(NEW) One-Click Triton & SageAttention installation guide: https://civarchive.com/articles/12248

V2.2 UPDATED (PENDING)

This update was tested with 8GB VRAM, removes the confusion between i2v and t2v, adds convenient resolution settings (thanks @redhome) and several other handy features. It also fixes a critical bug in shift order and adds perfectly support for WAN Lora.

Recommended Version

• ComfyUI v0.3.19 or higher

  Wan does not work on versions below that.

• ComfyUI-Manager v3.26 or higher

  There is an issue with missing nodes and conflict in previous versions

Require Nodes (Works on Install Missing Nodes)

• ComfyUI-KJNodes (need to update. Switch Version -> latest(nightly), doesn't work 1.0.6 version or lower)

• ComfyUI-GGUF (If you have other gguf custom nodes, remove them)

• ComfyUI-GIMM-VFI (need to installed cuda toolkit 12.x)

• ComfyUI-Easy-Use

• rgthree-comfy

• ComfyUI-Florence2

• ComfyUI-Impact-Pack

• ComfyUI-Custom-Scripts

• ComfyUI-VideoHelperSuite

• ComfyUI-Frame-Interpolation

• ComfyUI-HunyuanVideoMultiLora

v2 Update Notes

• Applied TeaCache

• Added Florence2 auto-captioning

• Added GIMM VFI

• Added LoRA Loader

• Introduced shift parameter

• Optimized workflow

Require Models

T2V Quant Model

🔗 Download
📂 Save to: models/diffusion_models

or

T2V Fastest Model (Works low vram)

🔗 Download
📂 Save to: models/diffusion_models

I2V Quant Model

🔗 Download
📂 Save to: models/diffusion_models

CLIP DO NOT USE ANOTHER CLIP

🔗 Download
📂 Save to: models/clip

CLIP-VISION

🔗 Download
📂 Save to: models/clip_vision

VAE

🔗 Download
📂 Save to: models/vae

UPSCALE MODEL

🛠 Recommended Model: 4x_foolhardy_Remacri
download via ComfyUI Model Manager