Adds selected_seed outputs for CADE Easy and MG_SuperSimple while preserving existing output slot order.
Adds regression coverage for output contracts.
Skip NAG when context key dim mismatches layer
DZRobo
cdc8e19
Adds a check to bypass NAG cross-attention if the context's last dimension does not match the expected key dimension for the layer, preventing matmul crashes with mismatched context inputs.
Add Z_image support and Improve latent/channel handling
DZRobo
e69f3b7
Adds functions to harmonize latent channel counts and condition token lengths to prevent mismatches, especially for models like FLUX/Z_image. Enhances error reporting with debug output and traceback printing. Updates mg_combinode to better validate VAE/CLIP presence for checkpoint and input selection. Fixes hybrid sigma schedule alignment in mg_zesmart_sampler_v1_1.
Add smart seed diversity and config options
DZRobo
0f11662
Introduces 'smart_seed_k', 'smart_seed_steps', and 'smart_seed_diversity' parameters to the ComfyAdaptiveDetailEnhancer25 class and updates the smart seed selection logic to support diversity-based scoring. Updates mg_cade25.cfg to include these new smart seed options for each step. Also improves LoRA slot gating in mg_combinode.py to ignore empty, None, or disabled slots more strictly.
Add NAG fallback and fp32 decode options, update presets
DZRobo
0503191
Introduces NAG fallback logic to both easy and hard modules for improved guidance when CrossAttention patch is inactive. Adds an option to force VAE decode output to fp32 in the easy module. Updates mg_cade25.cfg with new default values for steps, cfg, denoise, buffer, guidance, and other parameters to improve quality and stability.
Add CWN and AGC guidance options to CADE25 nodes
DZRobo
e42ce45
Introduces Conditioning Weight Normalization (CWN) and Adaptive Guidance Clipping (AGC) to both easy and hard CADE25 modules, with configurable parameters exposed in the code and presets. Updates mg_cade25.cfg to include default values for these new options and adjusts some scheduler and denoise settings for improved stability and control.
Improve depth model auto-detection and fallback logic
DZRobo
42c372b
Enhanced depth model path resolution to scan additional directories and prefer local weights, including comfyui_controlnet_aux locations. Updated fallback to use a constant mid-gray tensor for pseudo-depth. Added warnings when no local weights are found and improved robustness of path detection. Fix memory leaking.
per‑step CFG scheduling
DZRobo
c86b5e0
Introduces per‑step CFG schedules (cosine, warmup, U‑shape) driven by log‑sigma progress; schedules stack with existing cfg_curve.
Presset finetune.
Add configurable CFG scheduling to guidance wrapper
DZRobo
7dd5fc1
Introduces per-step CFG scheduling options (cosine, warmup, U-shape) to _wrap_model_with_guidance and exposes related parameters in ComfyAdaptiveDetailEnhancer25. Updates mg_cade25.cfg presets to utilize new scheduling controls and adjusts detail and denoise settings for improved flexibility.
Add latent buffer stabilization to enhancer
DZRobo
df38cb0
Introduces a latent buffer mechanism to ComfyAdaptiveDetailEnhancer25 for improved multi-iteration stability, including EMA, anchor rebasing, and masked injection. Updates presets to enable and tune latent buffer parameters, and adjusts detail and mid-frequency settings for better results. Minor README clarification and workflow file update.
Speed optimisation
DZRobo
4bfcccd
Speed optimisation for CADE_Easy and SuperSImple workflow.
Add preview downscale and mid-frequency stabilizer
DZRobo
92ca926
Introduces a 'preview_downscale' option to cap output image size for reduced RAM usage in mg_cade25_easy.py. Adds mid-frequency stabilizer controls to mg_cade25.py for improved hand/object stability at high resolution. Updates README with new module notes and workflow documentation, and revises mg_Easy-Workflow.json for node order and structure.
Optimize memory usage
DZRobo
bb62070
Improves memory management by releasing temporary tensors throughout the CADE2.5 easy and hard modules, reducing RAM/VRAM peaks. Adds optional auto-save functionality for final images with configurable compression. Updates README with memory tips and clarifies workflow steps. Adjusts presets for sharpen, guidance, NAG, and mid-frequency stabilizer parameters to further optimize performance and output quality.
Add MG_CleanUp node and improve memory management
DZRobo
9068525
Introduces the MG_CleanUp node for aggressive RAM/VRAM cleanup and system cache trimming. Improves safe_encode and safe_decode in mg_cade25_easy.py to better release VRAM and avoid autograd graphs. Updates MG_ControlFusion nodes to clear caches and intermediates after execution. Adjusts blend and edge parameters in mg_controlfusion.cfg for better mask blending. Updates workflow to integrate new cleanup and parameter changes.
Bugfixes (#11)
DZRobo
Unverified signature
478379a
* Add MagicLatentAdapter node
Introduces MagicLatentAdapter for generating or adapting latents to match model expectations, including support for 5D (NCDHW) layouts and experimental FLUX/Qwen models. Adds mask alignment utilities to CADE nodes for robust mask blending, improves Gaussian blur to support 5D tensors, and updates preset defaults for improved results. Documentation and workflow updated to reflect new node and features.
* Bugfixes
1. Adjust edge mask gain logic and preset parameters. Fix depth path for CF (hard).
2. Memory leaking fix.
3. Quality improvement at all steps (Easy and SuperSimple)
Add MagicLatentAdapter node (experimental support FLUX/Qwen models) (#10)
DZRobo
Unverified signature
ccadebd
Introduces MagicLatentAdapter (experimental support FLUX/Qwen models) for generating or adapting latents to match model expectations. Adds mask alignment utilities to CADE nodes for robust mask blending, improves Gaussian blur to support 5D tensors, and updates preset defaults for improved results. Documentation and workflow updated to reflect new node and features.
BUG FIX and QoL. Fix cancel support to MG nodes (#6)
DZRobo
Unverified signature
dc64aaa
Implements cooperative cancellation and user interrupt propagation in mg_cade25_easy.py, mg_supersimple_easy.py, mg_cade25.py, and related modules. Adds interruptible sampler and callback wrappers, sentinel handling, and frequent cancel checks throughout long-running loops and workflows. Also improves LoRA selection robustness in mg_combinode.py by handling missing files gracefully. Updates workflow JSON to match new node logic.
Initial project structure and core files added (#1)
DZRobo
Unverified signature
695fbf0
Add MagicNodes repository with documentation, licensing, assets, workflows, presets, and main Python modules for Easy/Hard nodes, CADE 2.5, ControlFusion, QSilk, and supporting utilities. Includes sample images, negative LoRA, Depth Anything v2 vendor code, and setup scripts for ComfyUI integration.
