Seedance reference video: the 60% rule and the 15-second assumption

If you send a reference video to a Seedance model on Sume, the price is not the output price. Sume bills the output seconds plus an assumed 15 seconds of reference footage, then applies a 0.6 multiplier. For Seedance 2.5 at 720p in 16:9, a 5-second clip with no reference video is $2.889; the same clip with one reference video is $6.9336, about 2.4 times as much.

This rule comes from the Sume pricing package and applies to the whole Seedance family (2.5, 2.0, Fast and Mini), as of 2026-10-08.

The arithmetic step by step

The 15-second figure is an assumption, not a measurement of your file. Sume does not inspect the length of the reference video for this estimate, so a 3-second reference and a 12-second reference are billed the same way.

• Billed seconds: 5 output + 15 assumed input = 20.
• Tokens: 1280 x 720 x 20 x 24 / 1024 = 432,000.
• List before the discount: 432,000 / 1000 x $0.0214 = $9.2448.
• After the 0.6 multiplier: $5.5469.
• After Sume's 1.25 margin: $6.9336.

What this means for planning

The 0.6 multiplier is a discount on the combined total, not on the input alone. As the output gets longer, the fixed 15-second addition matters less. At 30 seconds of output, a reference video raises the billed seconds from 30 to 45, so the ratio against the no-reference price is 45 x 0.6 / 30 = 0.9: the reference clip is actually cheaper than the same output without it.

At 4 seconds, the minimum, the ratio is 19 x 0.6 / 4 = 2.85. That is why the rule bites hardest on short clips.

Where the reference field lives

Seedance accepts reference_video_urls (1 to 3) on the Video Router and Video 1.0 shapes, or a video_url entry in input_references on POST /v1/videos. The Sume docs say the Seedance 2.x models, Wan 3.0, MiniMax H3 and MiniMax H3 Max accept audio and video references, while Gemini Omni Flash 1.1 accepts video references but not audio. Reference images alone do not trigger the video rule in the pricing code, which looks only for reference video fields.

A practical rule of thumb

If your workflow always attaches a reference video, price it with the combined formula and not the plain one. For output clips of 15 seconds or longer the reference adds less than the discount removes, so the longer takes are where a reference-driven workflow is cheapest per output second. For short 4 to 6 second clips, consider image references or a first-frame image instead when they give you enough control, because those skip the extra 15 billed seconds. Always confirm the figure on the first completed job: usage.cost on the poll response is the Sume billable amount.