ComfyUI on Mac: how do you install and run it?
Comfy Desktop supports macOS 13 or later on Apple Silicon. Manual installation uses the official PyTorch nightly channel, ComfyUI dependencies and python main.py. Apple Silicon acceleration uses MPS, not MLX.
For the simplest official route, use Comfy Desktop. Choose manual installation when you need to manage the Python environment yourself.
Get the local AI newsletter for your Mac
One practical Thursday email with new local AI models, memory requirements and Mac setup guides.
For: MacBook Air M5 16 GB
You'll also get the weekly ModelFit email. By signing up you agree to our Privacy Policy. Unsubscribe from either email anytime, in one click.
How do you install Comfy Desktop?
Download Comfy Desktop from the official site and follow its macOS installation guide. The stated requirements are macOS 13 Ventura or later and Apple Silicon M1 or later.
The documented 4.85 GB recommendation is disk space per installation. It is not a RAM requirement or a model-memory budget.
How do you install ComfyUI manually?
Use the official Mac ARM PyTorch command inside your chosen environment, install ComfyUI's dependencies from its directory, then start the server. The current requirements page recommends Python 3.13, with Python 3.12 as a fallback for custom-node dependency trouble.
Install PyTorch for Mac ARM from the official nightly channel
conda install pytorch-nightly::pytorch torchvision torchaudio -c pytorch-nightlyInstall ComfyUI dependencies from the ComfyUI directory
pip install -r requirements.txtStart ComfyUI
python main.pyPyTorch 2.7 and above is supported, but the official requirements strongly recommend a newer version.
Which models and workflows can ComfyUI load?
ComfyUI can load complete checkpoints or separate diffusion models, VAEs, text encoders, LoRAs, ControlNets, adapters and upscalers from supported formats. Its checkpoint instructions name CKPT and SafeTensors files.
Workflows can be saved as JSON and extended with custom nodes. GGUF diffusion models use the separate ComfyUI-GGUF pack, which its maintainer describes as work in progress. Do not treat a ComfyUI workflow JSON file as a Draw Things or mflux model file.
How does ComfyUI use memory on Apple Silicon?
ComfyUI uses PyTorch MPS on Apple Silicon and marks its memory state as shared. Its MPS memory accounting comes from system memory rather than a separate physical VRAM pool.
The generic default unloads models to CPU memory after use. On unified memory, moving a model between devices does not by itself prove that physical RAM was released. ComfyUI also unloads loaded models under memory pressure.
Useful official controls include --lowvram, --novram, --cpu-vae, --cache-none and --disable-smart-memory. Their effects differ; use the failure guide rather than enabling every flag at once.
Frequently asked questions
Does ComfyUI work on Apple Silicon?
Yes. Comfy Desktop supports Apple Silicon M1 or later with macOS 13 or later. Manual ComfyUI uses PyTorch with the MPS backend.
Does ComfyUI use MLX on Mac?
No. The official requirements say ComfyUI does not use MLX directly. Apple Silicon acceleration uses PyTorch MPS.
How much RAM does ComfyUI need on a Mac?
There is no universal RAM requirement for arbitrary workflows. The model build, components, resolution, cache and app route all matter. Desktop's 4.85 GB figure is disk space, not RAM.
Can ComfyUI open GGUF models on Mac?
GGUF diffusion models are handled through the separate ComfyUI-GGUF custom-node pack. Follow that pack's current installation and model instructions; support is not interchangeable with core CKPT or SafeTensors loading.
Sources
Official documentation and source code unless the label identifies an individual issue report.