_score": 189.0, + "selection_reasons": [ + "117 downloads", + "6 thumbs-up", + "top-ranked for its category" + ], + "example_media": [ + { + "file": "01_image.jpeg", + "type": "image", + "resolution": "1088x1936", + "prompt": "" + }, + { + "file": "02_image.jpeg", + "type": "image", + "resolution": "1088x1936", + "prompt": "" + }, + { + "file": "03_image.jpeg", + "type": "image", + "resolution": "1088x1936", + "prompt": "" + } + ], + "tags": [ + "sneakers", + "salomon", + "fashion", + "clothing" + ], + "scraped_at": "2026-09-05T01:50:01+00:00", + "path": "ideogram-4/salomon-xt-6-sneakers" +} diff --git a/salomon-xt-6-sneakers/salomon_id4.safetensors b/salomon-xt-6-sneakers/salomon_id4.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..e10e6794262deb47ce72493d9fffbe2a4aef6493 --- /dev/null +++ b/salomon-xt-6-sneakers/salomon_id4.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bcee988a63010113f9470d6d38f5213c829db5152ff01a79caffe576bd4b86b4 +size 211734000 diff --git a/super-robot-style/example_images/01_image.jpeg b/super-robot-style/example_images/01_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..fc4932cf99971fb4225103224f4b8fe017d3685d --- /dev/null +++ b/super-robot-style/example_images/01_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a98733270b334cdeac932a59f5168a8666e743d4dc7f2377fdc324ed4cb349ff +size 144343 diff --git a/super-robot-style/example_images/02_image.jpeg b/super-robot-style/example_images/02_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..158abaaf8fe1c6213f720a6dfd8efaf772533870 --- /dev/null +++ b/super-robot-style/example_images/02_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4bfc3b839529cdc691fa5d73570d8d4a4c4cd3673ad1ce81e1bb9398098230f6 +size 2248254 diff --git a/super-robot-style/example_images/03_image.jpeg b/super-robot-style/example_images/03_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..7a0efd7e495637f9d7b44fd8af8fd34db664215a --- /dev/null +++ b/super-robot-style/example_images/03_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5e982feff70eb52aec2835e4ac1d13ae6ba50d9ea5e4bba5ebeb9d0c7811f663 +size 2208812 diff --git a/super-robot-style/example_images/04_image.jpeg b/super-robot-style/example_images/04_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..6117a33fe1673879712c4b7ffacea1cd0efee213 --- /dev/null +++ b/super-robot-style/example_images/04_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:43b26eab3f7884ea7bcc36b24fdbc76ce16efeda061410f6e336522205de029a +size 182460 diff --git a/super-robot-style/example_images/05_image.jpeg b/super-robot-style/example_images/05_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..8b97e43076722e0325960ff0afde76f6e9052928 --- /dev/null +++ b/super-robot-style/example_images/05_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:daf5ea75799a51a65c8a31040ad35179837ffa7614efc856af031cefca7664f3 +size 2035377 diff --git a/super-robot-style/example_images/06_image.jpeg b/super-robot-style/example_images/06_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..1307744e84c293a89aa2b0c2f08617be1abef276 --- /dev/null +++ b/super-robot-style/example_images/06_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0d97758fae83af7468c67f521153cbd3bec2f1be2f6a73740c101d59784efaf1 +size 169380 diff --git a/super-robot-style/idegoram4_ROBT.safetensors b/super-robot-style/idegoram4_ROBT.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b58390e07c41add35039ad1cec681a83181f7542 --- /dev/null +++ b/super-robot-style/idegoram4_ROBT.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:97d3f116d163d084553af2083e253560b5a610ad98ef8b47772490aeede2cf9d +size 211734000 diff --git a/super-robot-style/info.txt b/super-robot-style/info.txt new file mode 100644 index 0000000000000000000000000000000000000000..2a8baa8f074cc14369729dc4c17571605c0c9849 --- /dev/null +++ b/super-robot-style/info.txt @@ -0,0 +1,262 @@ +═══════════════════════════════════════════════════════════════════════════════ + SUPER ROBOT STYLE +═══════════════════════════════════════════════════════════════════════════════ + +─────────────────────────────────────────────────────────────────────────────── + IDENTITY +─────────────────────────────────────────────────────────────────────────────── + Base model Ideogram 4 + Generation mode not declared by the author + Checkpoint not determined — verify before use + Category Camera movement + Author allpleoleo439 + Source https://civitai.com/models/1029997 + Provenance CivitAI · model 1029997 · version 3051798 + +─────────────────────────────────────────────────────────────────────────────── + WEIGHTS +─────────────────────────────────────────────────────────────────────────────── + File idegoram4_ROBT.safetensors + Version v1.0 | ideogram4 + Size 201.9 MB + Format SafeTensor / unknown precision + SHA256 97d3f116d163d084553af2083e253560b5a610ad98ef8b47772490aeede2cf9d + Published 2026-06-20 + +─────────────────────────────────────────────────────────────────────────────── + TRIGGER WORDS +─────────────────────────────────────────────────────────────────────────────── + No trigger word. The LoRA applies as soon as it is loaded — + strength is the only way to dial the effect up or down. + +─────────────────────────────────────────────────────────────────────────────── + RECOMMENDED SETTINGS +─────────────────────────────────────────────────────────────────────────────── + The author did not publish specific settings. + No author settings were published. Community baseline: steps 48 is the + reference default: docs/inference.md says "`V4_QUALITY_48` is the default" + and the pipeline's `num_steps` default is 48; diffusers documents the same + 48-step default. 20 is the practical ComfyUI default — the `V4_DEFAULT_20` + preset and the value the shipped template `image_ideogram4_t2i.json` + selects (its preset combo defaults to "Default"). 12 = `V4_TURBO_12`. With + the TurboTime LoRA: 8 in the nomadoor community workflow; the model card + claims usable output down to 2., cfg A guidance SCHEDULE, not one number, + and the negative branch comes from a second model rather than a negative + prompt. Verified from src/ideogram4/sampler_configs.py: V4_QUALITY_48 = 45 + steps @ gw 7 then 3 polish steps @ gw 3; V4_DEFAULT_20 = 18 @ 7 + 2 @ 3; + V4_TURBO_12 = 11 @ 7 + 1 @ 3. Pipeline default `guidance_scale` = 7.0 when + no schedule is given. In ComfyUI the shipped template sets + `DualModelGuider` cfg 7 plus a `CFGOverride` node with widgets [3, 0.7, 1] + (cfg 3 from 0.7→1.0 of the run); kijai recommends starting the override at + 0.7, not 0.9. TurboTime: cfg 1.0 / no CFG., sampler Euler flow-matching. In + ComfyUI: `KSamplerSelect` → euler, driven by `SamplerCustomAdvanced`, with + the model-specific `Ideogram4Scheduler` (logit-normal sigma schedule). Its + params are `mu` and `std`, not the usual `shift`. Reference presets: mu 0.0 + / std 1.5 (48-step), mu 0.0 / std 1.75 (20-step), mu 0.5 / std 1.75 + (12-step). The Ideogram4Scheduler node's own widget defaults are steps 20, + mu 0.5, std 1.75, but the shipped template overrides mu/std/steps from the + preset JSON, so out of the box it runs 20 / mu 0.0 / std 1.75. mu is auto- + shifted by resolution in both implementations: mean = mu + + 0.5*log((w*h)/(512*512))., resolution 1024×1024 default (pipeline + `height`/`width` default 1024). Any multiple of 16 per side from 256 to + 2048, aspect ratios up to 6:1 or 1:6 — one model covers all of it. + 2048×2048 with V4_QUALITY_48 for maximum quality. Buckets documented in + docs/inference.md: 1536×1024 (3:2), 1024×1536 (2:3), 1920×1088 (~16:9), + 2048×768 (~21:9), 1024×1792 (~9:16), 1600×400 (4:1)., lora_strength No + single official number; these are author-stated values. DeverStyle Archer: + 0.6, and explicitly "on the model only, not on the model_negative". + zjourney v2: 0.55 standalone (range 0.35–0.6), or v1 @ 0.35 + v2 @ 0.45 + stacked. Furry Enhancer: the model page says 0.6–0.8, but its own version + notes ask for 0.7–0.9 (v1.94) and 0.8–0.95 (v2.8). TurboTime: 0.6 in the + nomadoor workflow (the author gives no figure). So ~0.5–0.9 is the + realistic band, read the specific card.. + +─────────────────────────────────────────────────────────────────────────────── + ADOPTION & POPULARITY +─────────────────────────────────────────────────────────────────────────────── + Downloads 144 + Rating 100% positive (21/21) + Comments 0 + Published 77 days ago + Download rate 0.6/day since release + + Adoption rank #33 of 35 in this collection (Niche) + Momentum rank #29 of 35 by download rate (Long tail) + + Adoption tier is the percentile of total downloads within this collection; + momentum is the percentile of downloads-per-day since release. Momentum is + ranked rather than measured against a fixed rate, because this base model + is itself new and every LoRA looks fast on an absolute scale. Both numbers + are CivitAI's, read on the scrape date at the foot of this file. + +─────────────────────────────────────────────────────────────────────────────── + CHAINING / STACKING +─────────────────────────────────────────────────────────────────────────────── + Stacks with others not stated + + The author gave no chaining guidance. The rules below are the + general Ideogram 4 ones, not anything specific to this LoRA. + + General Ideogram 4 rules that apply here (see REFERENCE.txt): + + • Chain `LoraLoaderModelOnly` nodes in series off the CONDITIONAL model + loader, ahead of the guider. DeverStyle states its samples were made + "by setting the LORA @ 0.6 strength on the model only, not on the + model_negative". + • Model-only loaders throughout. The text encoder is a frozen Qwen3-VL-8B + with no LoRA-patchable CLIP branch, so the model+clip `LoraLoader` has + nothing useful to do. + • To affect the unconditional/negative branch you must load a second, + physically distinct model file and patch that one. Passing the same + patched model object into both DualModelGuider inputs is a verified no- + op (Comfy-Org issue #14521; kijai confirmed it is expected behaviour). + • Trigger words go INSIDE the JSON caption, in the semantically matching + field, not prepended to a prose string. DeverStyle instructs putting + `dvr_arch_tv` in `style_description.art_style` (and in the content + field); the Kiki LoRA card describes using `art_style: "Studio Ghibli + style"` with `medium: "anime_screencap"`. + • Two versions of the same LoRA can be stacked at reduced strengths: + zjourney's author recommends v1 @ 0.35 + v2 @ 0.45 over either alone at + full strength. + • TurboTime changes the graph, not just the strength: it runs at cfg 1.0 + and the reference community workflow swaps `DualModelGuider` for a + single-model `CFGGuider`, dropping the unconditional model. Re-tune any + stacked style LoRA downward. + • A LoRA can be applied to only part of the run. The Gray Screen bypass + LoRA is used at NEGATIVE weight (about -0.25) and, in its shipped + workflows, only on the first step and then disabled; its author warns + that "LoRA being active through the whole generation could lead to + degradation". + • The author never states which weight family this targets. Silent + degradation, not hard errors, in every case observed. (1) Loading a + conditional-trained LoRA onto the unconditional transformer succeeds + silently because the architectures match; it perturbs the negative + branch instead of the positive one. (2) Verified ComfyUI trap: in `Du + +─────────────────────────────────────────────────────────────────────────────── + PROS +─────────────────────────────────────────────────────────────────────────────── + + Compact at 202 MB — cheap to stack with other LoRAs. [source: file + size] + +─────────────────────────────────────────────────────────────────────────────── + CONS & CAVEATS +─────────────────────────────────────────────────────────────────────────────── + - Needs a reduced strength; the default 1.0 is likely too strong. + [source: description] + - No trigger word — the LoRA is always on once loaded, so strength is the + only control. [source: model metadata] + - Author never states which generation mode it was trained for; test + against your own workflow before relying on it. [source: description] + - Modest adoption (144 downloads) — less community feedback to rely on. + [source: source stats] + +─────────────────────────────────────────────────────────────────────────────── + COMPATIBILITY WARNING +─────────────────────────────────────────────────────────────────────────────── + Silent degradation, not hard errors, in every case observed. (1) Loading a + conditional-trained LoRA onto the unconditional transformer succeeds + silently because the architectures match; it perturbs the negative branch + instead of the positive one. (2) Verified ComfyUI trap: in + `DualModelGuider`, passing the SAME model object as both the conditional + model and the LoRA-patched unconditional model is a no-op. Comfy-Org issue + #14521 tabulates it — case 3 (DualModelGuider, conditional model as + positive, conditional model + LoRA as negative) was "exactly identical" to + plain conditional-only CFG, while case 4 with the real unconditional model + differed. kijai confirmed it is expected: "there's no (working) mechanic to + swap LoRA on each step on the same model, it would only work if the model + was a different one." To patch the negative branch you must load a + genuinely separate model file. (3) A diffusers/PEFT-prefixed LoRA in + ComfyUI is expected to produce only a "lora key not loaded" console warning + and no visible effect — this follows from the absent key mapping (see + checkpoint_families) rather than from a published Ideogram-4 bug report. + (4) Cross-quantization loading (fp8-trained LoRA onto nvfp4/nf4/GGUF) + generally works with some precision loss; int8_convrot had a specific, now- + fixed bug (see gotchas). + +─────────────────────────────────────────────────────────────────────────────── + AUTHOR'S DESCRIPTION +─────────────────────────────────────────────────────────────────────────────── + Using this LORA, you can easily generate images of Japanese-style robot + cartoons. The lower the weight, the closer it gets to chibi (two-head + proportion) illustrations. If you want a cool-looking robot without it + appearing too childish, use a weight between 0.45 and 0.65 . For images + that already depict a robot, a lower weight is sufficient, but for those + that do not, a higher weight may be required. + このLORAを使用すると、日本風のロボットアニメの画像を簡単に生成することができます。重みを低く設定すると、2頭身のイラストに近づきます。かっこいいロ + ボットを描きたいが、子供っぽすぎない場合は、重みを0.45～0.65の間で設定すると良いです。元々ロボットを描いた画像の場合は、低い重みで十分ですが、 + ロボット以外の画像を使用する場合は、より高い重みが必要になります。 + +─────────────────────────────────────────────────────────────────────────────── + EXAMPLE MEDIA +─────────────────────────────────────────────────────────────────────────────── + 6 preview(s) in ./example_images/ — community results for + this LoRA. Videos keep a matching _poster.jpg still frame. + + No prompts are recorded below: CivitAI's public API flags these posts as + having generation metadata but does not return it, so the prompts cannot be + scraped. Open the model page to read them. + + 01_image.jpeg (image, 1024x1024) + 02_image.jpeg (image, 1184x896) + 03_image.jpeg (image, 1376x768) + 04_image.jpeg (image, 1024x1024) + 05_image.jpeg (image, 1376x768) + 06_image.jpeg (image, 1024x1024) + +─────────────────────────────────────────────────────────────────────────────── + HOW TO USE (COMFYUI) +─────────────────────────────────────────────────────────────────────────────── + Use ComfyUI's NATIVE Ideogram 4 support (Comfy-Org/ComfyUI PR #14259 "feat: + Add Ideogram4 model support", merged 2026-06-03) — not the official + `ComfyUI-Ideogram4` custom-node wrapper, whose nodes (Magic Prompt / + Pipeline Loader / Generate) wrap Ideogram's reference pipeline and contain + no LoRA input at all (verified: no occurrence of "lora" in its nodes.py). + + 1. Update ComfyUI to a build containing `comfy_extras/nodes_ideogram4.py` + (which supplies `Ideogram4Scheduler`); the same PR added the generic + `DualModelGuider` and `CFGOverride` nodes, which live in + `comfy_extras/nodes_custom_sampler.py`. + 2. From `Comfy-Org/Ideogram-4` download into + `ComfyUI/models/diffusion_models/`: `ideogram4_fp8_scaled.safetensors` + (9.28 GB) AND `ideogram4_unconditional_fp8_scaled.safetensors` (9.28 GB). + Lower-VRAM alternatives in the same folder: `*_nvfp4_mixed.safetensors` + (5.49 GB each) or `*_int8_convrot.safetensors` (9.58 GB each). + 3. Download `qwen3vl_8b_fp8_scaled.safetensors` (10.59 GB; or + `qwen3vl_8b_nvfp4.safetensors`, 6.31 GB) into + `ComfyUI/models/text_encoders/`. The shipped template also lists an + optional `gemma4_e4b_it_fp8_scaled.safetensors` from `Comfy-Org/gemma-4` + for local prompt building. + 4. Download `flux2-vae.safetensors` (0.34 GB, from Comfy-Org/Ideogram-4 or + Comfy-Org/flux2-dev) into `ComfyUI/models/vae/`. + 5. Put LoRA .safetensors in `ComfyUI/models/loras/` (no subfolder + convention enforced). + 6. Load the built-in template `image_ideogram4_t2i` (Workflow → Browse + Templates), or the workflow from + docs.comfy.org/tutorials/image/ideogram/ideogram-v4. + 7. Insert a `LoraLoaderModelOnly` between the main `UNETLoader` and the + `DualModelGuider`'s conditional model input. There is no CLIP-side LoRA + here — the text encoder is a frozen Qwen3-VL — so model-only is the right + loader. Strength ~0.5–0.8. + 8. Do NOT reuse one patched model object for both guider inputs; it + silently does nothing. Patching the negative branch requires a second, + physically separate model file with its own loader. + 9. Write the prompt as a JSON caption into `CLIPTextEncode` (the CLIPLoader + type is `ideogram4`); keep `ConditioningZeroOut` feeding the unconditional + branch. + + Non-ComfyUI paths: diffusers — + `Ideogram4Pipeline.from_pretrained("ideogram-ai/ideogram-4-nf4")` (the + docs' own example; `ideogram-ai/ideogram-4-nf4-diffusers` also exists), + then `pipeline.load_lora_weights(...)`. Training — `ostris/ai-toolkit` + (built-in `extensions_built_in/diffusion_models/ideogram4/` plus an + `Ideogram4Captioner` that writes JSON captions) or diffusers + `examples/dreambooth/train_dreambooth_lora_ideogram4.py`. The official HF + weights are GATED (`gated: auto`): accept the license on the model page and + authenticate (`hf auth login` / `HF_TOKEN`) or downloads fail. + +─────────────────────────────────────────────────────────────────────────────── + Scraped by lorakit on 2026-09-05 01:49 UTC. + Stats and description are the author's; pros/cons are derived from the + evidence tagged beside each line. +─────────────────────────────────────────────────────────────────────────────── diff --git a/super-robot-style/metadata.json b/super-robot-style/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..754f820dd1ddf15a7f38b085669bcb3490c2626d --- /dev/null +++ b/super-robot-style/metadata.json @@ -0,0 +1,132 @@ +{ + "name": "Super Robot Style", + "slug": "super-robot-style", + "base_model": "Ideogram 4", + "media_type": "image", + "model_id": 1029997, + "version_id": 3051798, + "version_name": "v1.0 | ideogram4", + "source": "civitai", + "url": "https://civitai.com/models/1029997", + "repo_id": null, + "file_path": null, + "author": "allpleoleo439", + "category": { + "key": "camera", + "label": "Camera movement" + }, + "modalities": [], + "primary_modality": "unspecified", + "checkpoint_family": "unknown", + "trigger_words": [], + "recommended_settings": {}, + "chaining": { + "author": { + "stacks": "not stated", + "notes": "", + "avoid": [], + "requires": [], + "order": "", + "strength": "", + "stated": false + }, + "general_rules": [ + "Chain `LoraLoaderModelOnly` nodes in series off the CONDITIONAL model loader, ahead of the guider. DeverStyle states its samples were made \"by setting the LORA @ 0.6 strength on the model only, not on the model_negative\".", + "Model-only loaders throughout. The text encoder is a frozen Qwen3-VL-8B with no LoRA-patchable CLIP branch, so the model+clip `LoraLoader` has nothing useful to do.", + "To affect the unconditional/negative branch you must load a second, physically distinct model file and patch that one. Passing the same patched model object into both DualModelGuider inputs is a verified no-op (Comfy-Org issue #14521; kijai confirmed it is expected behaviour).", + "Trigger words go INSIDE the JSON caption, in the semantically matching field, not prepended to a prose string. DeverStyle instructs putting `dvr_arch_tv` in `style_description.art_style` (and in the content field); the Kiki LoRA card describes using `art_style: \"Studio Ghibli style\"` with `medium: \"anime_screencap\"`.", + "Two versions of the same LoRA can be stacked at reduced strengths: zjourney's author recommends v1 @ 0.35 + v2 @ 0.45 over either alone at full strength.", + "TurboTime changes the graph, not just the strength: it runs at cfg 1.0 and the reference community workflow swaps `DualModelGuider` for a single-model `CFGGuider`, dropping the unconditional model. Re-tune any stacked style LoRA downward.", + "A LoRA can be applied to only part of the run. The Gray Screen bypass LoRA is used at NEGATIVE weight (about -0.25) and, in its shipped workflows, only on the first step and then disabled; its author warns that \"LoRA being active through the whole generation could lead to degradation\".", + "The author never states which weight family this targets. Silent degradation, not hard errors, in every case observed. (1) Loading a conditional-trained LoRA onto the unconditional transformer succeeds silently because the architectures match; it perturbs the negative branch instead of the positive one. (2) Verified ComfyUI trap: in `Du" + ] + }, + "pros": [ + "Compact at 202 MB — cheap to stack with other LoRAs. [source: file size]" + ], + "cons": [ + "Needs a reduced strength; the default 1.0 is likely too strong. [source: description]", + "No trigger word — the LoRA is always on once loaded, so strength is the only control. [source: model metadata]", + "Author never states which generation mode it was trained for; test against your own workflow before relying on it. [source: description]", + "Modest adoption (144 downloads) — less community feedback to rely on. [source: source stats]" + ], + "file": { + "name": "idegoram4_ROBT.safetensors", + "size_bytes": 211734000, + "size_mb": 201.93, + "sha256": "97d3f116d163d084553af2083e253560b5a610ad98ef8b47772490aeede2cf9d" + }, + "stats": { + "downloads": 144, + "thumbs_up": 21, + "thumbs_down": 0 + }, + "popularity": { + "downloads": 144, + "thumbs_up": 21, + "thumbs_down": 0, + "comments": 0, + "approval": "100% positive (21/21)", + "age_days": 76.8, + "downloads_per_day": 0.6, + "rank_by_downloads": 33, + "rank_by_momentum": 29, + "of_total": 35, + "adoption_tier": "Niche", + "momentum": "Long tail" + }, + "selection_score": 396.0, + "selection_reasons": [ + "144 downloads", + "21 thumbs-up", + "top-ranked for its category" + ], + "example_media": [ + { + "file": "01_image.jpeg", + "type": "image", + "resolution": "1024x1024", + "prompt": "" + }, + { + "file": "02_image.jpeg", + "type": "image", + "resolution": "1184x896", + "prompt": "" + }, + { + "file": "03_image.jpeg", + "type": "image", + "resolution": "1376x768", + "prompt": "" + }, + { + "file": "04_image.jpeg", + "type": "image", + "resolution": "1024x1024", + "prompt": "" + }, + { + "file": "05_image.jpeg", + "type": "image", + "resolution": "1376x768", + "prompt": "" + }, + { + "file": "06_image.jpeg", + "type": "image", + "resolution": "1024x1024", + "prompt": "" + } + ], + "tags": [ + "anime", + "japanime", + "japan", + "cartoon", + "style", + "robot" + ], + "scraped_at": "2026-09-05T01:49:54+00:00", + "path": "ideogram-4/super-robot-style" +} diff --git a/taiwan-bento-packed-lunch/example_images/01_image.jpeg b/taiwan-bento-packed-lunch/example_images/01_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..5c96b0eb1c4c730ac2a7917a78c29bab320609c2 --- /dev/null +++ b/taiwan-bento-packed-lunch/example_images/01_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:33b8d646097942e8fbc598e20435099ca74a8dcb327b503bd412194802414960 +size 2031552 diff --git a/taiwan-bento-packed-lunch/example_images/02_image.jpeg b/taiwan-bento-packed-lunch/example_images/02_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..cc11f9262542fb4a39867db77b873e688212db59 --- /dev/null +++ b/taiwan-bento-packed-lunch/example_images/02_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5fe53a0fc160c67aad229c04b9f50f1c2d0e0d5dc0ad6b3f94bc8bb9cb3f0ef5 +size 2167975 diff --git a/taiwan-bento-packed-lunch/example_images/03_image.jpeg b/taiwan-bento-packed-lunch/example_images/03_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..1e73b7c3441d9ae7716d5c19806a1f035b81c982 --- /dev/null +++ b/taiwan-bento-packed-lunch/example_images/03_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:83c91b6304471aafb3d641bdd6f38f496922a32f68caf45a6124717a9a186ad2 +size 1985200 diff --git a/taiwan-bento-packed-lunch/example_images/04_image.jpeg b/taiwan-bento-packed-lunch/example_images/04_image.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..4221d6a95c0d8cf86baa4603fa6ee6d7dcbba385 --- /dev/null +++ b/taiwan-bento-packed-lunch/example_images/04_image.jpeg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9cfd680a8eabf6681422061eb184254c476e2b73d5b69c1377224d7e7197bbe0 +size 2013094 diff --git a/taiwan-bento-packed-lunch/idegoram4_bando.safetensors b/taiwan-bento-packed-lunch/idegoram4_bando.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..cbc4452f023b4c003301648140b0b50562302084 --- /dev/null +++ b/taiwan-bento-packed-lunch/idegoram4_bando.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:271f930729f3e3ccf32644faf7ebfaf75a010913605434eed03bc775d1cd4855 +size 211734000 diff --git a/taiwan-bento-packed-lunch/info.txt b/taiwan-bento-packed-lunch/info.txt new file mode 100644 index 0000000000000000000000000000000000000000..7d3d60a4c32b7f37d95614541b6186101c6286ff --- /dev/null +++ b/taiwan-bento-packed-lunch/info.txt @@ -0,0 +1,258 @@ +═══════════════════════════════════════════════════════════════════════════════ + TAIWAN BENTO (PACKED LUNCH) | 便當 +═══════════════════════════════════════════════════════════════════════════════ + +─────────────────────────────────────────────────────────────────────────────── + IDENTITY +─────────────────────────────────────────────────────────────────────────────── + Base model Ideogram 4 + Generation mode not declared by the author + Checkpoint not determined — verify before use + Category Visual style + Author allpleoleo439 + Source https://civitai.com/models/952849 + Provenance CivitAI · model 952849 · version 3052339 + +─────────────────────────────────────────────────────────────────────────────── + WEIGHTS +─────────────────────────────────────────────────────────────────────────────── + File idegoram4_bando.safetensors + Version v1.0 | ideogram4 + Size 201.9 MB + Format SafeTensor / unknown precision + SHA256 271f930729f3e3ccf32644faf7ebfaf75a010913605434eed03bc775d1cd4855 + Published 2026-06-20 + +─────────────────────────────────────────────────────────────────────────────── + TRIGGER WORDS +─────────────────────────────────────────────────────────────────────────────── + No trigger word. The LoRA applies as soon as it is loaded — + strength is the only way to dial the effect up or down. + +─────────────────────────────────────────────────────────────────────────────── + RECOMMENDED SETTINGS +─────────────────────────────────────────────────────────────────────────────── + The author did not publish specific settings. + No author settings were published. Community baseline: steps 48 is the + reference default: docs/inference.md says "`V4_QUALITY_48` is the default" + and the pipeline's `num_steps` default is 48; diffusers documents the same + 48-step default. 20 is the practical ComfyUI default — the `V4_DEFAULT_20` + preset and the value the shipped template `image_ideogram4_t2i.json` + selects (its preset combo defaults to "Default"). 12 = `V4_TURBO_12`. With + the TurboTime LoRA: 8 in the nomadoor community workflow; the model card + claims usable output down to 2., cfg A guidance SCHEDULE, not one number, + and the negative branch comes from a second model rather than a negative + prompt. Verified from src/ideogram4/sampler_configs.py: V4_QUALITY_48 = 45 + steps @ gw 7 then 3 polish steps @ gw 3; V4_DEFAULT_20 = 18 @ 7 + 2 @ 3; + V4_TURBO_12 = 11 @ 7 + 1 @ 3. Pipeline default `guidance_scale` = 7.0 when + no schedule is given. In ComfyUI the shipped template sets + `DualModelGuider` cfg 7 plus a `CFGOverride` node with widgets [3, 0.7, 1] + (cfg 3 from 0.7→1.0 of the run); kijai recommends starting the override at + 0.7, not 0.9. TurboTime: cfg 1.0 / no CFG., sampler Euler flow-matching. In + ComfyUI: `KSamplerSelect` → euler, driven by `SamplerCustomAdvanced`, with + the model-specific `Ideogram4Scheduler` (logit-normal sigma schedule). Its + params are `mu` and `std`, not the usual `shift`. Reference presets: mu 0.0 + / std 1.5 (48-step), mu 0.0 / std 1.75 (20-step), mu 0.5 / std 1.75 + (12-step). The Ideogram4Scheduler node's own widget defaults are steps 20, + mu 0.5, std 1.75, but the shipped template overrides mu/std/steps from the + preset JSON, so out of the box it runs 20 / mu 0.0 / std 1.75. mu is auto- + shifted by resolution in both implementations: mean = mu + + 0.5*log((w*h)/(512*512))., resolution 1024×1024 default (pipeline + `height`/`width` default 1024). Any multiple of 16 per side from 256 to + 2048, aspect ratios up to 6:1 or 1:6 — one model covers all of it. + 2048×2048 with V4_QUALITY_48 for maximum quality. Buckets documented in + docs/inference.md: 1536×1024 (3:2), 1024×1536 (2:3), 1920×1088 (~16:9), + 2048×768 (~21:9), 1024×1792 (~9:16), 1600×400 (4:1)., lora_strength No + single official number; these are author-stated values. DeverStyle Archer: + 0.6, and explicitly "on the model only, not on the model_negative". + zjourney v2: 0.55 standalone (range 0.35–0.6), or v1 @ 0.35 + v2 @ 0.45 + stacked. Furry Enhancer: the model page says 0.6–0.8, but its own version + notes ask for 0.7–0.9 (v1.94) and 0.8–0.95 (v2.8). TurboTime: 0.6 in the + nomadoor workflow (the author gives no figure). So ~0.5–0.9 is the + realistic band, read the specific card.. + +─────────────────────────────────────────────────────────────────────────────── + ADOPTION & POPULARITY +─────────────────────────────────────────────────────────────────────────────── + Downloads 196 + Rating 100% positive (21/21) + Comments 0 + Published 77 days ago + Download rate 0.6/day since release + + Adoption rank #30 of 35 in this collection (Niche) + Momentum rank #30 of 35 by download rate (Long tail) + + Adoption tier is the percentile of total downloads within this collection; + momentum is the percentile of downloads-per-day since release. Momentum is + ranked rather than measured against a fixed rate, because this base model + is itself new and every LoRA looks fast on an absolute scale. Both numbers + are CivitAI's, read on the scrape date at the foot of this file. + +─────────────────────────────────────────────────────────────────────────────── + CHAINING / STACKING +─────────────────────────────────────────────────────────────────────────────── + Stacks with others not stated + + The author gave no chaining guidance. The rules below are the + general Ideogram 4 ones, not anything specific to this LoRA. + + General Ideogram 4 rules that apply here (see REFERENCE.txt): + + • Chain `LoraLoaderModelOnly` nodes in series off the CONDITIONAL model + loader, ahead of the guider. DeverStyle states its samples were made + "by setting the LORA @ 0.6 strength on the model only, not on the + model_negative". + • Model-only loaders throughout. The text encoder is a frozen Qwen3-VL-8B + with no LoRA-patchable CLIP branch, so the model+clip `LoraLoader` has + nothing useful to do. + • To affect the unconditional/negative branch you must load a second, + physically distinct model file and patch that one. Passing the same + patched model object into both DualModelGuider inputs is a verified no- + op (Comfy-Org issue #14521; kijai confirmed it is expected behaviour). + • Trigger words go INSIDE the JSON caption, in the semantically matching + field, not prepended to a prose string. DeverStyle instructs putting + `dvr_arch_tv` in `style_description.art_style` (and in the content + field); the Kiki LoRA card describes using `art_style: "Studio Ghibli + style"` with `medium: "anime_screencap"`. + • Two versions of the same LoRA can be stacked at reduced strengths: + zjourney's author recommends v1 @ 0.35 + v2 @ 0.45 over either alone at + full strength. + • TurboTime changes the graph, not just the strength: it runs at cfg 1.0 + and the reference community workflow swaps `DualModelGuider` for a + single-model `CFGGuider`, dropping the unconditional model. Re-tune any + stacked style LoRA downward. + • A LoRA can be applied to only part of the run. The Gray Screen bypass + LoRA is used at NEGATIVE weight (about -0.25) and, in its shipped + workflows, only on the first step and then disabled; its author warns + that "LoRA being active through the whole generation could lead to + degradation". + • The author never states which weight family this targets. Silent + degradation, not hard errors, in every case observed. (1) Loading a + conditional-trained LoRA onto the unconditional transformer succeeds + silently because the architectures match; it perturbs the negative + branch instead of the positive one. (2) Verified ComfyUI trap: in `Du + +─────────────────────────────────────────────────────────────────────────────── + PROS +─────────────────────────────────────────────────────────────────────────────── + + Compact at 202 MB — cheap to stack with other LoRAs. [source: file + size] + +─────────────────────────────────────────────────────────────────────────────── + CONS & CAVEATS +─────────────────────────────────────────────────────────────────────────────── + - No trigger word — the LoRA is always on once loaded, so strength is the + only control. [source: model metadata] + - Author never states which generation mode it was trained for; test + against your own workflow before relying on it. [source: description] + - Modest adoption (196 downloads) — less community feedback to rely on. + [source: source stats] + +─────────────────────────────────────────────────────────────────────────────── + COMPATIBILITY WARNING +─────────────────────────────────────────────────────────────────────────────── + Silent degradation, not hard errors, in every case observed. (1) Loading a + conditional-trained LoRA onto the unconditional transformer succeeds + silently because the architectures match; it perturbs the negative branch + instead of the positive one. (2) Verified ComfyUI trap: in + `DualModelGuider`, passing the SAME model object as both the conditional + model and the LoRA-patched unconditional model is a no-op. Comfy-Org issue + #14521 tabulates it — case 3 (DualModelGuider, conditional model as + positive, conditional model + LoRA as negative) was "exactly identical" to + plain conditional-only CFG, while case 4 with the real unconditional model + differed. kijai confirmed it is expected: "there's no (working) mechanic to + swap LoRA on each step on the same model, it would only work if the model + was a different one." To patch the negative branch you must load a + genuinely separate model file. (3) A diffusers/PEFT-prefixed LoRA in + ComfyUI is expected to produce only a "lora key not loaded" console warning + and no visible effect — this follows from the absent key mapping (see + checkpoint_families) rather than from a published Ideogram-4 bug report. + (4) Cross-quantization loading (fp8-trained LoRA onto nvfp4/nf4/GGUF) + generally works with some precision loss; int8_convrot had a specific, now- + fixed bug (see gotchas). + +─────────────────────────────────────────────────────────────────────────────── + AUTHOR'S DESCRIPTION +─────────────────────────────────────────────────────────────────────────────── + Using this LORA, you can generate images of a typical Taiwanese bento box. + You can customize the contents, such as sausage, chicken drumstick, fish, + or pork chop. This LORA is specifically trained on the early "four- + compartment" bento style, which includes three small sections for side + dishes and one large section for the main dish and staple food. You can use + other types of containers, but the results may not be as ideal. + このLORAを使用すると、典型的な台湾の弁当の画像を生成することができます。弁当の内容を変更することができ、例えば、ソーセージ、鶏もも肉、魚、スペアリ + ブなどを選ぶことができます。このLORAは、初期の四分割スタイルの弁当を基にトレーニングしました。具体的には、小さな3つの仕切りに副菜を入れ、大きな仕 + 切りに主食と主菜を配置します。他の容器を使用することも可能ですが、その場合、効果があまり良くないかもしれません。 + +─────────────────────────────────────────────────────────────────────────────── + EXAMPLE MEDIA +─────────────────────────────────────────────────────────────────────────────── + 4 preview(s) in ./example_images/ — community results for + this LoRA. Videos keep a matching _poster.jpg still frame. + + No prompts are recorded below: CivitAI's public API flags these posts as + having generation metadata but does not return it, so the prompts cannot be + scraped. Open the model page to read them. + + 01_image.jpeg (image, 768x1376) + 02_image.jpeg (image, 768x1376) + 03_image.jpeg (image, 768x1376) + 04_image.jpeg (image, 768x1376) + +─────────────────────────────────────────────────────────────────────────────── + HOW TO USE (COMFYUI) +─────────────────────────────────────────────────────────────────────────────── + Use ComfyUI's NATIVE Ideogram 4 support (Comfy-Org/ComfyUI PR #14259 "feat: + Add Ideogram4 model support", merged 2026-06-03) — not the official + `ComfyUI-Ideogram4` custom-node wrapper, whose nodes (Magic Prompt / + Pipeline Loader / Generate) wrap Ideogram's reference pipeline and contain + no LoRA input at all (verified: no occurrence of "lora" in its nodes.py). + + 1. Update ComfyUI to a build containing `comfy_extras/nodes_ideogram4.py` + (which supplies `Ideogram4Scheduler`); the same PR added the generic + `DualModelGuider` and `CFGOverride` nodes, which live in + `comfy_extras/nodes_custom_sampler.py`. + 2. From `Comfy-Org/Ideogram-4` download into + `ComfyUI/models/diffusion_models/`: `ideogram4_fp8_scaled.safetensors` + (9.28 GB) AND `ideogram4_unconditional_fp8_scaled.safetensors` (9.28 GB). + Lower-VRAM alternatives in the same folder: `*_nvfp4_mixed.safetensors` + (5.49 GB each) or `*_int8_convrot.safetensors` (9.58 GB each). + 3. Download `qwen3vl_8b_fp8_scaled.safetensors` (10.59 GB; or + `qwen3vl_8b_nvfp4.safetensors`, 6.31 GB) into + `ComfyUI/models/text_encoders/`. The shipped template also lists an + optional `gemma4_e4b_it_fp8_scaled.safetensors` from `Comfy-Org/gemma-4` + for local prompt building. + 4. Download `flux2-vae.safetensors` (0.34 GB, from Comfy-Org/Ideogram-4 or + Comfy-Org/flux2-dev) into `ComfyUI/models/vae/`. + 5. Put LoRA .safetensors in `ComfyUI/models/loras/` (no subfolder + convention enforced). + 6. Load the built-in template `image_ideogram4_t2i` (Workflow → Browse + Templates), or the workflow from + docs.comfy.org/tutorials/image/ideogram/ideogram-v4. + 7. Insert a `LoraLoaderModelOnly` between the main `UNETLoader` and the + `DualModelGuider`'s conditional model input. There is no CLIP-side LoRA + here — the text encoder is a frozen Qwen3-VL — so model-only is the right + loader. Strength ~0.5–0.8. + 8. Do NOT reuse one patched model object for both guider inputs; it + silently does nothing. Patching the negative branch requires a second, + physically separate model file with its own loader. + 9. Write the prompt as a JSON caption into `CLIPTextEncode` (the CLIPLoader + type is `ideogram4`); keep `ConditioningZeroOut` feeding the unconditional + branch. + + Non-ComfyUI paths: diffusers — + `Ideogram4Pipeline.from_pretrained("ideogram-ai/ideogram-4-nf4")` (the + docs' own example; `ideogram-ai/ideogram-4-nf4-diffusers` also exists), + then `pipeline.load_lora_weights(...)`. Training — `ostris/ai-toolkit` + (built-in `extensions_built_in/diffusion_models/ideogram4/` plus an + `Ideogram4Captioner` that writes JSON captions) or diffusers + `examples/dreambooth/train_dreambooth_lora_ideogram4.py`. The official HF + weights are GATED (`gated: auto`): accept the license on the model page and + authenticate (`hf auth login` / `HF_TOKEN`) or downloads fail. + +─────────────────────────────────────────────────────────────────────────────── + Scraped by lorakit on 2026-09-05 01:49 UTC. + Stats and description are the author's; pros/cons are derived from the + evidence tagged beside each line. +─────────────────────────────────────────────────────────────────────────────── diff --git a/taiwan-bento-packed-lunch/metadata.json b/taiwan-bento-packed-lunch/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..f79bd4516d0799e75ef96ad07b36c73a36827c70 --- /dev/null +++ b/taiwan-bento-packed-lunch/metadata.json @@ -0,0 +1,116 @@ +{ + "name": "Taiwan bento (packed lunch) | 便當", + "slug": "taiwan-bento-packed-lunch", + "base_model": "Ideogram 4", + "media_type": "image", + "model_id": 952849, + "version_id": 3052339, + "version_name": "v1.0 | ideogram4", + "source": "civitai", + "url": "https://civitai.com/models/952849", + "repo_id": null, + "file_path": null, + "author": "allpleoleo439", + "category": { + "key": "style", + "label": "Visual style" + }, + "modalities": [], + "primary_modality": "unspecified", + "checkpoint_family": "unknown", + "trigger_words": [], + "recommended_settings": {}, + "chaining": { + "author": { + "stacks": "not stated", + "notes": "", + "avoid": [], + "requires": [], + "order": "", + "strength": "", + "stated": false + }, + "general_rules": [ + "Chain `LoraLoaderModelOnly` nodes in series off the CONDITIONAL model loader, ahead of the guider. DeverStyle states its samples were made \"by setting the LORA @ 0.6 strength on the model only, not on the model_negative\".", + "Model-only loaders throughout. The text encoder is a frozen Qwen3-VL-8B with no LoRA-patchable CLIP branch, so the model+clip `LoraLoader` has nothing useful to do.", + "To affect the unconditional/negative branch you must load a second, physically distinct model file and patch that one. Passing the same patched model object into both DualModelGuider inputs is a verified no-op (Comfy-Org issue #14521; kijai confirmed it is expected behaviour).", + "Trigger words go INSIDE the JSON caption, in the semantically matching field, not prepended to a prose string. DeverStyle instructs putting `dvr_arch_tv` in `style_description.art_style` (and in the content field); the Kiki LoRA card describes using `art_style: \"Studio Ghibli style\"` with `medium: \"anime_screencap\"`.", + "Two versions of the same LoRA can be stacked at reduced strengths: zjourney's author recommends v1 @ 0.35 + v2 @ 0.45 over either alone at full strength.", + "TurboTime changes the graph, not just the strength: it runs at cfg 1.0 and the reference community workflow swaps `DualModelGuider` for a single-model `CFGGuider`, dropping the unconditional model. Re-tune any stacked style LoRA downward.", + "A LoRA can be applied to only part of the run. The Gray Screen bypass LoRA is used at NEGATIVE weight (about -0.25) and, in its shipped workflows, only on the first step and then disabled; its author warns that \"LoRA being active through the whole generation could lead to degradation\".", + "The author never states which weight family this targets. Silent degradation, not hard errors, in every case observed. (1) Loading a conditional-trained LoRA onto the unconditional transformer succeeds silently because the architectures match; it perturbs the negative branch instead of the positive one. (2) Verified ComfyUI trap: in `Du" + ] + }, + "pros": [ + "Compact at 202 MB — cheap to stack with other LoRAs. [source: file size]" + ], + "cons": [ + "No trigger word — the LoRA is always on once loaded, so strength is the only control. [source: model metadata]", + "Author never states which generation mode it was trained for; test against your own workflow before relying on it. [source: description]", + "Modest adoption (196 downloads) — less community feedback to rely on. [source: source stats]" + ], + "file": { + "name": "idegoram4_bando.safetensors", + "size_bytes": 211734000, + "size_mb": 201.93, + "sha256": "271f930729f3e3ccf32644faf7ebfaf75a010913605434eed03bc775d1cd4855" + }, + "stats": { + "downloads": 196, + "thumbs_up": 21, + "thumbs_down": 0 + }, + "popularity": { + "downloads": 196, + "thumbs_up": 21, + "thumbs_down": 0, + "comments": 0, + "approval": "100% positive (21/21)", + "age_days": 76.6, + "downloads_per_day": 0.56, + "rank_by_downloads": 30, + "rank_by_momentum": 30, + "of_total": 35, + "adoption_tier": "Niche", + "momentum": "Long tail" + }, + "selection_score": 448.0, + "selection_reasons": [ + "196 downloads", + "21 thumbs-up", + "top-ranked for its category" + ], + "example_media": [ + { + "file": "01_image.jpeg", + "type": "image", + "resolution": "768x1376", + "prompt": "" + }, + { + "file": "02_image.jpeg", + "type": "image", + "resolution": "768x1376", + "prompt": "" + }, + { + "file": "03_image.jpeg", + "type": "image", + "resolution": "768x1376", + "prompt": "" + }, + { + "file": "04_image.jpeg", + "type": "image", + "resolution": "768x1376", + "prompt": "" + } + ], + "tags": [ + "objects", + "food", + "taiwan" + ], + "scraped_at": "2026-09-05T01:49:42+00:00", + "path": "ideogram-4/taiwan-bento-packed-lunch" +}