Syren's 3-minute cap and 30-second avatar scenes vs Sume's 60 s

Syren's help article says videos can be up to 3 minutes long, and also that avatar scenes longer than 30 seconds can time out (read 2026-10-11). Sume's avatar route accepts one talking video of 4 to 60 seconds per job, so a 3-minute presenter video is three jobs joined on a timeline rather than one request.

This post only compares the limits each side documents. It does not rank the products. The Syren facts come from How do I create a video with Syren?, read 2026-10-11; the Sume facts come from Generate avatar video and Timeline 1.0.

The two limits side by side

The numbers are not the same kind of limit: Syren's 30 seconds is a warning about reliability inside one scene, while Sume's 4 and 60 seconds are an accept-or-refuse window checked at submit.

How Sume checks length before you pay

Sume estimates the spoken duration from the script or from video_inputs and accepts it only when the estimate is between 4 and 60 seconds inclusive. If it is outside that range, the request is refused before any render starts, so the cost of finding out is zero. For multi-scene requests each spoken scene carries a duration, and silence beats are scenes with voice.type: "silence" and a required duration.

The practical effect: a 30-second scene is nothing special on Sume. A single scene may fill the whole 60 seconds.

Planning a 3-minute presenter video on Sume

Split the script at paragraph breaks into three parts of about 60 seconds each, one avatar job per part with the same avatar_handle. Each job gets its own Idempotency-Key, so a retry never bills twice. At the published per-second rates, three 60-second jobs on the standard tier are 180 seconds at $0.184, which is $33.12; on plus ($0.245) $44.10; on max ($0.55) $99.00, before any product image surcharge. The join is a Timeline 1.0 render at $0.10 per started output minute (about $0.30 for three minutes) plus the audio spine step described in the docs.

Mind your account limits: three jobs submitted together count against the plan's processing concurrency, and extra jobs wait as queued until a slot opens (Generation admission). A Free workspace processes one at a time with room for six accepted jobs, so a three-part series fits.

Order matters less than people expect. Because the avatar window is the same for every part, you can submit all three at once with three idempotency keys, store the three job ids, and poll each. If part two fails you retry only part two with its own key; the other two clips are untouched. Compare that with a single long scene on a tool where a timeout means resubmitting the whole thing.

Where the seam shows

Cutting a presenter mid-sentence is the visible risk of any split. Put the cut between sentences, start each part with a short hook line, and keep the same avatar and scene so continuity holds. Previewing the first frame of each part costs no full render and lets you check that the framing matches (Avatar video previews). If you use Syren for a 3-minute video instead, split any avatar scene that approaches 30 seconds, given what Synthesia's page says.

Finally, check what you are actually timing. Sume's window is measured on the estimated spoken length of the script, not on the file you hope to get, so a script near the limit can be refused even if you think it reads in under a minute. Split early and leave yourself some margin below 60 seconds for each job; a request that Sume refuses costs nothing, but it costs a round trip.