Player not loading? Watch on YouTube
This tutorial covers installing FaceFusion on Windows with NVIDIA CUDA support, then running a basic face swap. The title identifies version 3.7.1, while the speaker calls the demonstrated release 3.7. The introduction describes local AI features such as age modification, background removal, DFM swapping, expression editing and lip sync. The speaker also says FaceFusion supports CPU and AMD GPU execution, though this walkthrough follows the CUDA setup.
The stated prerequisites are a graphics card with at least 3GB, CUDA Toolkit 11.8 to 12.8, Python 3.11.0, FFmpeg, Git for Windows and Visual Studio. The installation starts with cloning the repository and creating a Python virtual environment. The presenter uses CUDA 11.8 and installs PyTorch before editing the requirements file to specify SciPy 1.15.1 and ONNX Runtime GPU 1.16.3. These dependency choices belong to the demonstrated configuration; the speaker says other CUDA versions need their own ONNX Runtime version.
After a NumPy error, the presenter removes and reinstalls NumPy, but the spoken transcript does not specify the replacement version. The app downloads models and provides a URL to open in a browser. A run.bat file simplifies later launches. The usage demo uploads a source face and a target photo or video, waits for the preview, and starts processing with optional face or frame enhancement.
