ComfyUI runs well on Apple Silicon Macs. You can generate images with SDXL, Flux and other models on an M1, M2, M3, M4 or newer chip, using the Mac’s GPU through Apple’s Metal (MPS) backend instead of an NVIDIA card.
There are two ways to install it. Comfy Desktop is a normal Mac app you drag into Applications, and it handles Python and PyTorch for you. The manual install uses Git, Python and a terminal, and gives you full control.
This guide walks through both, shows where models and outputs live on a Mac, how to update and fully uninstall, and how to fix the problems Mac users hit most. It follows the official Comfy docs as of November 2026.
- Needs: an Apple Silicon Mac (M1 or later) on macOS 13 Ventura or newer. Intel Macs aren’t supported by Comfy Desktop.
- Easiest: download Comfy Desktop from comfy.org/download, drag it to Applications, create an instance.
- Space: at least 4.85 GB per install, plus your models.
- RAM: 8 GB minimum, 16 GB recommended. More unified memory means bigger models.
- Manual route: conda env, PyTorch nightly for Mac,
git clone,pip install -r requirements.txt,python main.py. - Models: Desktop shares them from
~/ComfyUI-Shared.
Mac Requirements
| Item | Requirement |
|---|---|
| Chip | Apple Silicon, M1 or later |
| macOS | macOS 13 (Ventura) or later |
| Memory | 8 GB minimum, 16 GB recommended |
| Disk | At least 4.85 GB per installation, plus models (often 2 to 25 GB each) |
| Permissions | Admin rights for the first install |
| Browser (manual install) | A recent Chrome is recommended by Comfy |
Macs use unified memory, so the GPU shares RAM with the system. A 16 GB Mac can run SDXL comfortably; for large models like full Flux, 32 GB or more makes life much easier. If you’re buying hardware for AI, compare with our RTX 5070 for AI guide.
Desktop or Manual Install?
| Comfy Desktop | Manual install | |
|---|---|---|
| Setup | Drag and drop app, guided wizard | Terminal, Git, conda, Python |
| Python and PyTorch | Installed for you | You install them |
| Updates | Automatic for the app; one click for ComfyUI | git pull plus requirements |
| Multiple installs | Built-in instances with shared models | Separate folders by hand |
| Headless or server use | Not supported | Yes |
| Best for | Most people | Developers, servers, custom setups |
Install Comfy Desktop on Mac
- Download the .dmg from
comfy.org/download. The page detects your Mac. - Open the .dmg and drag Comfy Desktop into Applications.
- Launch it from Launchpad, Spotlight or the Applications folder.
- If macOS blocks it, open System Settings > Privacy & Security and click Open Anyway.
Already have an older ComfyUI install? Desktop can detect and migrate it, carrying over custom nodes, workflows, models and settings. Migration reuses your existing folder and Python environment in place, so make your own backup first if you want an untouched copy.
Create Your First Instance
Desktop runs ComfyUI as “instances”, separate installs that can share one model library.
- Click the + card in the instance chooser.
- Name it (the default is ComfyUI).
- Choose Public Builds, the recommended install method.
- Keep Release on Stable (recommended).
- Check Hardware. Desktop detects your Mac and preselects the right option, and shows the ComfyUI version, Python version and download size.
- Pick an install location or keep the default. If space is low, you’ll get a warning with the option to choose another drive.
- Wait for setup. It downloads, unpacks and builds the environment. Comfy warns the final update step can take an hour or more on a cold machine.
When it finishes, you can let starter templates download models in the background or choose Skip & open ComfyUI.
If you need a PyPI mirror (for example on a slow or restricted network), Desktop Settings > Advanced has a PyPI Mirror URL field.
Where Files Live on a Mac
| What | Path (Comfy Desktop) |
|---|---|
| App data | ~/Library/Application Support/Comfy Desktop |
| Logs | ~/Library/Logs/Comfy Desktop |
| Instances (installs) | ~/ComfyUI-Installs |
| Shared models, input, output | ~/ComfyUI-Shared |
Put checkpoints, LoRAs, VAEs and other models in the matching subfolders of the shared models folder so every instance can use them. You can add more model folders, such as an external SSD, in Desktop Settings > Storage with + Add Shared Directory. Custom nodes are installed per instance.
Not sure how to grab models? See how to download models from Hugging Face.
Manual Install With Terminal
Comfy recommends Python 3.13 (3.12 is the fallback) and a Miniconda environment.
- Install Miniconda for Apple Silicon, then create an environment:
conda create -n comfyenv python=3.13 conda activate comfyenv - Clone ComfyUI:
git clone https://github.com/Comfy-Org/ComfyUI.git - Install PyTorch for Mac (the docs use the nightly build with conda):
conda install pytorch-nightly::pytorch torchvision torchaudio -c pytorch-nightly - Check the GPU backend:
python -c "import torch; print(torch.backends.mps.is_available())"It should print
True. - Install ComfyUI’s requirements:
cd ComfyUI pip install -r requirements.txt - Start it:
python main.pyThen open
http://127.0.0.1:8188in your browser.
Stop the server with Ctrl + C in Terminal. Run python main.py --help to see every startup flag. Models go in ComfyUI/models/ (checkpoints, loras, vae and so on). To reuse models from another folder, copy extra_model_paths.yaml.example to extra_model_paths.yaml and add your paths.
Your First Image
- Open the templates browser and pick a basic text-to-image workflow.
- Download the model it asks for, or drop a checkpoint into your models folder and press
Rto refresh. - Type a prompt in the positive prompt box.
- Press Cmd + Enter to queue it.
New to the interface? Our ComfyUI guide covers nodes and workflows, and the ComfyUI keyboard shortcuts cheat sheet lists the Mac keys. Ready-made graphs are in our ComfyUI workflows collection.
What to Expect on Apple Silicon
- Slower than a big NVIDIA GPU for the same model, but very usable for SD 1.5 and SDXL.
- Memory is the limit. Unified memory is shared with macOS and your apps, so close heavy apps before big jobs.
- Smaller models win. Pick SDXL-class or distilled/turbo models and quantized versions of large models where available.
- Lower steps and resolution first, then scale up once a workflow works.
- Some custom nodes are CUDA-only. Nodes that depend on NVIDIA-specific libraries may not run on a Mac.
More speed tips that also apply on a Mac: how to speed up ComfyUI.
Update ComfyUI on Mac
| Install | How to update |
|---|---|
| Desktop app | Checks automatically; Settings > Updates > Check for updates, then Restart & Update |
| Desktop instance | Open the instance’s Update tab; choose Stable or Latest on GitHub |
| Manual install | cd ComfyUI, git pull, then pip install -r requirements.txt in your env |
Uninstall Completely
- Delete instances you don’t need from inside Desktop (Uninstall deletes files; Forget keeps them).
- Move the app to Trash: Applications > right-click Comfy Desktop > Move to Trash.
- Remove leftover data if you want a clean slate (this is permanent):
~/Library/Application Support/Comfy Desktop~/Library/Logs/Comfy Desktop~/ComfyUI-Installs~/ComfyUI-Shared(your models and outputs)
For a manual install, delete the ComfyUI folder and remove the environment with conda env remove -n comfyenv.
Troubleshooting
| Problem | Likely cause | Fix |
|---|---|---|
| “App can’t be opened” | Gatekeeper block | System Settings > Privacy & Security > Open Anyway |
| Installer not offered on Intel Mac | Desktop needs Apple Silicon | Use an M-series Mac, or a cloud GPU |
| Generation runs on CPU and is very slow | MPS not available | Run the MPS check; reinstall PyTorch for Mac in your env |
| Out of memory or Mac freezes | Model too big for unified memory | Lower resolution or batch size, close apps, use a smaller or quantized model |
| Black or blank images | Precision or VAE issue | See our ComfyUI black image fix |
| A custom node fails to import | CUDA-only dependency | Check the node’s README for Mac support or use an alternative |
| Low disk space warning | Install or models filling the drive | Choose another install location, move models to an external SSD via Storage |
| Need logs | Diagnosing a crash | Settings > Logs > Open logs folder, or ~/Library/Logs/Comfy Desktop |
FAQ
Can ComfyUI run on a Mac?
Yes. It runs on Apple Silicon Macs (M1 or later) with macOS 13 or newer, using the Metal (MPS) GPU backend.
Does ComfyUI work on Intel Macs?
Comfy Desktop lists Apple Silicon only. An Intel Mac would be limited to slow CPU generation, so a cloud GPU is a better option.
How much RAM do I need for ComfyUI on a Mac?
Comfy lists 8 GB minimum and 16 GB recommended. Larger models like Flux are far more comfortable with 32 GB or more.
Where do I put models on a Mac?
With Comfy Desktop, use the shared folder at ~/ComfyUI-Shared. With a manual install, use the ComfyUI/models subfolders.
Is Comfy Desktop free?
Yes. ComfyUI and Comfy Desktop are free to download and run locally.
Can I run ComfyUI headless on a Mac?
Not with Comfy Desktop. Use the manual install and start it with python main.py.
Which port does ComfyUI use?
8188 by default. Desktop picks the next free port if 8188 is busy.




