Seedance Reference Video Costs More: A 10-Second Clip Goes Up 50%

Seedance is the model family most often used with reference clips: ByteDance says Seedance 2.5 accepts up to 10 videos, 30 images and 10 audio clips (read 2026-10-03). Image and audio references do not change the price in Sume's pricing code. A reference video does, and by an amount that depends on how long your output is.

The rule in Sume's pricing code

Seedance is billed on video tokens: width times height times seconds times 24 frames, divided by 1,024. When a request carries a reference video, Sume's estimator adds an assumed 15 seconds of input to the output seconds, prices the total tokens, and then multiplies by 0.6. The input length is assumed, not measured, so a 3-second reference and a 15-second reference are estimated the same way.

The effect: the cost multiplier is (output plus 15) divided by output, times 0.6. That is 2.4 times at 5 seconds, 1.5 times at 10 seconds and 1.2 times at 15 seconds.

Worked example on seedance-2

At 720p 16:9, seedance-2 is about $0.378 per second on Sume without a reference video (provider list $0.014 per 1,000 tokens times 21,600 tokens per second, times 1.25).

What to do with it

Prefer fewer, longer outputs when a video reference is essential: the fixed 15-second charge is spread over more output seconds. If a reference image can carry the look, use it instead, since images add no tokens. For motion, a first and last frame is cheaper than a reference clip.

These figures are estimates computed from the pricing code, so confirm against the amount reserved on your own job, which is reported as usage.cost once the job completes. See which model fits the inputs you have.