ComfyUI-ProportionChanger

Note

: This README was automatically generated using Claude Code AI-assisted development tools.

Advanced DWPose-based body proportion manipulation and keypoint processing for ComfyUI

This custom node is created by decomposing and copying the WanVideo UniAnimate DWPose Detector node from kijai's ComfyUI-WanVideoWrapper. The key difference is that instead of using image inputs, it now accepts DWPose KeyPoint data as input, enabling more flexible and detailed body proportion manipulation.

Transform body proportions, manipulate pose keypoints, and create more natural and consistent pose sequences through advanced proportion scaling algorithms.

Features

Core Nodes

• ProportionChanger DWPose Detector: Basic DWPose detection with keypoint output
• ProportionChanger Reference: Apply proportion changes from reference poses
• ProportionChanger DWPose Render: Pose visualization and rendering
• ProportionChanger Params: Interactive parameter control for pose adjustments
• ProportionChanger Interpolator: Frame interpolation for fluid motion
• pose_keypoint input: Utility for JSON to POSE_KEYPOINT conversion
• pose_keypoint preview: Utility for POSE_KEYPOINT to JSON conversion

Key Capabilities

• Proportion Manipulation: Scale body parts (head, torso, limbs) independently
• Reference-based Scaling: Apply proportions from reference poses to target sequences
• Format Conversion: Seamless conversion between DWPose and POSE_KEYPOINT formats
• Batch Processing: Handle multiple frames and pose sequences efficiently
• Interactive UI: Real-time parameter adjustment with pose preview

Installation

1. Clone this repository into your custom_nodes folder:

cd ComfyUI/custom_nodes
git clone https://github.com/[your-repo]/ComfyUI-ProportionChanger.git

2. Install dependencies:

cd ComfyUI-ProportionChanger
pip install -r requirements.txt

3. Restart ComfyUI

Dependencies

Required Models

The following models need to be downloaded and placed in the correct directories:

DWPose Models (place in unianimate/models/DWPose/):

• dw-ll_ucoco_384_bs5.torchscript.pt - Main pose detection model
• yolox_l.torchscript.pt - Object detection model

Download Links:

• Hugging Face Model Repository

Python Dependencies

See requirements.txt for complete list. Main dependencies include:

• torch
• torchvision
• opencv-python
• numpy

Usage

Basic Workflow

1. Load Image: Input your source image
2. DWPose Detection: Use ProportionChanger DWPose Detector to extract pose keypoints
3. Apply Proportions: Use ProportionChanger Reference with reference pose for transformations
4. Render Result: Use ProportionChanger DWPose Render to visualize the final pose

Advanced Features

Proportion Reference System

• Load a reference pose with desired proportions
• Apply those proportions to target images automatically
• Supports different canvas sizes with automatic scaling

Interactive Parameter Control

• ProportionChanger Params provides real-time adjustment
• Preview changes before applying to full sequences
• Fine-tune individual body part scaling

Node Reference

ProportionChanger DWPose Detector

Purpose: Basic pose detection with keypoint output Inputs: image, score_threshold Outputs: pose_keypoint

ProportionChanger Reference

Purpose: Apply reference proportions to target poses Inputs: pose_keypoint, reference_pose_keypoint Outputs: changed_pose_keypoint

ProportionChanger Params

Purpose: Interactive parameter adjustment for pose manipulation Inputs: pose_keypoint, various scaling parameters Outputs: adjusted_pose_keypoint

Technical Implementation

Core Algorithms

• DWPose Detection: Modified from WanVideo UniAnimate implementation
• Proportion Scaling: Mathematical transformation of keypoint coordinates
• Canvas Scaling: Automatic adjustment for different image dimensions

Data Formats

• Input: Standard image formats (PNG, JPG, etc.)
• Internal: POSE_KEYPOINT format with normalized coordinates (0-1 range)
• Metadata: Canvas dimensions for accurate coordinate transformation

Attribution and Credits

This project builds upon excellent work from the open-source community:

Primary Sources

• kijai/ComfyUI-WanVideoWrapper: Primary foundation for DWPose detection implementation
• toyxyz/ComfyUI-ultimate-openpose-editor: Source for ProportionChanger Params node functionality

Development Methodology

• Claude Code: All development performed using AI-assisted programming
• Vibe Coding: Iterative development with AI guidance
• Issue-driven Development: Systematic approach to feature implementation and bug fixes

Special Thanks

• kijai: For the robust WanVideoWrapper foundation and DWPose implementation
• toyxyz: For the innovative openpose editing framework
• Anthropic: For Claude Code development tools
• Open Source Community: For continuous inspiration and collaboration

License

This project is licensed under GPL 3.0 due to the combination of source materials from different licenses:

• Primary source: kijai/ComfyUI-WanVideoWrapper (Apache 2.0)
• Secondary source: toyxyz/ComfyUI-ultimate-openpose-editor (GPL 3.0)

When combining code from Apache 2.0 and GPL 3.0 licensed projects, the resulting derivative work must be distributed under GPL 3.0 according to license compatibility rules.

Copyright Notice:

• Original WanVideo UniAnimate DWPose Detector: Copyright by kijai
• ProportionChanger Params functionality: Copyright by toyxyz
• Modifications and integration: This project's contributors

See the LICENSE file for the full GPL 3.0 license terms.

Development Status

This project is actively developed with Claude AI assistance. Issues and feature requests are managed through AI-driven development cycles.

Recent Updates

• Reference-based proportion scaling system
• Batch processing capabilities
• Interactive parameter control
• Canvas size compatibility improvements

Contributing

While this project is primarily developed using Claude Code methodology, community feedback and issue reports are welcome. Please note that development follows an AI-assisted workflow.

Troubleshooting

Common Issues

1. Model Loading Errors: Ensure DWPose models are in correct directories
2. Canvas Size Mismatches: Check that canvas_width/canvas_height metadata is present
3. Coordinate Range Issues: Verify pose data uses normalized coordinates (0-1)

Debug Information

Enable debug logging in utils.py for detailed processing information.

Example Workflows

Check the example_workflows/ directory for sample ComfyUI workflows demonstrating various features and use cases.

---

Note: This is a specialized tool for pose manipulation and proportion adjustment. For general video generation, consider using the original WanVideoWrapper or native ComfyUI implementations.