Alibaba speech-to-video avatar model
Use it when a portrait should speak with natural face movement, body presence, and scene-aware gestures.
Your generated avatar will appear here.
Upload a portrait image, add voice audio, and click 'Generate Avatar' to start.
Examples
Hover a card to preview, then open the player to watch with sound and review the full prompt.
A cooking host who gestures at ingredients on cue from the narration.
A warm cooking-show host speaks to the camera and gestures naturally toward the ingredients, with steady medium framing, soft kitchen light, realistic body motion, and stable identity.
Model context
Wan 2.2 Speech to Video turns portrait and audio into a talking clip with face and body motion. It is useful when the avatar should feel like a host, not only a mouth synced to sound.
That makes it a fit for cooking hosts, explainers, social presenters, and longer speech-driven clips where gesture and posture support the voice.
Model strengths
Use it when a portrait should speak with natural face movement, body presence, and scene-aware gestures.
Prompt guide
A head-and-torso image works better when you want hands, posture, or scene interaction.
The audio drives timing and length, so finalize the take before rendering.
Say when the host points, leans, smiles, or gestures, and keep movements natural.
Check face stability, mouth timing, and hand motion throughout longer clips.
It is best for speech-to-video avatars with natural face and body motion, especially presenter or host-style clips.
Use a clear portrait with enough torso or scene context if you want gestures or body movement.
Yes, prompt the desired gestures and make sure the source image provides space for that motion.
Use LTX 2.3 when precise lip sync in a short talking-head clip is more important than body motion.