Files
Ganaraj PermundaandClaude Sonnet 4.5 5fd336ed0a Initial commit: ComfyUI TurboDiffusion custom nodes
This is the current working implementation with custom model loader and dequantization.

Features:
- TurboWanModelLoader: Custom loader for quantized .pth models with block-wise int8 dequantization
- TurboWanSampler: I2V conditioning preparation based on WanImageToVideo pattern
- TurboDiffusionSaveVideo: Video export to MP4/GIF/WebM formats
- Complete workflow using ComfyUI-native nodes (UNETLoader, CLIPLoader, VAELoader)

Model support:
- TurboWan2.2-I2V-A14B quantized models (high/low noise)
- nsfw_wan_umt5-xxl_fp8_scaled text encoder
- wan_2.1_vae VAE

Implementation uses custom dequantization logic for int8 block-wise quantization.
Next iteration may wrap official TurboDiffusion inference code instead.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
2025-12-16 17:08:43 +05:30

2 lines
5 B
Plaintext