Seedance 2.5 with a reference video: 1.5x a plain 10 s clip

When a Seedance 2.5 request includes a reference video, Sume's quote counts the output seconds plus an assumed 15 seconds of input, then takes 60% of that. A 10-second 720p clip goes from $5.778 to $8.667, a factor of 1.5. A 30-second clip goes down, to $15.60 from $17.33.

What is the rule?

From Sume's pricing code: billable tokens = tokens for (duration + 15 seconds), then x 0.6. The multiplier on a plain clip is (d + 15) x 0.6 / d, which is above 1 for short clips and below 1 once the output passes 22.5 seconds. The 15 seconds is an assumption in the quote, not a measured input length.

What does it cost at 720p?

Does this apply to image references?

No. The multiplier and the 15-second assumption are keyed to reference video inputs only. Reference images and audio do not add to the quote in the estimator. Also note that models differ on inputs: the Video Router docs say the Seedance 2.x models accept audio and video references.

How do I keep the bill down?

• Use a reference image where an image carries the idea; the 5-second case costs 2.4x with a video reference.
• If you need a video reference, make the output longer rather than running several 5-second jobs: per second, 30 s with a reference video is $0.52 vs $0.58 without.
• Draft at 480p first; the same multipliers apply at every tier.

Where is the break-even?

The ratio is (d + 15) x 0.6 / d. It equals 1 at d = 22.5 seconds. Below that the reference-video job costs more than a plain job of the same length; above it, less. At 15 seconds the ratio is 1.2 ($10.40 vs $8.67 at 720p).

The quote counts 15 seconds of input whatever the clip you send is. A 3-second reference is billed as if it were 15, so a short reference does not make the job cheaper.