Does aspect ratio change Seedance 2.5 price? Barely: under 1%

Aspect ratio barely changes the price of a Seedance 2.5 clip on Sume. The model bills video tokens, and the token count comes from the output width times height, so 16:9, 9:16 and 1:1 land on the same pixel count at 720p and cost the same $2.89 for 5 seconds, while 21:9, 4:3 and 3:4 sit about 0.7% higher.

The only way ratio moves a bill by a visible amount is at 480p with a square, which is about 4.4% cheaper than the other ratios. If you are choosing between vertical and landscape to save money, you are choosing on the wrong axis; resolution and duration are the real levers. The formula comes from Sume's price tables on main, and the supported ratios and resolutions come from Video Generation.

How the token count works

Sume's Seedance estimate is width x height x seconds x 24 fps / 1024 video tokens, multiplied by the per-1000-token list rate and the 1.25 house margin, then rounded up to the cent for the job. Seedance 2.5 lists at $0.0214 per 1000 tokens, or $0.0234 at 1080p. Each resolution and ratio pair maps to a fixed output size, which is why the ratio changes the price only through the pixel count.

At 720p the sizes are 1280x720 for 16:9, 720x1280 for 9:16, 960x960 for 1:1, 1470x630 for 21:9, and 1112x834 for 4:3 (834x1112 for 3:4). The product of each pair is within 0.7% of 921,600, because the sizes are chosen to hold the total pixel budget roughly constant.

Price by ratio for a 5 second clip

The table uses 5 seconds because it is a common ad length. Multiply by 3 to get a 15 second estimate and round up to the cent yourself, since rounding happens once per job.

What this means for a mixed-ratio batch

A campaign that needs the same hook in 9:16, 1:1 and 16:9 can budget three identical Seedance 2.5 720p jobs at $2.89 each, $8.67 in total for 5 second versions. Adding 21:9 adds $2.91. You do not need separate cost columns per ratio, only per resolution.

Where it does matter is the wallet check. A job is admitted only if the balance covers its estimate, so a wallet holding exactly $2.90 accepts the 16:9 job and refuses the 4:3 one at $2.91. That is a rare edge, but it is the kind of one-cent surprise that shows up as 402 insufficient_credits in a loop that submits ratios in sequence; the rules are in Generation admission.

Where ratio does change the price

Other rows are not token based. Wan 3.0 and Kling 3 bill per output second, so ratio never enters their price, and Wan accepts a different set of ratios than Seedance. Choose the ratio the placement needs, then compare rows on resolution and seconds.

If a model rejects the ratio you want, the request fails before a reservation is taken, so the choice costs nothing to test. Check the ratio against the limits on the model page before you plan the batch.

A practical habit is to keep one price per resolution in your batch sheet and a single line of exceptions: 480p square is cheaper, and the wide or 4:3 shapes add a few cents. That keeps the sheet short and matches how the estimator behaves, since every ratio at a given resolution lands within about 0.7% of the same pixel count. It also keeps reviews fast, because a reader can see at a glance that a 9:16 cut and a 16:9 cut of the same shot are budgeted identically.

• Same pixel budget means the same price at 720p and 1080p for 16:9, 9:16 and 1:1.
• 21:9, 4:3 and 3:4 cost 2 to 5 cents more at 720p and 1080p for 5 s.
• Duration is linear, resolution is not: 1080p is about 2.5 times 720p.