35-second Omni video on Sume: 10+10+10+5 s clips, cost at 720p

A 35-second video from Gemini Omni 1.1 Flash on Sume is four jobs: three 10-second clips and one 5-second clip, joined by one Timeline 1.0 render. At the Sume 720p rate that is about $4.375 for the clips plus $0.10 for the timeline, roughly $4.48, before any draft passes. Sume does not ship Google's scene extension, so you plan the length yourself.

Google's announcement says Omni 1.1 Flash scene extension analyzes up to 10 seconds of prior context and extends in 10-second steps up to 40 seconds total (Google, read 2026-10-05). The Sume catalog entry gemini-omni-flash-1.1 takes 3 to 10 seconds per job, so a longer piece is several jobs.

The arithmetic

Sume bills the provider list times 1.25 for each output second, as a function of resolution (Sume docs: Video Router, read 2026-10-05). With the provider list rates in Sume's estimator at the time of writing ($0.03, $0.10, $0.15 and $0.30 per second at 360p, 720p, 1080p and 4K), the Sume rates are $0.0375, $0.125, $0.1875 and $0.375 per second. Treat these as a worked example and confirm the live numbers in GET /v1/videos/models and the API pricing page.

Why 10, 10, 10, 5

The cap is 10 seconds per job, and the minimum is 3. Any split that keeps each clip between 3 and 10 works, so 8+9+9+9 is also valid. The longest-first split uses the fewest jobs. Every seam is a cut or a short fade, so fewer seams means fewer places for a character to change.

The timeline rate is $0.10 per output minute rounded up, and the reserve is the audio length divided by 60, rounded up (Sume docs: Timeline 1.0, read 2026-10-05). A 35-second render is one billable minute.

Add the draft pass

A 360p pass of the same four clips costs 35 seconds at $0.0375, which is $1.31. If you keep the first take on each shot, the draft is waste. If you re-roll a shot twice before approving it, the drafts are cheap next to a 720p re-roll. The break-even depends on how often your prompts miss, so count your own misses on a first batch.

What the total does not include

Run the plan call before the render. POST /v1/timeline-1.0/plan is unbilled and returns estimated_cost_usd_micros, so the timeline line of this budget is checkable without spending.

• Re-rolls at any tier. Each is a new paid job.
• A music bed or a voice-over, which the timeline takes as its audio spine.
• A 1080p or 4K final, which changes the rate per second but not the job count.

Join them

Import nothing: a finished Sume job returns a Sume-hosted artifact, and the timeline accepts source_url values that are your workspace's media.sume.com artifacts. Declare audio.duration_seconds as 35 and the four slots at starts 0, 10, 20 and 30 with durations 10, 10, 10 and 5. Transitions go only on slots after the first, last at most one second, and no more than half of the shorter neighbor. A 0.25 second fade is the doc's own example; a hard cut needs no field.

If you would rather not stitch, seedance-2.5 takes 4 to 30 seconds in one job and wan-3.0 2 to 30 seconds (Sume docs: Video generation, read 2026-10-05). That changes the model and its look. It is an answer when the 35 seconds are one continuous move; it is not the same product as Omni.