Sand.ai says Magi-2 Preview is a 114B-parameter unified audio-video generation model built on MagiMoE that activates 6B parameters per token to explore an efficient path to scaling video generation.
Who it’s for
Creative partners and creators looking to generate audio-visual content such as advertisements, music videos, and cinematic scenes.
What it does, according to the maker
- Character Performance: Precisely generates dialogue, singing, and emotional shifts with natural lip-sync, facial expressions, eye movements, and body language.
- High-Fidelity Audio-Visual Synchronization: Generates visuals and audio jointly from a single model to ensure precise alignment.
- Cinematography: Understands shot scale, composition, perspective, and camera movement intent to mimic real-world filming.