Wan 2.7 is Alibaba Tongyi Lab's latest open-source video model. It is a 27B-parameter Mixture-of-Experts diffusion transformer with 14B active parameters per inference pass, built on a DiT architecture with Full Attention that processes spatial and temporal relationships across entire sequences simultaneously, ensuring strong character identity stability. Wan 2.7 supports text-to-video, image-to-video, reference-to-video with voice cloning, and instruction-based video editing, along with a "Thinking Mode" for planned multi-shot compositions. The reference-to-video feature accepts up to 5 combined image or video references for precise casting control and generates 720p-1080p clips ranging from 5 to 15 seconds. It is designed for directors, narrative creators, and workflows that require consistent character continuity across scenes.
