Context

If you are using an AI-generated avatar, 3D character, or cartoon face as your video input, you may notice poor lip sync quality, robotic-looking mouth movements, or no mouth movement at all.

Why this happens

Sync's lip sync models are trained primarily on real human faces. AI-generated characters, 3D avatars, and cartoon faces have different facial geometry and textures that the models may not handle well.

• Face detection may fail if the avatar's face does not match real human characteristics.
• Mouth region processing may produce artifacts because the model expects real skin texture.
• Movement may look robotic because patterns learned from real speech may not translate to stylized characters.

What you can try

• Use lipsync-2-pro — it handles a wider range of facial types due to diffusion-based super resolution.
• Use a photorealistic AI avatar rather than a cartoon or stylized character.
• Ensure the avatar's face is front-facing, well-lit, and clearly visible.
• When generating the avatar, include “person is speaking naturally” in your prompt for a more natural mouth starting position.

Known limitations

• Cartoon and anime faces are generally not supported.
• 3D-rendered characters with non-photorealistic textures may produce poor results.
• Static AI-generated images have limited lip sync potential.

Related docs: Improving Lip Sync Quality • Model Comparison • Troubleshooting