Personas and consistent characters
Create a reusable visual identity in Models.
The Models area holds reusable personas: a face identity with reference images that you can bring into image and video generation. It is distinct from choosing an AI generation model.
Create a persona
- Open Models in your workspace.
- Create a persona from a description or upload reference photographs.
- Give it a recognizable name. The upload workflow expects 3–10 reference images.
- Review the generated or uploaded preview before using it.
- Choose whether to share the persona with the workspace where that control is available.
Use clear images of the same person and only material you have the right to use. A description-based persona generates its initial face and consumes generation credits.
Generate with the persona
Open the persona, select image or video, and describe the scene. The references preserve the face; your prompt controls the clothing, framing, expression, style, and action. Review every output for consistency.
The legacy AI Avatars page described a dedicated talking-head workflow. The current persona feature does not by itself promise lip synchronization or the former avatar editor. Use the capabilities actually offered by the selected video model.
Developers can create and retrieve these resources with the Personas API.