Seedance 2.5 4K: fal lists 720p, Sume lists 1080p, who offers 4K?

You cannot order 4K from Seedance 2.5 on Sume: the seedance-2.5 row lists 480p, 720p and 1080p. The fal page I read lists 480p and 720p, and ByteDance's page does not state a resolution at all. If you need 4K today on Sume, the model that lists it is gemini-omni-flash-1.1.

The 4K figure for Seedance 2.5 appears in third-party write-ups. I could not confirm it on the vendor's own page, so I would not plan a 4K delivery around it.

What each source says

The fal page lists native 30-second generation, any whole second from 4 to 30, and resolution options of 480p or 720p. The ByteDance page confirms generation up to 30 seconds with the option to extend twice, and it does not say the resolution.

Sume's catalog is the arbiter of what the API accepts. GET /v1/videos/models returns supported_resolutions for each id, and a value outside the list returns 400 unsupported_capability rather than a silent downgrade.

The 1080p gap between fal and Sume

The Sume catalog lists 1080p for seedance-2.5 while the fal page I read lists up to 720p. The catalog is what the Sume API validates against, so a 1080p request is accepted by Sume. Treat provider pages and Sume's catalog as separate documents and read the one you call.

If this matters for a production job, request a short 1080p clip first and check the output dimensions before you scale up.

Ask the catalog, not a blog post

Run this once and keep the output with your project notes.

curl -s https://api.sume.com/v1/videos/models \
  -H "Authorization: Bearer $SUME_API_KEY" \
  | python3 -c "import json,sys; [print(m['id'], m['supported_resolutions']) for m in json.load(sys.stdin)['data']]"

Plan for the 4K you can order

For a 4K master, use Omni and accept its 10-second ceiling, or generate at 1080p with Seedance and upscale elsewhere. The 4K model list shows the resolutions by model. Compare Seedance and Kling on duration, references and audio in this side-by-side.

How to read a spec claim

Launch coverage mixes what a model can do in a lab with what a hosted API exposes. A few habits keep a budget honest.

• Find the vendor page first, then the host page. They can disagree, and the host decides what you can request.
• Look for the field in the request schema, not the adjective in the headline.
• Test a short clip at the resolution you plan to ship, and check the file's real dimensions.
• Write down the date you read each page. Limits change, and the table in this post is dated for that reason.