AI avatar video length limit: Sume accepts 4 to 60 seconds

Sume Avatar 1.0 accepts scripts that it estimates at 4-60 seconds. What happens outside that window, and how to split a longer script into jobs.

5 min readSume
All posts

Sume Avatar 1.0 accepts a script or multi-scene plan only when it estimates the video duration at 4 to 60 seconds inclusive. A shorter or longer script is not rendered as one job. To make something longer than a minute, split the script into several jobs and join the clips afterwards.

Where the limit applies

The window is the same on every avatar-video surface, which matters because people often assume previews or captions are looser. From the docs read on 2026-10-08:

Duration windows for avatar surfaces, as of 2026-10-08
SurfaceDuration rule
Generate avatar video (talking-video)Estimated 4-60 seconds inclusive
Avatar video previewsSame window as Avatar Video
Inline captions on talking-videoSume rejects an estimated duration over 60 seconds
Face swap (Beta)Source video approximately 4-15 seconds with usable audio

The duration is an estimate

Sume estimates length from the script text. For multi-scene video_inputs, each spoken scene carries a duration field and the total planned duration must still land in the same window. That means you control the length in two ways: with the amount of text in a single script, or with explicit scene durations in video_inputs.

If you are near an edge, aim for the middle of the window. A script that you think is 58 seconds may be estimated at 61. A 3-second line may be estimated below 4. Run the preview first if you are unsure; it uses the same window and costs no full render.

Splitting a long script

Cut at natural boundaries: a topic change, a question and its answer, or the end of a point. Each job gets its own script and the same avatar_handle, so the presenter looks the same across clips. Use a separate Idempotency-Key per segment so a retry does not create a duplicate. A 3-minute piece is then three jobs of up to 60 seconds each; the pricing arithmetic for that is in the existing 3-minute presenter reel post.

  • Keep every segment well inside 4-60 seconds.
  • Give each segment the same avatar_handle, aspect_ratio and quality.
  • Name segments in your own system by order, not by job id, so assembly is deterministic.
  • Join clips in order with a timeline job or your own editor.

What to do when a request is rejected

A request outside the window is a client error. Do not retry the same body. Shorten or split it, then submit with a new Idempotency-Key. For the submit and poll pattern see Jobs and results.

Planning around the window

A practical way to plan is to write the script first, read it aloud at a natural pace, and time it. Spoken English at a steady pace is often cited at around 150 words a minute, so a 60-second ceiling is roughly 150 words and a 4-second floor is a short sentence. Sume's estimate may differ from your stopwatch, so leave margin.

If a script is too short, do not pad it with filler. Combine it with a second short line, or add a silence scene with an explicit duration in video_inputs to reach the floor honestly. If it is too long, remove detail rather than speeding up delivery, because viewers retain more from a clean 45 seconds than a rushed 70.

Batch jobs should respect the account's limits too. Check the generation_limits in submit responses before queuing many segments, as described in Generation admission.

Sources

Related posts

More in Sume Avatar 1.0

All Sume Avatar 1.0 posts

Written by Sume