README.md

metadata

title: Wan2.1 T2I GGUF
emoji: 🎨
colorFrom: blue
colorTo: indigo
sdk: gradio
sdk_version: 5.9.1
app_file: app.py
pinned: false
license: mit

Wan2.1 Text-to-Image (ComfyUI GGUF Backend)

This Space runs Wan2.1 T2I using ComfyUI with GGUF quantization for efficiency.

🚀 Features

• Model: Wan2.2-T2V-A14B-LowNoise-Q3_K_S.gguf (Quantized for lower VRAM usage)
• Encoder: umt5-xxl-encoder-Q3_K_S.gguf
• LoRA: Wan2.1_T2V_14B_FusionX_LoRA
• Backend: ComfyUI (Headless) + Gradio UI

⚠️ Hardware Requirements

This model requires significant RAM/VRAM even in quantized form.

• Recommended: GPU Space (T4 or A10G).
• Minimum: 16GB+ System RAM (if running on CPU, which will be very slow).

📂 Architecture

The app.py script automatically:

1. Clones ComfyUI and ComfyUI-GGUF if missing.
2. Downloads the necessary models (Unet, CLIP, VAE, LoRA) from Hugging Face using aria2 or python fallback.
3. Loads the ComfyUI nodes and serves a Gradio Interface.

🔗 Credits

• ComfyUI
• ComfyUI-GGUF
• Models by QuantStack, City96, Comfy-Org, and vrgamedevgirl84.