Fast, repeatable ComfyUI setup for Windows 11 with CUDA 12.8. This repo installs a clean Conda environment, the right PyTorch build for your GPU, and a lightweight supervisor that launches ComfyUI headlessly and shuts it down cleanly when you close the window.
Highlights
- Windows 11 tested end-to-end
- Conda env: Python 3.12.11 (falls back to latest 3.12.x if needed)
- PyTorch CUDA 12.8 wheels (fallback to CUDA 12.6 if 12.8 is unavailable)
- Foreground launcher: Ctrl+C or closing the window kills all subprocesses
- Log-based watcher (no HTTP probes), gentler restarts, rotating logs
Status (2025-08-12)
ComfyUI Desktop exists, but this installer is still maintained for users who want to host or run ComfyUI as a service/foreground process on Windows. The scripts and CUDA 12.8/Python 3.12.11 stack are up to date and ready.
Make sure you have:
-
NVIDIA GPU drivers (latest) — https://www.nvidia.com/Download/index.aspx
-
Git — https://git-scm.com/downloads (Git LFS recommended)
-
Miniconda — https://docs.conda.io/en/latest/miniconda.html
After install, in an elevated Conda prompt run:
conda init --all --system
-
(Optional) Microsoft Visual C++ Build Tools — some nodes may need it: https://visualstudio.microsoft.com/visual-cpp-build-tools/
-
Clone or download this repository.
-
Run the installer:
install_update_comfyui.bat
What it does
- Creates/uses Conda env "ComfyUI" with Python 3.12.11
- Installs PyTorch + CUDA 12.8 (falls back to 12.6 if needed)
- Installs onnxruntime-gpu (falls back to CPU)
- Clones/updates ComfyUI into
%USERPROFILE%\ComfyUI
- Adds a few helpful custom nodes (see below)
- Verifies versions and cleans pip cache
Use the helper to fetch models defined in model_config.yaml
:
install_models.bat
model_config.yaml
example:
unet:
my_model:
repo_url: "https://huggingface.co/..."
include_files:
- example.safetensors
vae:
another:
repo_url: "https://huggingface.co/..."
include_folder: vae
Start ComfyUI:
launch_comfyui.bat
- Opens at: http://127.0.0.1:8188 (default)
- Foreground by default — close the window to stop everything
- Logs live under
logs/
(rotated)
Quiet output
-
Set
HEADLESS=1
before launching to suppress the environment summary prints (process still runs foreground):set HEADLESS=1 && launch_comfyui.bat
You don’t need a .env
file. Defaults work out of the box. If you want to customize paths/port/flags:
- HEADLESS: Quiet mode for launcher prints (1=quiet, 0=show env summary)
- COMFYUI_ENV_NAME: Conda environment name (default: ComfyUI)
Watcher tuning (advanced)
- ENABLE_NO_OUTPUT_RESTART=1
- NO_OUTPUT_RESTART_SECS=600
- QUIET_CPU_THRESHOLD=2.0
- QUIET_CPU_WINDOW_SECS=300
- BOOT_WAIT_TIME=1600
- MONITOR_INTERVAL=10
The installer clones/updates these by default:
- ComfyUI-Manager — https://github.com/ltdrdata/ComfyUI-Manager
- rgthree-comfy — https://github.com/rgthree/rgthree-comfy
- ComfyUI_ExtraModels — https://github.com/city96/ComfyUI_ExtraModels
- ComfyUI-GGUF — https://github.com/city96/ComfyUI-GGUF
You can add more by cloning into ComfyUI/custom_nodes/
.
If you prefer running as a background service, tools like NSSM work well. Point the service to launch_comfyui.bat
and set a working directory for logs. Note that Ctrl+C behavior doesn’t apply to services.
- Update: re-run
install_update_comfyui.bat
- Uninstall: remove
%USERPROFILE%\ComfyUI
and the Conda envComfyUI
(e.g.,conda remove -n ComfyUI --all
)
- Environment: Conda
ComfyUI
, Python 3.12.11 (or latest 3.12.x) - PyTorch: CUDA 12.8 wheels by default, CUDA 12.6 fallback
- Supervisor: no HTTP/TCP health checks; uses logs + CPU quiet periods
- Clean shutdown: Ctrl+C or closing the console terminates the whole process tree
MIT. See LICENSE.
- ComfyUI by https://github.com/comfyanonymous/ComfyUI
- The authors of the included custom nodes