Shortest valid clip on each Sume video model and what it costs

The cheapest single clip you can order on Sume is a 2-second Wan 3.0 clip at 480p, which is 2 x $0.0625 = $0.125. The most expensive minimum among the six video models listed here is a 4-second Seedance 2.5 clip at 720p at $2.3112. Each model has its own valid duration range, and a request below a model's minimum is out of range (Seedance 2.5 bills a minimum of 4 seconds).

The minimum-clip table

Video prices are per output second and linear in duration inside each model's valid range. Multiplying the model's lowest valid duration by the per-second price gives the floor for one clip.

Why this matters for tests

Teams that run many throwaway drafts tend to pick a length first and a model second. The table reverses that: if the thing being tested is a prompt or a reference image, the 2-second Wan clip answers it for $0.125, and a Seedance 2.5 draft of the same prompt costs $1.0747 at 480p, about 8.6 times as much (1.074708 / 0.125 = 8.6).

If the test needs 5 seconds of motion, Wan 3.0 at 480p costs 5 x 0.0625 = $0.3125, the same as MiniMax H3 at its 5-second floor and 480p. MiniMax H3 and H3 Max share that $0.3125 floor at 480p; at 768p the floors split to $0.375 and $0.50.

Scaling a minimum-clip test

Say you want to test 50 prompts before spending on the real shoot. At each model's cheapest size, 50 minimum clips cost $6.25 on Wan 3.0 (50 x 0.125), $5.625 on Omni Flash 1.1 (50 x 0.1125) and $53.74 on Seedance 2.5 (50 x 1.074708).

At the mid sizes the gap is wider: 20 prompts at 720p cost $5.00 on Wan (2 s), $7.50 on Omni Flash (3 s) and $46.224 on Seedance 2.5 (4 s). A short test on a cheap model tells you whether a prompt is broken; it does not tell you how a pricier model will render it. Use the cheap floor to reject bad prompts, then pay for the model you will deliver with.

One caution: a 2-second clip tests framing and subject, not motion over time. For a motion test use the model's 5-second price, such as 5 x $0.0625 = $0.3125 on Wan at 480p.

Check the live range before you hard-code it

The ranges above come from the catalog and the video models doc. Gemini Omni Flash 1.1 also accepts only 16:9 or 9:16 and always carries native audio, so its $0.1125 floor already includes sound. Read GET /v1/videos/models at run time instead of copying these numbers into code.

Reading the range at run time

Do not copy the minimums into code. Ask the API.

• Call GET /v1/videos/models and read each model's duration range.
• Clamp the requested duration to that range before you estimate.
• Estimate with the clamped value, so the number you show matches the number Sume will reserve.