Latest in AI
Seedance 2.5 US Draft Mode on fal: API, Pricing and the 14-Day Limit
The US-hosted API now separates choosing a shot from paying for its final render, with a few constraints developers need to handle.
fal added draft generation to its US-hosted Seedance 2.5 API on October 30. Developers can preview a video at 480p, then complete the selected draft at 1080p within 14 days. The useful change is deciding which shot deserves a final render before committing to the higher-resolution output.
The official announcement covers text-to-video, image-to-video and reference-to-video workflows. This is an addition to fal’s US endpoints, not the first release of Seedance 2.5 or of draft generation across every provider.
How the Seedance 2.5 US draft API works
The text-to-video API schema exposes a draft boolean. Enabling it produces a 480p preview and overrides the normal resolution setting. The response includes a draft_id, which must be completed using the same account within 14 days.
Pass that identifier to bytedance/seedance-2.5/us/draft/complete to request the 1080p result. Keep the draft identifier alongside the preview, creation time and originating account rather than storing only the video URL. Otherwise, a usable-looking preview in your application can lose its path to completion.
This is particularly relevant for products with an approval queue. If a client takes two weeks to review a clip, the application should warn about the deadline before approval, not discover the expired draft when the final render is requested.
Seedance 2.5 US pricing
fal’s generation page lists approximate per-second rates alongside token-based billing. These are estimates for the listed output settings, not fixed prices for every possible reference-video request.
| Output | Listed approximate cost per generated second |
|---|---|
| 480p | $0.2205 |
| 720p | $0.4730 |
| 1080p | $1.164 |
The US draft-completion page lists roughly $1.164 per generated second, or $0.0234 per 1,000 tokens, for 1080p completion. It also explains that reference-video inputs affect the calculation and that input and output video are both charged when video references are used.
A draft is a paid selection step, not a free preview. BytePlus’s underlying draft documentation describes the draft and final tasks as independently billed. Its 480p draft uses the normal 480p token rate, and final completion supports 1080p.
When drafting saves money
Using the displayed estimates, five five-second drafts followed by one five-second final would cost about $11.33, before any reference-video effects. Five direct five-second 1080p generations would be about $29.10. That is an illustrative calculation from the listed rates, not a measured saving from testing.
The comparison reverses when every shot already works. A five-second draft plus its final is about $6.92 at those estimates, compared with $5.82 for one direct final. Drafting pays off by avoiding high-resolution rejects; it adds a step when there are no rejects to avoid.
For a production interface, the important design decision is therefore where to put human approval. Let users judge composition and motion in the draft, make the remaining cost visible, and reserve completion for the clips they actually intend to use. The new endpoint is most useful when the application changes its workflow around that decision rather than treating draft mode as another checkbox.
I’m Paras. I work at the intersection of AI, engineering and distribution.
I help technical founders turn API launches and product updates into clear stories developers can understand and share.