2026-09-26Models

AI video generation API length limits: up to 30 seconds per clip

On Sume, seedance-2.5 (4–30 s) and wan-3.0 (2–30 s) make 30-second clips; other models stop at 15 s or less. Limits by model, the error, and going longer.

4 min readSume

All posts

In the Sume video generation API, seedance-2.5 (4–30 seconds) and wan-3.0 (2–30 seconds) accept a 30-second duration. Every other model in the catalog stops at 15 seconds or less: 15 for most, 10 for gemini-omni-flash-1.1. Each model lists the whole-second lengths it accepts in supported_durations on GET /v1/videos/models.

The ranges come from the Video generation docs and the catalog behind GET /v1/videos/models, read on 2026-09-26.

What is the longest clip each video model can make?

Limits are not uniform, so check the model before you submit. supported_durations lists every whole second from the model's minimum to its maximum. On the legacy Video Router the same ids carry the same limits, as a duration_seconds minimum and maximum under capabilities in GET /v1/video-router/models.

Clip lengths from `supported_durations` in the catalog behind `GET /v1/videos/models` and Video generation, read 2026-09-26. Confirm with `GET /v1/videos/models` before you submit.
Model id	Duration (whole seconds)
`seedance-2.5`	4–30
`wan-3.0`	2–30
`seedance-2`, `seedance-2-fast`, `seedance-2-mini`	4–15
`kling-3`	4–15
`grok-imagine-video-1.5`	4–15
`minimax-h3`, `minimax-h3-max`	5–15
`gemini-omni-flash-1.1`	3–10

How do I request a 30-second clip?

Send duration as an integer number of seconds to a model that lists it. This adapts a docs prompt to a 30-second seedance-2.5 clip at 720p:

The call returns 202 with a job id and a polling_url. Poll until status is completed, then download from unsigned_urls[0].
Send Idempotency-Key to make retries safe: a replay returns the original job.

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: t2v-30s-001" \
  -d '{
    "model": "seedance-2.5",
    "prompt": "A time-lapse of a flower blooming",
    "resolution": "720p",
    "duration": 30
  }'

What happens if I ask for a length the model does not list?

POST /v1/videos refuses the request with a 400, the error code unsupported_capability, and a message such as “kling-3 does not support duration 20.” The error details include the model's supported values, so you can correct the value or pick a model whose range covers it.

One mode takes no length at all: Gemini Omni Flash 1.1's edit mode does not send duration to the model, and a value you pass only sizes the reserve estimate. See Edit a video with a prompt.

Does a longer clip cost more?

On per-second models, yes. Wan 3.0, the MiniMax H3 models, and Gemini Omni Flash 1.1 list per-video-second-<resolution> rates, and kling-3 and grok-imagine-video-1.5 list a per-video-second rate, so each added second adds to the price. Seedance models price per 1,000 video tokens. Every rate is the provider's list price × 1.25, reserved on submit, plus a 5.5% agent fee by default. Read the rates in pricing_skus or see API pricing.

How do I make a video longer than 30 seconds?

Generate it as several clips within each model's range, then assemble them with Timeline 1.0, which joins an audio spine and ordered video[] clips into one MP4. The steps are in How to assemble a long-form video with the Timeline 1.0 API. Carrying a shot from one clip into the next is covered in Get the last frame of a video to chain clips.

Sources

Written by Sume