Run Stable Diffusion on Mac Mini M4
How to run SDXL and FLUX on the Mac Mini M4 using Diffusers or ComfyUI — with expected generation times and optimization tips.
Generation Time
18–25s per 1024×1024 (SDXL)
Min Memory
16 GB
Software
Python 3.11, ComfyUI, pytorch-nightly (MPS)
Hardware Used in This Guide
mini-pc · Check Price on Amazon
Step-by-Step Setup
- 01
Install ComfyUI
ComfyUI has the best Apple Silicon support in the image-gen ecosystem. Clone it and install dependencies.
git clone https://github.com/comfyanonymous/ComfyUI cd ComfyUI pip install -r requirements.txt
- 02
Download SDXL base model
Place the model checkpoint in ComfyUI's models/checkpoints directory. Use the fp16 variant to fit in 16 GB.
# ~6.5 GB download cd models/checkpoints wget https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/resolve/main/sd_xl_base_1.0_0.9vae.safetensors
- 03
Launch ComfyUI with MPS backend
The --force-fp16 flag prevents OOM on 16 GB systems running SDXL.
python main.py --force-fp16
- 04
Open the UI
Navigate to http://127.0.0.1:8188 and load a basic SDXL workflow. Queue a prompt to test generation speed.
- 05
Try FLUX.1 Schnell for faster generations
FLUX.1 Schnell generates in 4 steps vs SDXL's 20–30, cutting time to ~8–12s on M4. Download the fp8 variant.
# ~8 GB huggingface-cli download black-forest-labs/FLUX.1-schnell \ --local-dir models/checkpoints/flux-schnell
Optimization Tips
- ›
Use fp16 precision — fp32 doubles VRAM usage with no perceptible quality gain on M-series.
- ›
FLUX.1 Schnell (4-step) is 3–4× faster than SDXL on M4; use it for iteration, SDXL for final renders.
- ›
Keep image size at 1024×1024 or below on 16 GB to avoid system RAM spillover.
- ›
ComfyUI's MPS backend is more stable than the A1111 WebUI for Apple Silicon in 2026.
Other Hardware for SDXL / FLUX.1
mini-pc · Check Price on Amazon · 24 GB Unified