The PyTorch build that fits your GPU
ComfyPilot detects your graphics card and driver and picks the stack: CUDA 13 on NVIDIA RTX 20-series and newer, ROCm on AMD, XPU on Intel Arc. You can still choose another one.
Free and open source · Linux and Windows
Install, run and keep ComfyUI up to date, the easy way.
ComfyPilot sets up ComfyUI with the right PyTorch build for your graphics card, keeps several installs side by side, and gets you the models, LoRAs and workflows you need. No terminal, no guessing which CUDA wheel goes with which driver.
Version 1.0.0 · updates itself after that
- NVIDIACUDA 13 · RTX 20-series and newer
- AMDROCm
- Intel ArcXPU
Features
Everything between you and your first image
ComfyPilot handles the parts of ComfyUI that usually mean a terminal, a forum thread and a broken environment.
Safe updates
The PyTorch stack is pinned per install, so updating ComfyUI never swaps it out, and a snapshot of your custom nodes is taken before every update.
Add-ons in one click
SageAttention, FlashAttention, Nunchaku, InsightFace, Trellis2 and the built-in ComfyUI-Manager, with the compatibility rules handled for you.
As many installs as you like
Each with its own version, stack and status at a glance in the sidebar. Try a new ComfyUI without touching the one that works.
Run it from anywhere
Start, stop and restart from the app or the system tray. ComfyUI's output shows in a built-in console with filters for errors and warnings.
Launch options that make sense
Options follow ComfyUI's current memory model, and only the ones your ComfyUI version understands are shown.
Your tokens stay in your keychain
Hugging Face and Civitai tokens are kept in your system keychain, with step-by-step help to create them. Bug reports from the built-in diagnostics never include token values.
Updates itself
ComfyPilot tells you when a new version is out and installs it: the Windows installer, the right .deb, .rpm or Arch package, or the AppImage in place.
A quick tour
From a fresh install to a working workflow
01
A new install, set up for your GPU
Pick a folder, and ComfyPilot recommends the PyTorch build for the card it finds. Choose add-ons and popular custom nodes, installed at versions known to work together.
- Optional shared models folder across installs
- Checks for missing system tools and installs them after asking
- Windows: per-user, no administrator rights needed
02
Models that fit your VRAM
A curated catalog of FLUX, WAN, LTX and more, with the variants that fit your graphics card and RAM. Tick what you want and it lands in the right folders.
- Large Hugging Face files download fast via Xet
- Gated models with your Hugging Face token
03
Browse LoRAs on Civitai
Search, sort and filter Civitai's LoRAs by base model and see previews, with videos playing in the detail view. Download straight into the right folder of your ComfyUI.
04
Open a workflow, get what it needs
Open any workflow and see which models and custom nodes your install is missing, then download or install them. Works with workflows from anywhere.
- Checks workflows saved in ComfyUI or any .json file
- Shows the folder each missing model belongs in
05
Find workflows on Civitai
Browse ComfyUI workflows shared on Civitai, download one, and ComfyPilot checks what it needs before you open it in ComfyUI.
Download
Get ComfyPilot v1.0.0
Free and open source. Pick the file for your system; after that, ComfyPilot keeps itself up to date.
Windows
Windows 10 (1809) or 11
Run it. Installs for your user, no administrator rights needed.
Debian · Ubuntu
Debian 13+, Ubuntu 25.04+ and derivatives
sudo apt install ./comfypilot_*.debFedora
Fedora 43+
sudo dnf install ./comfypilot-*.rpmArch Linux
CachyOS, EndeavourOS, Manjaro…
sudo pacman -U comfypilot-*.pkg.tar.zstAny other Linux · AppImage
Ubuntu 22.04+, Mint, openSUSE, Bazzite, Silverblue…
chmod +x ComfyPilot-*.AppImageThen run it. Gear Lever adds it to your app menu. It updates itself in place, without asking for a password.
What you need: a 64-bit system and, for image and video generation, an NVIDIA (RTX 20-series or newer recommended), AMD or Intel Arc graphics card with a current driver.
FAQ
Questions
Is ComfyPilot free?
Yes. ComfyPilot is free and open source under the Apache-2.0 license. The source is on GitHub.
Does it work on macOS?
Not at the moment. ComfyPilot runs on Windows 10 (1809) and 11, and on 64-bit Linux.
Will updating ComfyUI break my PyTorch or custom nodes?
The PyTorch stack is pinned per install, so a ComfyUI update can't swap it out. Built-in custom nodes are pinned to tested versions, and a snapshot of your custom nodes is taken before every update.
Can I keep my existing ComfyUI?
Yes. You can run several installs side by side, each with its own version and PyTorch stack, and share one models folder between them if you like.
Something isn't working. How do I report it?
Click the clipboard button at the bottom of the left bar (Copy diagnostics for a bug report). It copies the app and system versions, your GPU, the selected install's settings and the last 150 lines of the log; token values are never included. Then open a bug report and paste it in.
