Seedance 2.5 with a reference video: how Sume prices a 30 s job

When a Seedance 2.5 request includes a reference video, Sume's price estimator counts the output seconds plus 15 assumed input seconds, then applies a 0.6 multiplier to the token cost. For a 10-second 720p 16:9 clip the Sume price moves from $5.78 to $8.67.

This is the estimator in the repo as of 2026-10-07; the final charge is always the usage.cost on the completed job.

The numbers

Reading the table

The surcharge is a fixed 15 seconds of tokens, so it hurts short clips most. A 5-second job rises by about 140 percent, while a 30-second job falls by about 10 percent.

Practical advice

• For motion you can describe, skip the reference video and use images.
• For long jobs the reference video is a small share of the bill.
• Test at 480p first; the same multiplier applies.

Why the estimator assumes 15 seconds

The estimator counts a fixed 15 seconds of input for a reference video, whatever the length of your file, and applies a 0.6 multiplier to the total.

The reserved amount is an estimate. The charge is reported as usage.cost on the poll response.

Cheaper ways to carry motion

If a still pose is enough, use reference images. A video reference makes sense when the motion itself is the point: a dance, a camera move, a product rotation.

Keep video references short. The cap on Sume is 3 per Seedance 2.5 job, and Gemini Omni Flash accepts clips of up to 3 seconds each.

• Compare with and without on 5-second tests.
• Expect the biggest relative increase on short clips.
• Do not assume the multiplier applies to Wan 3.0; its price is per second.