Easy Fresh ComfyUI Install on Windows 11 with Flash-Attention 2 & SageAttention (Low VRAM Boost!) скачать в хорошем качестве

Easy Fresh ComfyUI Install on Windows 11 with Flash-Attention 2 & SageAttention (Low VRAM Boost!)

Step 1: Prepare Prerequisites *Install NVIDIA CUDA Toolkit* Download and install the latest version: https://developer.nvidia.com/cuda-dow... *Download and add FFmpeg to PATH* Download the latest FFmpeg build: https://github.com/BtbN/FFmpeg-Builds... Unzip the downloaded file. Copy the full path to the `bin` folder inside the unzipped directory. Open *System Properties* → *Advanced* → **Environment Variables**. Under *User variables**, select **Path* → *Edit* → **New**. Paste the `bin` path → *OK* to save and exit. *Install Git* (recommended; skip if you prefer manually downloading ZIP files) https://git-scm.com/download/win Keeping ComfyUI updated with Git is the easiest way to get first-day support for the latest models. You can update via the ComfyUI Manager or by running `git pull` in the ComfyUI directory. Step 2: Install ComfyUI with Miniconda 1. *Clone the repositories* Open a terminal (Command Prompt or PowerShell) and run: ``` git clone https://github.com/comfyanonymous/Com... cd ComfyUI git clone https://github.com/Comfy-Org/ComfyUI-... custom_nodes/ComfyUI-Manager ``` 2. *Install Miniconda* (if not already installed) Download and install from: https://repo.anaconda.com/miniconda/ 3. *Create and activate a Conda environment* Open *Anaconda PowerShell Prompt (Miniconda3)* from the Start menu. Create a new environment (replace `comfyui` with your preferred name): ``` conda create -n comfyui python=3.12 ``` When prompted, accept the terms and let it finish creating. Activate the environment: ``` conda activate comfyui ``` 4. *Install ComfyUI requirements* Navigate to your ComfyUI folder: ``` cd path\to\your\ComfyUI ``` Install dependencies: ``` pip install -r requirements.txt ``` 5. *Launch ComfyUI* ``` python main.py ``` If you encounter PyTorch-related errors, follow the reinstall steps shown in the video (using `--force-reinstall` for PyTorch). Step 3: Install Performance Boosters (Flash-Attention, SageAttention, Triton) These optional wheels (Flash-Attention, SageAttention, and Triton for Windows) significantly improve performance and allow you to run demanding workflows like: Qwen-Image-Edit in BF16 WAN2.2 in FP16 even on GPUs with 10 GB VRAM (when paired with 64 GB system RAM). *Flash-Attention 2 for Windows* Download the appropriate wheel for your setup: https://huggingface.co/ussoewwin/Flas... *SageAttention (precompiled wheels by Kijai)* Download the matching wheel: https://huggingface.co/Kijai/Precompi... After downloading the `.whl` files, install them in your activated Conda environment: ``` pip install path\to\downloaded-wheel.whl ``` Restart ComfyUI after installation to apply the changes. Enjoy the extra speed and reduced VRAM usage!