Wan 2.5 (officially "Wan2.5-Preview") is a multimodal text-to-video and image-to-video model from Alibaba's Tongyi Lab, unveiled on September 24, 2025 at the Yunqi/Apsara Conference. It generates videos with natively synchronized audio tracks (voices, sound effects, music, lip-sync) directly from text or image inputs. The model is available exclusively via Alibaba Cloud's DashScope/Model Studio API and various third-party APIs; unlike Wan 2.1/2.2, no Open Source model weights have been released to date. It supports resolutions up to 1080p at 24fps and video lengths of 5 or 10 seconds.

Sources (2)