YouTube Shorts up to 3 min: how many Sume clips each model needs

YouTube's own Shorts help page says Shorts can be up to 3 minutes long. No Sume video model makes 180 seconds in one request: the longest single clip is 30 seconds (Seedance 2.5 and Wan 3.0), so a full-length Short is at least six clips joined in Timeline. Models capped at 15 or 10 seconds need 12 or 18.

Clip count by model

The counts below are plain arithmetic: 180 seconds divided by the longest clip a model accepts, rounded up. You can use shorter clips, but you cannot use longer ones, because the catalog rejects a duration the model does not list with a 400 instead of clipping it.

Why fewer clips is not always better

A 30-second clip is one continuous generation, so characters and lighting hold together. But a longer clip is also a bigger bet: if the last five seconds are wrong, you regenerate all of it. Twelve 15-second clips cost the same seconds in total, but you can swap one bad clip without touching the other eleven. Pick by how much continuity the script needs. A talking-head-style Short wants few clips. A montage of product shots does not care.

Joining the clips

Timeline takes one audio spine plus ordered video slots and returns one MP4. The spine length is audio.duration_seconds, which can be 1 to 1800, so 180 is well inside the limit. The default output is 1080x1920. The public rate in the Timeline doc is $0.10 per output minute, rounded up, so a 180-second render reserves three minutes, or $0.30. Confirm the live rate in the catalog before you budget.

Record the voice first and cut the pictures to it. The voiceover order for Shorts walks through TTS, then captions, then Timeline, and the six-clip Wan 3.0 example prices one full run.

A worked plan for 180 seconds

Say the script is a 180-second product walkthrough in nine beats of 20 seconds. Seedance 2.5 or Wan 3.0 can make each beat in one request, so you submit nine jobs, not six, and each lands on a natural cut. That is the useful way to read the table: the minimum clip count is a floor, and a beat structure usually puts you above it.

Submit the jobs together with distinct idempotency keys so a retry of one does not create a duplicate. Poll each, download the MP4s, and import them to Sume's media host before Timeline can use them, because Timeline sources must be Sume-hosted. Then build the spine from your voiceover and set each slot's start and duration to the beat. The slots must start at zero and increase, and a slot cannot stop more than half a second before the end of the spine.

If one beat comes back wrong, only that slot changes. The other eight stay as they were, and the Timeline job is the only thing you rerun.

What I could not confirm

The YouTube page says Shorts are up to 3 minutes. It does not say anything about how long a render may take or how the player treats a 179-second file. Check the upload in the YouTube app before you schedule a batch.