Seedance pixel sizes by ratio on Sume: 864x496 at 480p, 1920x1080

A Seedance clip on Sume is priced from a fixed pixel size for each resolution and aspect ratio. 16:9 at 480p is 864 x 496, not 854 x 480; at 720p it is 1280 x 720 and at 1080p 1920 x 1080. The six ratios are 21:9, 16:9, 4:3, 1:1, 3:4 and 9:16. If a Sora client requested a size such as 1280x720, Sume returns 400 unsupported_parameter on /v1/videos and wants resolution plus aspect_ratio instead; this table says what you get.

OpenAI's Sora Videos API ended on 2026-09-24 (Pondero). Anyone with a downstream layout that assumed one fixed size now needs the real numbers.

The size table

These are the dimensions in Sume's Seedance pricing code, which it uses to count video tokens. If a ratio is missing from a request, the pricing code treats auto as 16:9.

Things to notice

Every 480p dimension is a multiple of 16 (432, 496, 560, 640, 752, 864, 992). That is why 16:9 at 480p is 864 x 496 and not 854 x 480. At 720p, 21:9 is 1470 x 630, which is wider than the 1280-pixel frame an editor might expect. And 3:4 and 9:16 are the 4:3 and 16:9 sizes turned on their side, so a vertical 720p clip has exactly the pixels of a horizontal one.

What it means for cost and layout

Because price tracks pixels x seconds, 9:16 and 16:9 cost the same at the same resolution and duration. A 4:3 720p clip has 927,408 pixels per frame against 921,600 for 16:9, close to the same. 21:9 at 720p is 926,100. The ratio does not move the price much; the resolution does, since 1080p has 2.25 times the tokens of 720p.

For layout, treat the numbers as the delivered size to design against, and verify a delivered file with video inspect rather than trusting a table. Sora-era crops and safe zones were tuned to different sizes, so re-check any text overlay positions.

Which ratios a given model lists

Not every model supports every ratio. The docs' example catalog entry for seedance-2 lists the six above, while gemini-omni-flash-1.1 takes only 16:9 and 9:16. Read supported_aspect_ratios from GET /v1/videos/models for the model you pin. The platform vocabulary is wider (3:2, 2:3, 9:21 also exist), and each model reports its own subset.

Cost at each size for 8 seconds

Using the same token formula, an 8-second 16:9 clip at 480p has 864 x 496 x 8 x 24 / 1024 = 80,352 tokens, at 720p 172,800 tokens, and at 1080p 388,800 tokens. At the seedance-2 rate of $0.014 per 1,000 tokens that is about $1.12, $2.42 and $5.44 at list, or $1.41, $3.02 and $6.80 at Sume's 1.25 multiple. Resolution is the lever; ratio is nearly free.