Can a 30-second AI clip be a YouTube Short? Yes, up to 3 minutes
YouTube Help says a Short can be up to 3 minutes, so a 30-second AI clip fits. How to pick the model length and cut it to the file you upload, with Sume docs.

Yes. YouTube Help says a Short can be up to 3 minutes long, so a 30-second AI clip is well inside the limit. The work is not the length; it is getting a clean 9:16 file of the right duration out of the model.
What the YouTube page says
The YouTube Help page for Shorts (read 2026-10-05) is the only YouTube page used for the limits below. It gives the maximum Short length and an upload resolution ceiling. I could not find an aspect ratio stated on the page text I read, so I do not state one for YouTube here; a vertical frame is simply the common choice for a feed that scrolls.
| Item | Source | Value |
|---|---|---|
| Maximum Short length | YouTube Help | 3 minutes (180 s) |
| seedance-2.5 duration | Sume Video Router docs | 4 to 30 s |
| wan-3.0 duration | Sume Video Router docs | 2 to 30 s |
| Most other catalog models | Sume Video Router docs | Up to 15 s |
| gemini-omni-flash-1.1 | Sume Video Router docs | 3 to 10 s |
Nothing in the catalog overshoots
Every row in the Sume column is shorter than 180 seconds, so no catalog model can produce a clip that is too long for a Short on its own. A single call at the top of the longest range gives 30 seconds, one sixth of the limit.
Ask for the vertical clip directly
Choose the model by what the clip needs, then set the shape in the request. The Video Router docs show aspect_ratio: "9:16", a resolution, and a duration in one body:
curl -X POST https://api.sume.com/v1/video-router/generate \
-H "Authorization: Bearer $SUME_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: short-30s-001" \
-d '{
"model": "seedance-2.5",
"prompt": "A product demo on a kitchen counter, handheld, warm light",
"resolution": "1080p",
"duration": 30,
"aspect_ratio": "9:16",
"mode": "async"
}'Then wait for the job
The job is async. Poll GET /v1/jobs/:id/status and read GET /v1/jobs/:id/result when it is ready. Billing is the provider list rate times 1.25 for each model, so read the live price on GET /v1/catalog before you run a long clip at 1080p.
Trim before you upload
A 30-second clip rarely needs all 30 seconds. If the first three seconds are a slow establishing shot, cut them. Video trim takes start and duration, accepts a source up to 1800 seconds, and costs $0.02 per job. Send precision: exact when you want the cut to land on the frame you asked for.
Probe the file
Before you upload, probe the file with video inspect. Send frames: false for a probe only, and read the duration, whether it has audio, and the frame size. That is a free check that the file is what you meant to make.
Pick length by the story, not the ceiling
The longest model range in the catalog is a good default only when the story needs it. A product clip that shows one action rarely benefits from 30 seconds, and a shorter clip is cheaper, because Sume bills each model at the provider list rate times 1.25 for each output second. If you are not sure, generate a short proof first at a lower resolution, check composition, and only then spend on the long 1080p version.
Where each number comes from
Treat the three numbers in the table as separate things. The 3 minute figure is a platform limit read from YouTube Help on 2026-10-05. The model ranges are what the Sume docs list today and can change when the catalog changes, so read capabilities from GET /v1/video-router/models before you hard-code a duration in a script. The two do not constrain each other until you want something longer than any single model call.
Longer than one call
If you need a Short longer than one model call, the pieces can be joined. Timeline 1.0 takes up to 200 video slots and an output of up to 1800 seconds, which is more than the 3 minutes a Short can be. Generate the segments, import them as media, and render one MP4 from a plan. Keep each cut tied to one spine so the audio does not jump at joins.
A short checklist
Before you upload, confirm four things: the file is vertical, the duration is under 180 seconds, the audio is what you intended, and the first frame is the one you want a viewer to see. Each one takes a single probe or a still, and all four are free with video inspect.
The takeaway
A Short does not need the model's full range. Pick the shortest model call that tells the story, trim the dead frames, and probe the result.
Sources
Related posts
More in Use cases
- Cancellation save video with an AI avatar: one 30-second clip, one ask
A cancellation-save clip should be 30 seconds and make one offer. At Plus it costs $7.35 to render; send it only to accounts where a human would also reach out.
- Candle holiday ad with AI: a jar photo as IMAGE_REF_0 and a Lyria bed
Make a candle holiday ad: Gemini Omni Flash reference-to-video from your jar photo, a Lyria 3.5 bed and a Timeline render, about $1.48 on Sume.
- Car walkaround clip: a 30-second Seedance 2.5 take for dealer ads
A one-take car walkaround prompt for Seedance 2.5: a 30-second path in 16:9, the Sume price at 720p and 1080p, and what to verify before posting.
- Cart-abandoner retargeting cuts from one hero ad: three trims, $0.06
TikTok's holiday guide says to retarget cart abandoners and video engagers at peak. Cut a 15-second hero into three retargeting clips with video-trim for $0.06.
Written by Sume