Alibaba's Wan2.6 brings native lip-sync and multi-shot video

Alibaba has unveiled Wan2.6, a native multimodal model built to tackle the biggest headaches in AI filmmaking: character drift and awkward lip-syncing.

The new model can generate 15-second, 1080p HD videos with upgraded motion physics and instruction following. But its standout feature is something called Starring, which lets creators pull a human or human-like character from a reference video and drop them into new scenes while preserving their appearance and voice.

Moving past single-shot clips

Instead of forcing users to generate single, disconnected shots, Wan2.6 introduces intelligent multi-shot narrative capabilities. A simple prompt can now turn into an auto-storyboarded sequence that maintains visual continuity across different camera angles.

It also brings native audio and video synchronization. The model produces studio-quality sound and realistic lip-sync for multi-speaker dialogues right out of the box. Beyond video, Wan2.6 handles advanced image synthesis with fine-grained lens and lighting control, alongside structured image-and-text storytelling powered by real-world reasoning.

Why it matters

For digital creators and video teams, Wan2.6 looks like a major push from generating random B-roll clips to assembling actual, ready-to-use scenes. Bundling character consistency, auto-storyboarding, and native lip-sync into one system eliminates a messy multi-tool editing pipeline.

In our view, making voice consistency and lip-sync native is a smart move that hits real creator pain points. It clearly sets a new baseline for the industry, pushing competitors to treat video, audio, and narrative structure as a single unified generation task.