Seedance 2.5 reference video price on Sume: the 15-second assumption

A reference video changes what a Seedance 2.5 job costs on Sume, and not in the direction you might expect for long clips. The repo's pricing code assumes 15 seconds of input video, adds those seconds to the token count, then multiplies the result by 0.6. Short outputs get dearer; a 30-second output gets cheaper.

The numbers

Reference images and reference audio did not change the estimate when tested; only a reference video did. These figures are billed amounts for 720p, 16:9, from the repo's Seedance estimator.

Why the ratio falls

The formula is (output seconds + 15) divided by output seconds, times 0.6. At 5 seconds that is 4 times 0.6, or 2.4. At 30 seconds it is 1.5 times 0.6, or 0.9. The 15-second figure is an assumption, not a measurement of your upload, so the hold does not change if your reference clip is 3 seconds or 12.

What to do with it

• If you need a short clip guided by a reference video, expect two to three times the plain price.
• If you can make the output longer anyway, longer clips amortize the assumed input.
• Do not rely on this for other models; only Seedance 2.5 was priced this way here.

Hold versus final cost

The estimator quotes the amount that is held at submit. Because the assumed input length is a constant rather than a measurement, a short reference clip does not make the job cheaper and a long one does not make it dearer. If you want a cheaper job, change the output length or drop the reference video, not the length of the upload.

To check a real job, submit it and read its usage row. The poll response's usage.cost is the billable amount, and the docs describe it as the Sume billable amount for the job.

Hold sizes are reserved at submit, so the reservation is the number in the table. Re-check the live catalog before a large batch.