Longest AI video Sume can render: 1800 s from 60 clips

The longest single video Sume's Timeline API renders is 1800 seconds, which is 30 minutes. Built from the 30-second clips that seedance-2.5 and wan-3.0 accept, that is 60 clips.

A single generation is much shorter. This post is about the assembled ceiling, not the one-shot ceiling.

Three numbers decide the ceiling

From the Timeline 1.0 docs: audio.duration_seconds is 1 to 1800, video[] has 1 to 200 slots, and each slot must be at least 0.2 seconds long. The audio spine sets the output length, so the output cannot exceed 1800 seconds even if 200 slots would cover more.

Price of the assembly

Timeline renders are $0.10 per ceil(output minute). A full 1800-second render is 30 minutes, so $3.00 for the render. The docs confirm the live rate is in GET /v1/catalog. The video jobs that make the clips are billed separately, per model.

There is no provider inference in the render step, only worker ffmpeg. So the cost of a long video is almost entirely the clips.

Do the join safely

With more than 12 slots the render chunks by default. Do not force render.strategy: single, which is refused above 12 slots. Keep fades to 8 in a row. Use POST /v1/timeline-1.0/plan first; it is unbilled and returns segment_count, billable_minutes and estimated_cost_usd_micros.

Remember the other ceilings in the media chain. Video trim accepts a source up to 1800 seconds but an output only up to 900 seconds, so a 30-minute master cannot come back through trim in one cut.

A realistic long video

A 10-minute video from 30-second clips is 20 clips and 20 video jobs, then one render of ceil(600/60) = 10 minutes, so $1.00 for the join. The practical limit is rarely Timeline. It is keeping a character or a set consistent across 20 independent jobs, which is why chaining the last frame into the next first frame is the usual technique.

Where a long build usually breaks

Three things break first. The first is cost drift: 60 jobs reserve 60 amounts, so check balance before you fire the batch and keep each submit idempotent with a stable key. The second is look drift: independent jobs do not share state, so use the same model id and the last-frame chain. The third is frame-rate mismatch: if clips have different rates, set output.fps to one of 24, 25, 30 or 60 and expect a warning if it resamples.

Imports matter too. Every clip has to be a media.sume.com file in your workspace before it can be a slot, so import generated videos first with POST /v1/media-imports.

A plan call for the maximum

Run POST /v1/timeline-1.0/plan with audio.duration_seconds: 1800 and your slot list. The reply gives segment_count, billable_minutes (30 for a full render) and estimated_cost_usd_micros, so a $3.00 render shows as 3000000 micros. It does not create a job or reserve credits. Fix any refusal at the plan stage, where it is free.

Short answer for a brief

If a client asks how long a Sume video can be, answer in layers. One generation: up to 30 seconds on seedance-2.5 and wan-3.0, 15 seconds on most other ids. One assembled file: up to 1800 seconds from up to 200 slots. One media step on an existing file: it depends on the step, from 300 to 1800 seconds.

Quote the render price too: $0.10 for each started output minute, plus the clips.

Before you build

Before you build, read the linked Sume docs page for the exact request fields, limits and prices, because those pages are the source of truth and can change. Run one short, cheap test with your own material first, check the output in a player and in your editor, and only then scale to the full shot list. Keep every job id and file you approve, so a later change never forces you to regenerate work that was already signed off. Note that this post describes Sume's catalog and tools; Sume does not run Luma Ray 3.2, and nothing here claims HDR or EXR output.