DaVinci MagiHuman

DaVinci MagiHuman is an open-source model for generating lip-synced talking video. You give it one portrait photo plus a script or an audio track, and it produces a clip where the mouth movement matches the speech, with the audio generated alongside the video in a single pass rather than stitched from a separate text-to-speech step. It has 15 billion parameters and is released under Apache 2.0, with a free hosted demo available.

- Drive a talking head from a single reference portrait.

- Generate audio and video together, avoiding a separate TTS stage.

- Support multiple languages for lip sync, subject to the training data.

- Choose output resolution from 256p up to 1080p.

- Self-host by pulling the weights from Hugging Face and running the CLI.

Content creators producing presenter or virtual anchor video, researchers working on audio-visual generation, and marketers who need localised product explanation clips.

The hosted demo and the model licence are separate things, so read the demo's own terms before relying on it. Realistic talking heads are a recognised impersonation risk: only use images and voices you have the right to use, and be aware that synthetic media of a real person may breach platform rules or the law, with the consequences falling on you. The quoted inference speed assumes an H100-class GPU, so self-hosting is not cheap. Run one full task end to end first, then expand once you are comfortable with the workflow and the result quality. Behavior may vary across browsers and devices, so test before production use and consult the help docs if errors appear.