ℹ️ Note:

I have quantized the model to Q4km and Q3kl. The results are good, but this model takes a long time to process. I will try to prepare a good workflow soon.

workflow(3 step)

https://civarchive.com/models/1620800

🌸 WAN 2.1 VACE: Next-Gen Video Diffusion Model

🎬 Explore the Power of WAN 2.1 VACE for Video Generation! 🎬

📌 Overview

WAN 2.1 VACE is a state-of-the-art video diffusion model designed for high-quality, efficient video generation.

🔑 Key Components

1. Diffusion Model: wan2.1_vace_14B_fp16.safetensors (Q4km.gguf or Q3kl.gguf )

2. VAE: wan_2.1_vae.safetensors

3. Text Encoder: Choose one:

   • umt5_xxl_fp16.safetensors (recommended for compatibility)

   • umt5_xxl_fp8_e4m3fn_scaled.safetensors

📁 File Organization

Place the downloaded files in the following structure within your ComfyUI directory:

ComfyUI/
├── models/
│   ├── diffusion_models/
│   │   └── wan2.1_vace_14B_fp16.safetensors  (Q4km or Q3kl here)
│   ├── text_encoders/
│   │   └── umt5_xxl_fp16.safetensors  # or umt5_xxl_fp8_e4m3fn_scaled.safetensors
│   └── vae/
│       └── wan_2.1_vae.safetensors

🛠️ Step-by-Step Tutorial

• English Tutorial

• 中文教程

Follow the official tutorial for detailed setup and usage instructions. The guide covers everything from installation to advanced workflow tips.

💡 workflow

https://civarchive.com/models/1620800

👨‍💻 Developer Information

Those models was Quantized by Abdallah Al-Swaiti:

1. Hugging Face

2. GitHub

3. LinkedIn

4. ComfyUI-OllamaGemini

For additional tools and updates, check out my other repositories.

✨ Create Dreamy Videos with WAN 2.1 VACE and Pastel Dream! ✨