1. Download comfyui and unzip to a suitable location.
(comfyui0.3.49, pytorch2.7, python3.12, cuda12.8)
goto ComfyUI_windows_portable\update and Run update_comfyui_stable.bat for update.
2. Download library
unzip two folders into YOUR-FOLDER-PATH\ComfyUI_windows_portable\python_embeded\
3. Install Triton windows.
Open Command Prompt and run.
YOUR-FOLDER-PATH\ComfyUI_windows_portable\python_embeded\python.exe -m pip install triton-windows<3.4
(Example. D:\AI\ComfyUI_windows_portable\python_embeded\python.exe -m pip install triton-windows<3.4)
4. Download sageattention
Install sageattention (Run from the downloaded location)
YOUR-FOLDER-PATH\ComfyUI_windows_portable\python_embeded\python.exe -m pip install sageattention-2.2.0+cu128torch2.7.1.post2-cp39-abi3-win_amd64.whl
(Example. D:\AI\ComfyUI_windows_portable\python_embeded\python.exe -m pip install sageattention-2.2.0+cu128torch2.7.1.post2-cp39-abi3-win_amd64.whl)
5. goto ComfyUI_windows_portable and Edit run_nvidia_gpu.bat
add --use-sage-attention after --windows-standalone-build
(Example. .\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build --use-sage-attention
pause)
6. Run run_nvidia_gpu.bat
Check the "Using sage attention" statement and make sure there are no errors.
WAN 2.2 basic GGUF 720p workflow - v1.0 | Wan Video Workflows | Civitai
Using this workflow, took 185 seconds to generate a wan2.2 14B T2V.
(after all loading was complete, RTX4070S 12G VRAM, 32G RAM, 720x1280x81)