AI Video Face Swap, the first time I’ve seen a really easy offline desktop Windows AI tool for this. Yup, desktop PC, offline, TensorRT support, no watermarks, one-time-payment ($80, though no PayPal), and as simple to install and run as one of those Topaz AI software packages.
However it’s still 4,000 Python files sitting inside the bundled .EXE file. Users will need a good CUDA graphics card with 12Gb memory and CUDA installed, Windows 10 with 10Gb spare on the hard-drive, and the K-lite Codec pack for video input-output.
It seems to have obvious potential for making comics. Make a thirty-second set of all expressions you want, using your own face. Then use this to swop out the face for that of your character. Then extract the frames from the resulting video, for use in the comic. Or for pushing through a style-transfer AI.
Maybe only good for a front-view with a normal press-photo camera lens though, which is where 3D Poser/DAZ figures still have the advantage. Overhead, from the side, foreshortened lens… all no problem for Poser/DAZ.
Users of AI Video Face Swap report a high quality face-swap on a RTX 3080 card, transferring onto a three-minute 1280px video, takes about 30 minutes. Faster at standard quality. Multiple faces can be swopped at once. Edge-blending is good.
The dedicated alternative in free / open-source / offline software is the namesake Faceswap, but it looks like it may take a bit more setup. Still, not as much as most Python AI software does. It can also run on Linux.
With a lot more technical savvy and the right nodes you can also face-swap in general AI ‘Stable Diffusion’ image generation software such as ComfyUI and InvokeAI.