Cinematic Avatar: HeyGen pilots its avatars by prompt via API

HeyGen's new Cinematic Avatar API lets developers control video avatars using natural language prompts to automate scene composition, framing, and movements.

HeyGen's video API introduces Cinematic Avatar, a natural language-driven variant of the POST /v3/videos endpoint. Instead of providing a script and a voice, the developer describes the desired shot, scene, action, framing, and mood, then provides HeyGen with one to three avatar looks. The Seedance pipeline handles composing the scene, movements, and framing.

Optional references, images or videos transmitted via URL, assetid, or base64, allow guiding the style, movement, or composition, with a combined limit of three videos and nine images. The request is then configured via a few parameters: duration from 4 to 15 seconds or automatic, 720p or 1080p resolution, 16:9, 9:16, or 1:1 aspect ratio, with an option for automatic prompt enrichment. Generation remains asynchronous and can be tracked via polling or webhooks.

The entire system targets programmatic pipelines: an existing HeyGen avatar retains its likeness while cinematic direction is built entirely by code, without going through the interface.