How long can an AI video be? Shorts 3 min, TikTok ads 10 min
YouTube allows Shorts up to 3 minutes and TikTok non-Spark ads up to 10. A Sume clip is at most 30 s, but Timeline renders up to 1800 s from several clips.

The platforms allow far longer videos than a single AI generation. YouTube's Shorts page says up to 3 minutes. TikTok's ad page says up to 10 minutes for non-Spark ads and lists no duration limit for Spark Ads. A Sume video model makes between 2 and 30 seconds per request, and Timeline joins clips into a render of up to 1800 seconds.
Three different limits
The platform limit is the ceiling for what you upload. The model limit is how long one generation can be. The assembly limit is how long your finished file can be. They stack: you can only reach the platform limit by joining clips.
| Layer | Limit |
|---|---|
| YouTube Shorts | Up to 3 minutes |
| TikTok non-Spark ad | Up to 10 minutes |
| TikTok Spark Ad | No restrictions listed |
| One Sume video request | 2-30 s depending on model |
| sume/auto, one request | 3-10 s |
| Timeline render | 1-1800 s |
| Video trim output | 0.2-900 s, source up to 1800 s |
What to do with it
For a Short, the limit you hit is your own script, not the platform's. Most buyers do not need 3 minutes. If you want 15 or 30 seconds, one clip from a 30-second model does it, and the 15-second ad post shows which rows reach that in one request.
For anything over 30 seconds, plan the audio first. Timeline's length is set by audio.duration_seconds, and the video slots must cover the spine to within half a second at the end. Starting from a voice track with a known length avoids a render that ends early.
Planning backwards from the platform
Plan from the end. Decide the final length first, then divide by the clip limit to get the clip count, then check the assembly limit. A 45-second ad is two clips from a 30-second model or three from a 15-second one; a 3-minute Short is six or more, as the clip-count post shows. Every extra clip is an extra seam and an extra chance of a continuity break.
The assembly step has its own limits. A Timeline job refuses more than 20 audio parts, and render.strategy: single is refused above 12 video slots, so a very cut-heavy edit uses the chunked strategy, which is the default past 12 segments. That is a rendering detail you rarely meet, but it matters if you cut every two seconds.
None of this is a platform rule. The platform pages state upper bounds on length. They do not say a longer video does better, and they do not guarantee a file at the limit uploads quickly.
A caution on TikTok's Spark row
The Spark Ads row says no restrictions on duration. That is the text on the page, and the page lists blanks for other attributes on that row. Do not assume a 30-minute file works. Test with a real upload before you rely on it.
Sources
Related posts
More in Use cases
- How to split a training lesson into avatar clips under 60 seconds
Sume refuses avatar videos over 60 seconds. Split a 6-minute lesson by words, render each part, and stitch with Timeline 1.0. Word counts and cost by tier.
- HR onboarding welcome video with an AI avatar: day-one clip
A 30-second welcome clip that lists the three things a new hire does on day one. Script size, cost per tier, and what must stay in text, not video.
- Bulk queue idempotency: header or body key, and the header wins
A Sume bulk queue takes an Idempotency-Key as a header or as a body idempotency_key. If you send both, the header wins. Replays return 202 with the old queue.
- Insert a sponsor read into a podcast: TTS plus one audio concat
Generate the ad read with Sume TTS, then join it into the episode at the second you choose with one Timeline audio concat: sample-exact, $0.01 per join.
Written by Sume