HeyGen Video

HeyGen describes this as a general-purpose video model built on MiniMax H3 and post-trained by HeyGen. A prompt can specify the subject, setting, lighting, and sound together. The result is a short scene with its own audio track, rather than a talking-head avatar chosen in advance and animated from a prepared script.

The distinction matters when choosing an endpoint. The model page identifies POST /v3/models/videos and GET /v3/models/videos/{video_id} for HeyGen Video. It explicitly contrasts that model workflow with POST /v3/videos, which belongs to avatar-rendering engines. If the goal is to animate a selected digital presenter reading supplied copy, consult the avatar API instead of treating these endpoints as interchangeable.

The model identifier in the API reference is heygen-video-1. The documentation separates creation into three modes, each with a distinct role for the supplied material. Reference-to-video is described as producing a new scene while retaining the products, people, or places in supplied references; the docs allow references that include images, videos, or audio.

HeyGen says the model works best on short, contained shots: one subject, one place, one action, with a locked or barely moving camera. The documentation calls out coherent scenes, literal brief-following, picture and sound generated together, and holding details from a supplied object reference as strengths.

The same page cautions about long on-screen text, soft organic motion such as petals, paper, and hair, and close hand work such as assembling or operating equipment. Those are useful constraints for prompt design: simplify the shot when reliability matters, avoid making small readable lettering the central requirement, and review hand-object interaction rather than assuming it will be correct.

The documented create call is POST /v3/models/videos and returns HTTP 202 with a pending job identifier. Retrieve status with GET /v3/models/videos/{video_id}; pending means queued, processing means running, and completed, failed, or cancelled are terminal states. The API uses an x-api-key header. The page names videos:write for creating, videos:read for retrieving, and assets:write for uploading references.

A completed response includes a signed video_url, which the docs say can be refreshed by polling again. The example request uses a 5-second duration, 768p resolution, and 16:9 ratio; treat those as an example, not a complete list of available settings. The model documentation does not publish a price table, so consult HeyGen’s current account or pricing information before estimating cost.