Wan 2.2 TI2V FP8

# Wan2.2-TI2V for ComfyUI This repository contains the converted [Wan2.2-TI2V](https://huggingface.co/Wan-AI/Wan2.2-TI2V-5B-Diffusers) model, adapted for **ComfyUI** usage. ## ✅ Model Components | Module | Format | File | |----------------|----------------|------| | Transformer | FP8 + scale | `wan2.2_ti2v_transformer_fp8_scaled.safetensors` | | VAE | FP16 | `wan2.2_ti2v_vae_fp16.safetensors` | | Text Encoder | BF16 (shared) | `wan2.2_ti2v_text_encoder_bf16.safetensors` | ## 📦 How to Use in ComfyUI Place the files in your ComfyUI model paths: - **Transformer** → `models/diffusers/` - **VAE** → `models/vae/` - **Text Encoder** → `models/clip/` (or appropriate location) ## 🛠️ Notes - FP8 conversion used Tencent HunyuanVideo's `fp8_optimization.py` - Safe export with `safetensors` - Includes special handling of shared tensors (`shared.weight`) ## 🔗 Credits - Original model: [Wan2.2-TI2V](https://huggingface.co/Wan-AI/Wan2.2-TI2V-5B-Diffusers) - FP8 optimization: [Tencent Hunyuan](https://github.com/Tencent-Hunyuan/HunyuanVideo)

Source context: 0 downloads · 2 likes · Repo Goldlionren00/Wan2.2-TI2V-FP8-ComfyUI