HeyGen video scenes API vs Sume job result previews

HeyGen added GET /v3/videos/{video_id}/scenes in August 2026 to read back a finished video's scene-by-scene composition. Sume has no equivalent endpoint named in its docs: a completed Sume job result carries media.sume.com artifacts plus public-safe preview fields such as preview_image_url and scene_previews.

HeyGen facts are from its changelog; Sume's from Generate avatar video, read 2026-09-30.

What does HeyGen's scenes endpoint return?

Per the changelog, it returns what each scene shows and says, plus video-level context: title, aspect ratio, resolution, captions and brand glossary. It works for any video in the workspace, however it was created, and describes the video as it stands now, including later editor changes. Each scene splits into background, elements and script; walking scenes[].script[].text yields the narration in playback order.

What does Sume return after an avatar render?

After you submit to the avatar talking-video route, you read the job with /v1/jobs/{id}/status, /events and /result. Completed results can include public media.sume.com video artifacts plus preview_image_url and scene_previews. You can also list or fetch avatar-video resources at /v1/avatar-videos and /v1/avatar-videos/{id}.

How do I keep a record of what a Sume video said?

Store the request you sent. Multi-scene plans go in ordered video_inputs, which is the same text you would otherwise read back. Pair that with the job id and the scene_previews from the result so each scene has a still. To approve composition before a full render, use first-frame previews; see avatar video first-frame previews.

Is there a way to inspect a video I did not generate?

Sume's docs describe reads over avatar-video resources and jobs in your workspace, not a composition inspector for arbitrary videos. HeyGen's changelog says its endpoint covers videos made by any surface in a workspace, including hand-edited ones. That is the difference to plan around if you audit edited videos.