gdhe17/Self-Forcing & Wan-AI/Wan2.1-VACE-1.3B Scopes Addon Experiment
⚠️ Notice:
This project is intended for experimental use only.CausVid/Self-Forcing is not natively supported in ComfyUI, proper sampling is not implemented.
This is an addon experiment of gdhe17/Self-Forcing and Wan-AI/Wan2.1-VACE-1.3B scopes.
The process involved extracting VACE scopes and injecting into the target models, using scripts provided by wsbagnsv1.
The FP16 model weights were then quantized to specific FP8 formats (E4M3FN and E5M2) using ComfyUI custom node ComfyUI-ModelQuantizer by lum3on.
LoRA extraction was done by subtracting and saving the weight differences using the ModelMergeSubtract node and the Extract and Save LoRA node in ComfyUI.
Usage
The model files can be used in ComfyUI with the WanVaceToVideo node. Place the required model(s) in the following folders:
| Type | Name | Location | Download |
|---|---|---|---|
| Main Model | Wan2.1-T2V-1.3B-Self-Forcing-VACE | ComfyUI/models/diffusion_models |
Safetensors (this repo) |
| Text Encoder | umt5-xxl-encoder | ComfyUI/models/text_encoders |
Safetensors / GGUF |
| VAE | Wan2_1_VAE_bf16 | ComfyUI/models/vae |
Safetensors |
Notes
All original licenses and restrictions from the base models still apply.
Reference
- For an overview of Safetensors format, please see the Safetensors format.