Tongyi launches Fun-CineForge, the first film-grade dubbing multimodal large model.

Tongyi Labs has open-sourced Fun-CineForge, the first multimodal model supporting film-level dubbing, along with the CineDub method for building high-quality datasets. Based on the underlying capabilities of CosyVoice3, the model innovatively introduces a "temporal modality," integrating visual, text, audio, and temporal modalities to achieve four key functions: lip-sync, emotional expression, consistent timbre, and time alignment. It supports monologue, narration, and two-person and multi-person dialogue scenarios.