OpenMOSS/MOVA

Based on the README, here's a technical summary: Generates synchronized video and audio in a single forward pass using an asymmetric dual-tower architecture with bidirectional cross-attention fusion, eliminating cascaded pipeline errors and enabling precise multilingual lip-sync. Supports multiple resolution tiers (360p/720p), flexible memory management via component-wise or layerwise CPU offloading, and provides inference scripts alongside full training pipelines and LoRA fine-tuning capabilities. Available via HuggingFace model downloads, ComfyUI integration, and REST API.

833 stars. Actively maintained with 4 commits in the last 30 days.