avatarModel when you create a persona or session token.
For most new production builds, set avatarModel to cara-4.
If you omit
avatarModel, Anam uses the latest model version available for the selected avatar. This default can change as new versions become available, so set avatarModel explicitly when you need predictable rendering behavior.Available models
Cara 3
SetavatarModel to cara-3 when an existing deployment needs to stay on the previous stable model. Cara 3 outputs 720x480 landscape video.
Cara 4
SetavatarModel to cara-4 for new production avatars, higher-resolution sessions, and stronger custom avatars from a single photo. Cara 4 supports 1152x768 landscape output and 768x1152 portrait output.
For custom avatar setup, see Create custom avatars and Custom avatar best practices.
Cara 4 Latest
SetavatarModel to cara-4-latest to test Anam’s newest model capabilities before they are pinned to a stable release. Cara 4 Latest is experimental and is not recommended for production sessions.
Cara 3 vs. Cara 4
Cara 4 is the current generation of Anam’s avatar model. Compared with Cara 3, it produces higher-resolution video, better lip sync, more natural expressivity, smoother transitions between speaking and idle states, and stronger identity preservation for custom avatars. Cara 4 also supports portrait output for mobile and vertical layouts. To request a portrait session, setvideoWidth to 768 and videoHeight to 1152 in sessionOptions. See Video options for supported dimensions and crop behavior.
Use Cara 3 only when you need continuity with an existing integration or avatar that was built around Cara 3 output.

