1. Add Random Prompter Geek Node, more convenient random control.
2. Add ABC and non-ABC version of Random Prompter Nodes switch to config.ini.
3. Add Random Prompter config to config.ini.
4. Add more choices for Random Prompter Nodes with more refined and accurate categories
1. Fix Make Image Grid `from torchvision import transforms as TF` -> `import torchvision.transforms.functional as TF` issue.
2. Upgrade Split Image Grid node:
① Reasonable default values: Ensure at least 1 row/column to avoid division by zero errors.
② Better error messages: Provide specific resolution suggestions.
③ Boundary checks: Ensure cropping does not exceed the image area.
④ Empty list protection: Ensure a valid image tensor is always returned.
⑤ Exception handling: Gracefully handle errors in crop mode.
3. Separated the `install.py` file for ComfyUI and the `install_manual.py` file for the `install.bat` file. It should work fine now.
4. Zip IPAdapterPlus.py to IPAdapterPlus.zip to avoid node-conflict attention in ComfyUI Manager.
5. Random Prompter supports both string dictionaries, lists or txt format for YAML files.
6. Add more choices to Random Prompter.
1. Update model sheet workflow.
2. Add Hunyuan Prompt Enhancer and fal LLM nodes to portrait master + random prompter workflow.
3. Add EXCLUSION_MARK parameter in the Random Prompter to exclude files in random mode.
4. Add more choices for to Random Prompter data.
The functools module is part of Python’s standard library and must not be listed in requirements.txt. The PyPI package with the same name is outdated, unmaintained, and fails to build on modern Python environments, breaking the installation process.
This PR removes the erroneous functools entry from requirements.txt to ensure smooth dependency installation.
✅ Changes:
Removed functools from requirements.txt
🔧 Testing:
Verified successful installation of dependencies in Python 3.10/3.11
Confirmed node functionality remains intact (no third-party functools was used)
Note: Standard library usage (e.g., from functools import lru_cache) continues to work without any changes.
1. Re-write all nodes:
2. Code Organization: Nodes are logically grouped by data type and functionality.
3. Modular nodes loading control, including deprecated nodes.
4. Comments: All key functions and classes have detailed.
5. Documentation: Each node class has an description and tooltip.
6. Type Hints: Type hints have been added to all functions.
7. Naming Conventions: Naming conventions for return values have been standardized.
8. Error Handling: Necessary imports and error handling have been added.
9. Add Random Prompter JK🐉 node, deprecate Flux Prompt Generator.
10. Add QWen_ModelSheet and QWen_ModelSheet adv imgen/imgedit workflow.
11. Add Portrait Master + Random Prompter imgen workflow.
1. Update Hunyuan3D img2mesh workflow: ① supports 4K even 8K texture generation; ② Resize the input image to a valid resolution.
2. Create Hunyuan3D2.1 one-click version img2mesh workflow for runninghub.
3. Deprecate Hy3D Cam Config 20to21 JK🐉 node.
4. Fixed INT and Float min | max value for Math nodes.
1. Fix `comfy_extras.nodes_compositing.SplitImageWithAlpha.split_image_with_alpha()` naming issue. Rename it to `comfy_extras.nodes_compositing.SplitImageWithAlpha.excute()`.
1. Add Native Wan HuMo workflows.
2. Add Wan Video "Cuts" series workflows (ff | fun | HuMo), supports multiple video generation based on scene and audio cuts.
1. Add HuMo t2v | ref2v + a2v | infinitetalk workflows.
2. Add a description about the fp8_e5m2 model, which is available for Triton.
3. Add Audio Cuts JK🐉 | Cut Audio JK🐉 | Cut Audio Index JK🐉 | Cut Audio Cuts JK🐉 | Cut Audio Loop JK🐉 | Get Nth String JK🐉 | Wan Frame Count JK🐉 nodes.
4. Add first two of Wan Video "Cuts" series workflows (VACE | Talk), supports multiple video generation based on scene and audio cuts.
Jake Chai