MagicMirror uses CPU by default for inference and currently does not support hardware acceleration options. However, you can try MagicMirror’s CLI version, TinyFace, or use FaceFusion.
This is thanks to Tauri, which uses the system’s built-in WebView component, unlike Electron, which packages an entire Chromium browser and Node.js runtime into the app. If you're interested in the technical details, click here.
Every design element in MagicMirror - from the logo and typography to the UI - was created using AI ✨
I used Tensor.ART, a free AI image generation platform. It offers access to the latest models like Flux and SD 3.5, plus the ability to train your own custom models.
For example, I created MagicMirror's logo by training a Flux-based model on reference images collected from Dribbble - quick and simple.
Compared to platforms like Civitai and LibLib.AI, Tensor.ART has more models at a lower price, making it the most cost-effective option.
If you're looking to try AI image generation or want a more affordable platform, I recommend giving Tensor.ART a try.
This issue occurs because the app isn't signed with an Apple developer certificate. To resolve this:
- Open Finder and select Applications from the sidebar.
- Locate the app with the warning, right-click it, and choose Open.
If the app is stuck at the startup screen, check if the server.exe file exists on your system. This file, compiled using Nuitka from Python source, is essential for MagicMirror to run.
Windows Defender may flag the server.exe as malicious due to the use of Nuitka by some malware, leading to it being quarantined.
To fix this:
- Open Windows Security Center and go to Protection History.
- Find and restore the quarantined
server.exe. - Restart MagicMirror.
For details on Nuitka false positives in Windows Defender, click here.
If none of the above work, your system likely has compatibility issues with the server. Consider using alternative face swap tools like TinyFace (the underlying library used by MagicMirror), or FaceFusion.
Learn more: #6 (comment)
First, check if the server.exe file exists in $HOME/MagicMirror/. If it’s missing, follow the steps in the previous section to restore it.
If the issue persists, please submit an issue.
MagicMirror currently only supports x64 and ARM64 architectures. Older x32 systems are not supported.
The recommended environment is Windows 11 x64. MagicMirror may not run properly on versions lower than Windows 10.
First-time startup on macOS can be slow. If the app doesn’t start within 30 minutes, close it and try again.
If it still doesn’t launch, ensure your macOS version is at least 14 (Sonoma), as older versions may not be compatible.
Learn more: 521xueweihan/HelloGitHub#2859 (comment)
In simple terms, MagicMirror is a simplified version of FaceFusion.
Starting from FaceFusion, I removed unnecessary modules and kept only the core face-swapping functionality, which led to the creation of TinyFace, a lightweight Python face-swapping tool.
MagicMirror is a GUI project built on top of TinyFace, making it easier to use.
A typical face-swapping workflow includes multiple steps, such as face detection, recognition, swapping, and quality enhancement. These steps rely on different models. The inswapper_128.onnx model from InsightFace is a core part of the face-swapping process.