Hi folks,
in this video, I’ll show you how to get Triton and Sage Attention running in ComfyUI for massive speed gains—up to 50% faster for some workflows. Perfect if you’re working with Wan Video, Flux, or even SDXL.
This video will be paramount when I release my Wan 2.2 Video in a couple of days. Without Sage Attention, you’ll need much longer to generate videos. So consider this as Part I of a 2-Part Wan 2.2 Video Tutorial.
One Click installer for my Patrons:🎁
https://www.patreon.com/posts/edit-31-08-2025-135625892
📌 Links from the video & what they do:
Python 3.12 – Required for running ComfyUI and installing Sage Attention. We use Python 3.12 specifically to ensure compatibility with PyTorch 2.8 and the latest ComfyUI builds.
NVIDIA Drivers – The latest drivers are essential to unlock Triton and Sage Attention performance gains. Older drivers can cause compatibility issues.
Git – Needed to download ComfyUI and related repositories directly from GitHub using the git clone command.
ComfyUI Repo – The core application for building AI image and video workflows. We clone this as the main foundation.
Sage Attention – An optimized attention mechanism that can drastically reduce generation times for both images and videos.
ComfyUI Manager – A handy extension that adds package management and workflow installation tools directly inside ComfyUI.
KJ Nodes – A set of custom nodes for ComfyUI, including models with Sage Attention built in, making integration seamless.
💻 PyTorch 2.8 Install Command (CUDA 12.8):
python -m pip install torch==2.8.0+cu128 torchvision==0.23.0+cu128 torchaudio==2.8.0+cu128 --index-url https://download.pytorch.org/whl/cu128