wtf is ltx-faceid-lipsync?

This project turns a single photo and a voice recording into a talking video of that person saying exactly what your recording says. There is no driving video and no face swapping involved, the AI generates the whole talking clip from scratch while keeping the face matched to your reference photo and the lips synced to your actual audio. It is built on top of an existing AI video model called LTX-2.3 along with an add-on that locks identity to a photo, both run through a visual tool called ComfyUI. The standard version of that setup lets the AI invent its own voice for the person. This project adds a small change that instead freezes your own voice recording into the process, so the video's mouth movements are generated to match your actual audio rather than a made up one. There are two versions of the workflow file, one tuned for Nvidia graphics cards and one tuned for Apple Silicon Macs, since the two run different compressed versions of the underlying model for performance reasons. To use it, you install a handful of ComfyUI add-ons, load the workflow file, drop in a close up frontal photo of a face, drop in your voice recording as a WAV file, set the video length to match the audio length, and write a text prompt describing the person's appearance along with the action and setting. How well the video matches the reference photo depends heavily on how detailed that description is. Generating video is fairly slow on a Mac, taking several minutes for just a couple of seconds of footage, so this suits short takes rather than long videos. This tool is aimed at people already comfortable with ComfyUI who want to create synthetic talking head videos of a person speaking specific words, for things like demos, mockups, or creative projects, using their own voice rather than an AI generated one.