DeepFaceLab is free, open-source software for creating deepfake videos. It replaces one person's face with another, de-ages a face, or swaps an entire head, and it ships with an XSeg editor for masking and a merger for compositing the result back into footage. The code is released under the GPL-3.0 license and runs on TensorFlow with CUDA, with Windows builds distributed by torrent and file hosts and a Linux build maintained by contributors. The repository was archived by its owner on November 13, 2024, so it is now read-only, but the code and releases remain available.
It is aimed at VFX artists, YouTube creators, film hobbyists and machine learning enthusiasts who want full control over a face-swap pipeline instead of a one-click app. The developers state plainly that there is no button that makes everything perfect: users need to study the workflow and build skills, and experience with After Effects or DaVinci Resolve helps with finishing. Because training happens on your own hardware, you need a capable NVIDIA graphics card, along with patience for long training runs. Researchers can also read the accompanying paper on arXiv.
The project is used by well-known deepfake channels, including those behind the Tom Cruise clips, as well as groups producing parody videos and film effects. Typical jobs include replacing an actor's face in a short clip, de-aging a performer for a scene, swapping a whole head including hairstyle, and producing before-and-after progress comparisons at native resolution. Related work from the same author, DeepFaceLive, covers real-time face swap for streaming or video calls. A Discord server, available in English and Russian, offers community help and tutorials.
