480p draft, 1080p final on Seedance 2.5: no seed, so a new take
Seedance 2.5 on Sume accepts 480p, 720p and 1080p, but no v1 model takes a seed. A cheap draft previews the idea, not the shot. How to use drafts honestly.

A 480p draft on Sume will not become your 1080p final: no v1 video model accepts seed, so each resolution is a fresh generation. Per the docs, seedance-2.5 accepts 4 to 30 seconds at 480p, 720p and 1080p, so use the draft to test a prompt and framing, and treat the final as a new take you review again.
This matters now that 30-second clips are common, because a wasted long render is the costly mistake. Facts below are from the video generation docs, read 2026-10-02.
Why can't I reuse the draft's randomness?
The docs state that no v1 model accepts seed; each reports seed: false and rejects the field. A request that includes seed fails, so there is no way to ask for the same shot at a higher resolution.
How should a draft and a final differ?
Change only resolution between the two calls and keep the prompt, duration and aspect ratio identical so you are comparing like with like.
| Setting | Draft | Final |
|---|---|---|
model | seedance-2.5 | seedance-2.5 |
resolution | 480p | 1080p |
duration | Same, 4 to 30 s | Same |
seed | Not accepted | Not accepted |
Idempotency-Key | Its own key | A different key |
What does a draft tell me?
Whether the prompt is read the way you meant, whether the subject and camera move make sense, and whether the length is right. It does not tell you the final's exact framing or motion. Pricing is shown per model in pricing_skus from the catalog; read it before choosing.
When should I skip the draft?
For short clips where one extra take is cheap, go straight to the final and keep the one you like. Use drafts when the clip is long, the prompt is new, or a client needs to approve direction first.
Sources
Related posts
More in Use cases
- Accessible transcript for a video page: Sume transcript, then review
Get a timed transcript of your video from Sume's video inspect for $0.01 a minute, then correct it by ear. Silent clips fail, and no output is pre-certified.
- Turn a finished ad still into a short clip with a first frame
Send the approved still as frame_images with frame_type first_frame on POST /v1/videos. Which models accept it, and how to pick a duration, from Sume's docs.
- ADA Title II April 26, 2027: caption course videos with an API
ADA.gov sets WCAG 2.1 AA with a 2027 deadline for larger public bodies. See how to burn captions into lecture clips with Sume and what it leaves undone.
- Add a cold open to a YouTube video: move the best 10 seconds first
Put your best 10 seconds at the start of a video: detach audio once, then one Timeline 1.0 render with audio.parts and source_in. Plan call is free.
Written by Sume