HeyGen Studio video scene voiceover: freeze, loop or fit_to_scene

In HeyGen's Studio API, a video scene can take an optional voiceover, and playback.mode decides how the clip fills that voiceover: freeze, loop or fit_to_scene. Sume has no per-scene playback mode; you control timing with a silence beat in avatar video or with Timeline 1.0 slots.

What the HeyGen changelog says

From the HeyGen API changelog, read 2026-10-02:

• Voiceover and clip audio compose together.
• Studio scenes can be avatar_video, image or video, 1 to 50 per request, per the same changelog.

The Sume equivalent for talking scenes

In an avatar video with video_inputs, each scene has a voice object. A spoken scene uses type: "text" with a duration; a non-speaking beat uses type: "silence" with a required duration. Total planned length must stay in 4 to 60 seconds.

That gives you a held beat, which is similar to freeze, but only inside an avatar clip with one avatar and one shared scene. There is no looping or speed-fitting of an existing clip in that request.

For existing footage

For B-roll that must match narration length, use Timeline 1.0. You give it an audio spine and ordered video[] slots, and the compiler pads or loops short sources, with warnings it cannot predict in the unbilled plan step. Use the plan call first to see duration and segment count.

• Want a freeze: put a silence beat, or let a short source hold.
• Want a loop: choose slot lengths so the source repeats, and read the warnings.
• Want speed-fit: Sume's avatar request does not retime an existing clip; pick or trim the source length yourself.

Summary

HeyGen exposes the fill behaviour as a field on the scene. Sume splits it into separate documented steps. If you need the one-field version for a clip under a voiceover, HeyGen's is simpler; if you want to keep each stage inspectable, Sume's separate jobs are easier to reason about.