Add 10 LoRA(s): pagani-zonda-r-2009, mihoyo-style, disco-elysium, gray-screen-bypass-lora-and-workflow…

.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

art-the-clown-terrifier/4rtcl0wn_id4.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d6794f4b0c0866d9e6adb5be781c2ebd82ad268d5c4e31b82d5f3710cc0ccb0a
+size 1123105152

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.
+───────────────────────────────────────────────────────────────────────────────

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"
+}

disco-elysium/MixedElysium_Ideogram.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e762d7696337f6e50f78a7316a37ef2bcea847ab9a712f5ac5f1fa91697c0caa
+size 211734080

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.
+───────────────────────────────────────────────────────────────────────────────

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"
+}

dream-loom/Sime2_Ideogram.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fdf61f3f7f4bb948c30c5ed01e87e0b14fbee745c3cfac7dd0944833ee98f16e
+size 211734048

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.
+───────────────────────────────────────────────────────────────────────────────

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"
+}

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

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.
+───────────────────────────────────────────────────────────────────────────────

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"
+}

mihoyo-style/idegoram4_miha_re3_000003000.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8d6b75860935f6af02b2f89d5ae6a38e46d1312b60d12b3433db03da5c93ed86
+size 211734008

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.
+───────────────────────────────────────────────────────────────────────────────

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"
+}