How to Install and Use FaceFusion on Mac (Apple Silicon)
FaceFusion runs natively on Mac with Apple Silicon — M1, M2, M3, and M4 chips are all supported. But getting it set up and running smoothly requires knowing a few Mac-specific tricks that the official docs barely mention.
This guide covers everything: installation (three methods), the CoreML execution provider, realistic performance expectations, Mac-specific bugs and fixes, optimal settings, and what to do when your Mac is too slow for video processing.
Already running FaceFusion but having issues? Check our common issues guide or the installation troubleshooting page.
Which Macs Can Run FaceFusion?
Before you start, make sure your Mac is compatible. FaceFusion dropped Intel Mac support entirely — only Apple Silicon is supported now.
| Mac Chip | Supported? | Performance |
|---|---|---|
| M1 / M1 Pro / M1 Max / M1 Ultra | Yes | Usable for images, slow for video |
| M2 / M2 Pro / M2 Max / M2 Ultra | Yes | Moderate — workable for short videos |
| M3 / M3 Pro / M3 Max | Yes | Moderate — slight improvement over M2 |
| M4 / M4 Pro / M4 Max | Yes | Best Mac experience, still slower than NVIDIA |
| Intel Mac (any) | No | Not supported since FaceFusion 3.5+ |
macOS version: Sequoia and Tahoe are fully supported. Older versions may work but are not officially tested.
The honest truth about Mac performance: FaceFusion's developer henryruhs said it directly on Reddit (June 2026):
"While Apple Silicone was a huge step compared to their Intel Arch, it is still known for overall bad performance for AI inference."
This does not mean FaceFusion is unusable on Mac. It means you should set realistic expectations — especially for video processing. More on performance numbers below.
Method 1: Official macOS Installer (Easiest)
FaceFusion offers a paid one-click macOS installer (v3.6.1, updated April 2026). This is the easiest way to get started if you are not comfortable with the terminal.
What It Does
- Automatically installs Git, Conda, and FFmpeg
- Creates a desktop shortcut and start menu entry
- Lets you launch FaceFusion without manually activating any environment
- Works on macOS Sequoia and macOS Tahoe
How to Get It
The official macOS installer is available on Ko-fi and Buy Me a Coffee.
Gatekeeper Warning
macOS will likely block the installer the first time you try to open it. This is normal. To fix it:
- Right-click the installer and select Open (do not double-click)
- Or drag
install.shto the Terminal app and press Enter
Method 2: Manual Installation via Conda (Recommended)
This is the method recommended by the official FaceFusion documentation. It gives you full control over your environment and is free.
Prerequisites
You need three things installed before starting:
- Homebrew — the Mac package manager
- Git — for cloning the FaceFusion repository
- Miniconda — for managing the Python environment
If you do not have Homebrew, install it first:
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
Then install Git:
brew install git
Download and install Miniconda from the official website. Choose the Apple Silicon (ARM64) version.
Step-by-Step Installation
Open Terminal and run these commands one at a time:
# Initialize Conda (only needed once)
conda init --all
# Create a Python 3.12 environment
conda create --name facefusion python=3.12 pip=25.0
# Activate the environment
conda activate facefusion
# Clone the FaceFusion repository
git clone https://github.com/facefusion/facefusion
# Enter the directory
cd facefusion
# Install with CoreML support (IMPORTANT: do not use --onnxruntime default)
python install.py --onnxruntime coreml
# Install FFmpeg
conda install ffmpeg
# Launch FaceFusion
python facefusion.py run
Critical: Use CoreML, Not Default
The most important line is python install.py --onnxruntime coreml. If you use --onnxruntime default instead, FaceFusion will run on your CPU only — which is dramatically slower. CoreML tells FaceFusion to use your Mac's GPU and Neural Engine.
First Launch
On first launch, FaceFusion downloads hundreds of MB of AI models. This takes a few minutes depending on your internet speed. Once finished, a URL appears in the terminal (usually http://127.0.0.1:7860). Open this URL in Safari or Chrome to use FaceFusion.
Reopening FaceFusion Later
Every time you want to use FaceFusion:
conda activate facefusion
cd facefusion
python facefusion.py run
Method 3: Pinokio One-Click Installer (Free but Quirky)
Pinokio is a free app that installs AI tools with one click. It works on Mac, but has a well-known issue that trips up many users.
The Stalling Problem
On fresh Mac setups (especially M4 machines), Pinokio installation silently stalls at step "(6/11) Installing brew." It sits there forever with no error message and no progress.
The Fix: Install Xcode Command Line Tools First
Before launching Pinokio, open Terminal and run:
xcode-select --install
A dialog box appears asking you to install the Command Line Developer Tools. Click Install and wait about 10 minutes. Once it finishes, restart Pinokio and the FaceFusion installation will proceed normally.
This fix comes from Reddit user samuraxxx and has been confirmed by dozens of Mac users.
Installation Methods Compared
| Factor | Official Installer | Manual (Conda) | Pinokio |
|---|---|---|---|
| Difficulty | Easy | Intermediate | Easy (with fix) |
| Cost | Paid | Free | Free |
| CoreML auto-configured | Yes | You choose | Yes |
| Auto-updates | Yes | Manual git pull | Via Pinokio UI |
| Known issues | Gatekeeper block | Dependency conflicts | Stalls without xcode-select |
| Best for | Non-technical users | Developers | Users who want GUI management |
Understanding CoreML: How FaceFusion Uses Your Mac's GPU
When you install FaceFusion with --onnxruntime coreml, it uses Apple's CoreML framework through ONNX Runtime. This is the only GPU acceleration path currently available for Mac.
What CoreML Actually Does
CoreML can run AI operations on three compute units inside your Apple Silicon chip:
- CPU — always available, slowest
- GPU — Apple's integrated graphics, significant speedup
- Neural Engine (ANE) — dedicated AI accelerator, fastest for supported operations
FaceFusion enables all three by default. However, not all model operations are supported by CoreML — unsupported operations automatically fall back to CPU. This is why you may see high CPU usage even with CoreML enabled.
Verifying Your Execution Provider
Before processing any media, check the Execution Provider dropdown in the FaceFusion UI. If it says "cpu" instead of "coreml," your Mac's GPU is not being used.
The performance difference is massive:
| Execution Provider | Estimated Speed (Face Swap) | 1-Minute Video (30fps) |
|---|---|---|
| CoreML (correct) | ~1.5 fps | ~20 minutes |
| CPU only (wrong) | ~0.3 fps | ~100 minutes |
Performance: How Fast Is FaceFusion on Mac?
Let us be honest about this. Mac is usable for FaceFusion, but it is significantly slower than a Windows PC with an NVIDIA GPU.
Real-World Benchmark Data
| Hardware | Approx. FPS | 1-Min Video (swap only) | 1-Min Video (swap + enhance) |
|---|---|---|---|
| Mac M4 (CoreML) | ~1.5 fps | ~20 min | ~60 min |
| RTX 4080 (CUDA) | ~2.5 fps (HD) | ~12 min | ~25 min |
| RTX 5090 (CUDA) | ~5–20 fps | ~3 min | ~6 min |
| RTX 4090 (CUDA) | ~4–8 fps | ~4 min | ~10 min |
Benchmarks use face_swapper + face_enhancer (2 processors). Source: community reports and dinhanhthi.com.
M4 Pro vs M4 Max: Is Upgrading Worth It?
FaceFusion's developer shared benchmark screenshots comparing M1 Max to M4 Max and found a small difference. His advice:
"The difference is small, I'd advise to get a PC with an NVIDIA GPU if you really want to get into AI."
This means upgrading from one Apple Silicon generation to another will not dramatically improve your FaceFusion experience. The bottleneck is the ONNX Runtime CoreML execution path, not the raw chip capability.
What Mac Is Actually Good For
- Image face swaps — single images process in seconds, perfectly practical
- Short video clips (under 30 seconds) — wait time is reasonable
- Testing and previewing — try settings on short clips before committing to long renders
- Experimenting with different models and processors
What Mac Struggles With
- Long videos (over 2 minutes) — processing times become painful
- Real-time webcam face swap — at ~1.5 fps, there is noticeable lag
- Running multiple processors (swap + enhance + expression restore) — multiplies already-slow processing time
- Frame enhancement — extremely GPU-intensive, not recommended on Mac
Best Settings for Mac Users
Mac hardware is limited, so the right settings make a big difference. Here is what works best.
Execution Settings
| Setting | Recommended Value | Why |
|---|---|---|
| Execution Provider | CoreML | Uses GPU + Neural Engine instead of CPU only |
| Execution Thread Count | 10–16 | Matches Apple Silicon's shared memory architecture |
| Video Memory Strategy | Default | Apple Silicon uses unified memory — no separate VRAM to manage |
| System Memory Limit | 0 (no limit) | Let FaceFusion use what it needs unless you need memory for other apps |
Processor Settings (Balanced Quality + Speed)
| Setting | Value | Notes |
|---|---|---|
| Processors | face_swapper + face_enhancer | Adding expression_restorer triples processing time |
| Face Swap Model | hyperswap_1a_256 | Best balance of quality and speed |
| Face Enhancer Model | gfpgan_1.4 | Lighter than GPEN models, good results |
| Enhancer Blend | 80 | High enough for clarity, low enough to avoid flicker |
| Pixel Boost | 768x768 | Higher values barely improve quality but massively slow processing |
What to Avoid on Mac
- frame_enhancer — extremely GPU-intensive, recommended only for RTX 4090 or better
- gpen_bfr_2048 — stunning quality but painfully slow on Mac; use gfpgan_1.4 instead
- Running 3+ processors simultaneously — each one multiplies your processing time
- Pixel Boost above 768 — diminishing returns with significant time cost on Mac hardware
Mac-Specific Bugs and How to Fix Them
CoreML Cache Corruption After macOS Updates
This is the most commonly reported Mac-specific bug. After updating macOS (particularly to Tahoe 26.3), FaceFusion throws this error:
[E:onnxruntime] Non-zero status code returned while running CoreML node.
Status Message: output_features has no value for 682
What happened: The macOS update changed the CoreML framework version, which invalidates previously cached model compilations.
How to fix it (in order of preference):
- Upgrade FaceFusion to 3.5.1+ — includes an automatic fix for this issue
- Delete the .caches folder in your FaceFusion installation directory — this forces recompilation
- Upgrade onnxruntime-silicon:
pip install --upgrade onnxruntime-silicon - Temporary fallback: Switch the execution provider to CPU in the FaceFusion UI
Webcam/Camera Not Working
macOS requires explicit camera permission for the app that launches Python. FaceFusion uses OpenCV for webcam access, and the permission needs to go to your terminal app, not to Python itself.
How to fix it:
- Open System Settings → Privacy & Security → Camera
- Find your terminal app (Terminal.app, iTerm2, or Pinokio) and toggle it ON
- If the permission prompt never appeared:
tccutil reset Camera - If the camera still does not work, kill stuck camera processes:
sudo killall VDCAssistant sudo killall AppleCameraAssistant - Make sure no other app is using the camera at the same time
Videos Modified by QuickTime Fail to Process
FaceFusion often cannot process videos that were saved or trimmed using QuickTime Player. The file format is technically valid but uses a codec configuration that confuses FaceFusion.
Fix: Re-encode the video with FFmpeg first:
ffmpeg -i input.mov -c:v libx264 -c:a aac fixed_input.mp4
This also has the benefit of compressing large .mov files — a 1 GB .mov often shrinks to ~50 MB as .mp4.
Rosetta Emulation Issues
If you installed Python through an Intel-compatible (Rosetta) path, FaceFusion may run with degraded performance or encounter unexpected errors.
How to check: Run this in Terminal:
python -c "import platform; print(platform.machine())"
If it prints x86_64 instead of arm64, you are running under Rosetta. Reinstall Python and Conda using the Apple Silicon (ARM64) versions.
Cloud GPU: The Fast Alternative for Mac Users
If your Mac is too slow for video processing, renting a cloud GPU is surprisingly cheap and easy. You access it through your browser — no special software needed.
RunPod (Recommended)
RunPod offers pre-built GPU environments that work perfectly with FaceFusion.
Pricing (as of 2026):
| GPU | Price/Hour | Speed vs Mac M4 |
|---|---|---|
| RTX 4090 | $0.34/hr | ~3–5x faster |
| A100 | $1.19/hr | ~5–8x faster |
| H100 | $1.99/hr | ~8–12x faster |
Quick setup:
- Create a Pod with a PyTorch template
- Add
7860to the Exposed HTTP Ports field - Open a terminal in the pod and install FaceFusion
- Launch with:
python facefusion.py run --host 0.0.0.0 - Access FaceFusion through the pod's public URL from your Mac's browser
Vast.ai (Budget Option)
Vast.ai is a decentralized GPU marketplace with even lower prices. The trade-off is less reliability — hosts can disappear mid-job. Use SSH for file uploads instead of the Gradio tunnel, which is slow.
Cost Math: Cloud vs Buying a PC
Processing a 5-minute video (9,000 frames) at ~5 fps on an RTX 5090 takes about 30 minutes. At RTX 4090 rates ($0.34/hr), that costs about $0.17 per 5-minute video.
Even if you process 100 videos, that is only $17 in cloud GPU costs. Compare that to buying an NVIDIA GPU PC ($800–$2,000+) and the cloud option makes a lot of sense for casual Mac users.
The Future: PyTorch MPS Support in FaceFusion 4
FaceFusion currently uses ONNX Runtime with CoreML, which is not the most efficient way to use Apple Silicon's GPU. The good news: this is changing.
What the Developer Said
FaceFusion developer henryruhs confirmed on Reddit (January 2026):
"I plan to support different execution drivers like onnxruntime and PyTorch once the FF4 API is stable... it's actually just a swap of the internal inference method."
What This Means for Mac Users
PyTorch MPS (Metal Performance Shaders) provides direct Metal GPU access — it bypasses the ONNX-to-CoreML translation layer. Based on general ML benchmarks, this could potentially reduce the Mac-vs-NVIDIA performance gap from 10–15x to 3–5x for supported operations.
Caveats
- No timeline has been announced for when FF4 will be ready
- Some PyTorch operations are not yet implemented in MPS
- Real-world FaceFusion performance improvement is speculative until the implementation exists
- Even with MPS, Mac will still be slower than an NVIDIA GPU with CUDA
Useful Tips and Tricks for Mac Users
Pre-process Your Videos with FFmpeg
Many issues on Mac disappear if you pre-process your input video:
ffmpeg -i input.mp4 -c:v libx264 -c:a aac -vf scale=-2:720 preprocessed.mp4
This converts to a clean H.264 format and downscales to 720p — which processes much faster on Mac hardware with minimal quality difference for face swaps.
Use the Trim Frame Feature
Instead of processing an entire 10-minute video, use FaceFusion's built-in Trim Frame Start/End to test your settings on a short clip first. Get the settings right on 5 seconds of footage, then apply them to the full video.
Split Long Videos Into Segments
For videos longer than 2 minutes, split them into segments before processing:
ffmpeg -i input.mp4 -c copy -segment_time 120 -f segment segment_%03d.mp4
Process each segment individually and rejoin them afterward. This prevents crashes from memory buildup during long processing sessions.
Monitor Activity Monitor
While FaceFusion is running, open Activity Monitor (Applications → Utilities) and check:
- CPU usage — should be high (expected when CoreML falls back to CPU for some operations)
- GPU usage — should show some activity if CoreML is working correctly
- Memory pressure — if this turns yellow or red, close other apps
Video Tutorials
These YouTube tutorials walk through FaceFusion installation on Mac step by step:
- FaceFusion macOS Install Guide — quick walkthrough of the Conda installation method
- Face Fusion 3.5 Complete Installation Guide — covers both Windows and Mac, with Mac-specific notes
- FaceFusion 2026 Installation Tutorial — updated for the latest version
Frequently Asked Questions
Can I run FaceFusion on an Intel Mac?
No. FaceFusion dropped Intel Mac support starting with version 3.5. You need a Mac with Apple Silicon (M1 or newer). If you have an Intel Mac, you can use FaceFusion online instead — it runs entirely in your browser.
How long does a 1-minute video take on Mac?
With face swap only on an M4 Mac, about 20 minutes. Adding the face enhancer brings it to about 60 minutes. For comparison, an NVIDIA RTX 4090 does the same job in 4–10 minutes.
Is the M4 Max much faster than M4 for FaceFusion?
Not dramatically. The FaceFusion developer shared benchmarks showing a "small difference" between M1 Max and M4 Max. The bottleneck is the ONNX Runtime CoreML execution path, not the chip's raw power.
Why is my Mac using 100% CPU even with CoreML enabled?
This is expected. CoreML can only accelerate certain operations — the rest fall back to CPU. Additionally, FaceFusion's data pipeline (image loading, augmentation, pre-processing) runs on CPU before feeding data to the GPU.
Should I buy a more expensive Mac for FaceFusion?
Probably not, unless you need the Mac for other reasons too. For the price difference between an M4 MacBook Air and an M4 Max MacBook Pro, you could rent cloud GPUs for years of occasional FaceFusion use. See the Cloud GPU section above for cost math.
How do I fix the "output_features has no value" error?
This happens when macOS updates corrupt the CoreML cache. Delete the .caches folder in your FaceFusion installation directory, or upgrade to FaceFusion 3.5.1+. See the Bugs and Fixes section above for the full solution.
Can I use FaceFusion's webcam mode on Mac?
Yes, but you need to grant camera permission to your terminal app first. Go to System Settings → Privacy & Security → Camera and toggle on Terminal.app (or whichever terminal you use). Real-time performance is limited at ~1.5 fps.
Is there a free alternative to the paid macOS installer?
Yes. The Conda manual installation method is completely free. Pinokio is also free. The paid installer just automates the same steps — it does not give you any additional features.
Still Stuck?
If FaceFusion is too slow on your Mac or you want a hassle-free experience, try FaceFusion online. It runs in your browser with no installation, no GPU required, and no command line. Upload your video, pick a face, and you are done.