Alibaba Wan 2.6 AI Video Generator | Clivio AI

You provide a reference video containing your character's appearance and voice, then describe each new scene in text. Wan 2.6 generates subsequent shots that preserve the character's face, clothing, body proportions, and vocal timbre while placing them in entirely different environments. This lets you build a narrative arc across multiple clips without the identity drift that plagues single-shot models stitched together manually.

Wan 2.6 produces dialogue with natural lip-sync, ambient environmental sound, and foley effects in a single generation pass. It supports multi-person conversations where each speaker maintains a distinct voice. The audio is not layered on after video generation; both modalities are co-produced, which eliminates the timing mismatches common in post-dubbed workflows.

Three headline upgrades: 30% faster generation speed from an optimized diffusion scheduler, native audio-visual co-generation that Wan 2.5 lacks entirely, and multi-shot scene continuity with reference video support. Prompt comprehension is also sharper, particularly for complex compositional instructions involving multiple subjects and spatial relationships.

Individual clips run up to 15 seconds at 1080p resolution. For longer narratives, you chain multiple 15-second shots using the multi-shot system, where each new clip inherits visual and audio continuity from the reference. This approach scales to minutes of coherent content while keeping each generation fast and controllable.

Yes, and this is one of its standout capabilities. You can describe a conversation between two or more characters, and the model generates each person speaking with distinct lip movements, vocal tone, and timing. Turn-taking feels natural rather than robotic, and the camera framing adjusts to follow the active speaker when prompted to do so.

A 3-5 second clip showing the character's face from a roughly frontal angle with clear lighting and at least a few words of speech. The model extracts facial geometry, skin tone, hair style, clothing details, and voice characteristics from this reference. Avoid heavy filters or extreme angles in the reference, as these can introduce artifacts in the generated scenes.

Wan 2.6 is Alibaba's current flagship video generation model, succeeding the open-source Wan 2.5 line. While Wan 2.5 remains available and cost-effective for simpler tasks, Wan 2.6 represents Alibaba's push into narrative-grade video AI with audio. The multi-shot and dialogue capabilities position it as a direct competitor to Google's Veo line for storytelling applications.