Seedance reference video cost: 2.5, 2.0, Fast and Mini priced

A 10-second 720p Seedance clip with a reference video estimates at $8.67 on seedance-2.5, $5.67 on seedance-2, $4.54 on seedance-2-fast and $2.84 on seedance-2-mini. The same clip without a reference video is $5.78, $3.78, $3.03 and $1.89. Every id uses the same rule; only the rate per 1,000 video tokens differs.

All four ids are in Sume's Video Router catalog, and each accepts reference images, reference videos and reference audio. This post puts the reference-video cost in one table so you can pick the id before you spend on 2.5.

Why does a reference video add so much on a short clip?

Sume does not measure your reference file. When any reference video is present it adds 15 assumed input seconds to the output length, then multiplies the token cost by 0.6. For a 5-second clip that means pricing 20 seconds at 0.6, which is 12 seconds' worth, about 2.4 times the plain price. At 15 seconds it is (15 + 15) x 0.6 = 18 seconds' worth, or 1.2 times. The shorter the clip, the larger the relative jump.

What do the four ids cost with and without a reference video?

Estimates are list times 1.25, 16:9, 720p, rounded up to the cent. Rates per 1,000 video tokens in Sume's pricing table are $0.0214 (2.5), $0.014 (2.0), $0.0112 (Fast) and $0.007 (Mini).

Which id should you test a reference setup on?

Start on seedance-2-mini. It takes the same prompt, image, video and audio reference shape as the others, runs at 480p and 720p, and costs about a third as much as seedance-2.5 per clip. Settle the composition, the prompt and which reference does what, then change only the model field. The Fast vs Mini post covers how the cheaper ids differ.

Move to seedance-2.5 when you need a clip longer than 15 seconds, since the 2.0 family is capped at 15 and 2.5 runs 4 to 30 seconds, or when you need 1080p with the newest model (seedance-2 offers 1080p too). 2.0 Fast and Mini offer 480p and 720p only in the catalog. The 2.5 vs 2.0 post has the other differences.

What does a batch of 100 clips look like?

Multiply the table by your volume. One hundred 10-second 720p clips, each with a reference video, estimate at about $867 on seedance-2.5, $567 on seedance-2, $454 on seedance-2-fast and $284 on seedance-2-mini. Without reference videos the same hundred are $578, $378, $303 and $189. The reference video is the difference between those two sets of numbers, and it is the same 15-second assumption on every id.

If the reference clip is the same for every job, such as a brand camera move, treat it as a fixed per-clip cost: the estimate is built per request from the output length and the presence of a video, not from a stored copy of your file. If only a handful of clips need the motion, use a reference video there and plain text or image inputs everywhere else. Budget with the with-video column for any request that includes one, since the estimate does not look at how long your file is.

For job statuses and what a failed job returns, see the Jobs and results docs.

Is a reference video the only way to carry motion?

For motion and camera work, yes: reference images carry appearance, not movement. If you only need the first frame to match a product shot, use a frame image or a reference image and skip the video. Neither adds anything to the estimate, as the price-effect post shows.

The request shape for a reference video on /v1/videos is an entry in input_references with type: "video_url", per the Video generation docs. On /v1/video-router/generate the field is reference_video_urls. Use public HTTPS URLs on either surface.

What can change these numbers?

Rates, margin and the 15-second assumption are all Sume settings, and the catalog's per-model billable_formula is the source of truth. Different ratios change the pixel count: 9:16 at 720p is 720 x 1280, the same area as 16:9, so its token count matches, while 21:9 at 720p is 1470 x 630 and slightly larger. Re-read the catalog before budgeting a batch.