DaVinci MagiHuman
Open-source model that makes lip-synced talking videos from one portrait plus audio or text
Introduction
daVinci-MagiHuman is a 15B-parameter open-source model from Sand.ai and GAIR Lab. It generates talking videos from a single portrait and a script or audio, producing sound and picture in one pass. It uses the Apache 2.0 license.
- Added on
- Nov 2, 2026
- Website
- davinci-magihuman.com
- Company
- Sand.ai and GAIR Lab (Shanghai Jiao Tong University)
Product information
What is DaVinci MagiHuman?
daVinci-MagiHuman is an open-source 15B model that produces talking videos from a portrait image and a script or audio file, handling video and audio together in a single Transformer.
How to use DaVinci MagiHuman?
Provide a portrait photo and either a script or an audio file, then run the model locally (weights are available) to produce a lip-synced clip.
Core features of DaVinci MagiHuman
- Joint audio and video generation
- Talking head from a single photo
- Multilingual lip sync
- Apache 2.0 license
- Fast inference on H100-class GPUs
- Low word error rates
Use cases of DaVinci MagiHuman
- 1Marketing avatars from static portraits
- 2Multilingual educational content with synced lips
- 3Low-latency digital humans for interactive apps
- 4Prototyping talking-head clips for social media
Frequently asked questions about DaVinci MagiHuman
What is the main advantage of daVinci-MagiHuman?+
It generates audio and video together in one pass, so no separate text-to-speech and video pipelines are needed, which helps sync.
Can daVinci-MagiHuman be used commercially?+
Yes, the Apache 2.0 license permits commercial use, modification and local hosting.
DaVinci MagiHuman: What hardware does it need?+
Speed depends on the GPU; an NVIDIA H100-class card is reported to make a 2-second clip in about 2 seconds.
Which category is DaVinci MagiHuman in?+
DaVinci MagiHuman is listed under AI Avatar Video.