This project provides ComfyUI nodes for video generation using the Wan2.1 model, specifically enabling control over start and end frames for guided video creation. It targets users of ComfyUI looking to enhance their video generation workflows with frame-based conditioning, offering improved control over video narrative and character consistency.

How It Works

The nodes are a modification of Kijai's existing nodes, integrating start and end frame guidance directly into the video generation process. This approach allows for more precise control over the temporal evolution of generated videos by conditioning on specific initial and final frames, aiming to improve character identity consistency, especially with frame counts of 25 or higher.

Quick Start & Requirements

• Install by following instructions for ComfyUI-WanVideoWrapper.
• Requires Wan2.1 models (720P and 480P).
• Recommended frame count of 25+.
• Positive prompts are essential to avoid severe video distortion.
• Refer to ComfyUI-WanVideoWrapper for model download and installation.

Highlighted Details

• Supports Wan2.1 models in 720P and 480P.
• Includes start and end frame weight controls for video transitions.
• Compatible with SLG functionality in KJ's nodes.
• Positive prompts are crucial for stable video output.

Maintenance & Community

• Project is a modification of Kijai's nodes; gratitude extended to Kijai.
• Discussions are welcomed in the issues section.

Licensing & Compatibility

• License details are not explicitly stated in the README.
• Compatibility with commercial or closed-source projects is not specified.

Limitations & Caveats

The start and end frame video generation approach is in its early stages, with future work planned for model/LoRA fine-tuning. Incorporating end frame guidance in Image-to-Video (I2V) currently degrades quality. Positive prompts are mandatory for acceptable video generation.