Seedance, Wan and GPT Image behind one prepaid key. Send a prompt, some reference frames and a duration; get a finished render back. Every price on this page is the price the API quotes you before the job runs, and you can check it yourself with a dry run that costs nothing.
Prepaid in US dollars. No seat licence, no minimum commitment, no demo call.
Rates, out in the open
One row per model and mode. The rate is what you pay; there is no platform fee stacked on top of it.
Each row says how its rate was arrived at. measured is taken from real paid renders. published is the generation service's own price list, not yet re-measured here. provisional is derived from a sibling model and awaiting its first live render.
Seedance 2.5Reference to video
Billed per token
480p · 720p · 1080p · 4–30 s · up to 9 images · 30 s of reference video
With a reference video
$8.99 per million tokens
Without a reference video
$14.98 per million tokens
Tokens per output second, with a reference video: 480p 18,601 tokens/s · 720p 41,820 tokens/s · 1080p 41,820 tokens/s
Tokens per output second, without a reference video: 480p 9,707 tokens/s · 720p 21,824 tokens/s · 1080p 21,824 tokens/s
Upscale add-on at 1080p: +$0.046 per output second
A typical job (5 s at 480p from one reference image):$0.73
One request produces one image. The charge is the price above plus $0.02 for every reference image you send, and an edit sends between 1 and 16 of them, so the price above is never the whole charge.
A typical job (One 1k edit at medium quality from 1 reference image):$0.04
measured
Check the arithmetic
Three real quotes, with the multiplication shown. Run the same request with dry_run and you should get the same number back.
Seedance 2.0 Mini
Reference to video
15 s at 480p with a 14.56 s reference video
29 billable seconds × $0.05 per second
$1.45
Wan 2.6 Flash
Reference to video
10 s at 1080p with generated sound
10 output seconds × $0.10 per second
$1.00
GPT Image 2.5
Image edit
One 1k edit at high quality from 3 reference images
$0.05 for the image + 3 × $0.02 per reference image
$0.11
How billing works
Prepaid, in US dollars
You top up a balance and jobs draw it down. No invoice, no seat licence, no monthly minimum. Accepting a live job places a hold of 1.25× its quote on your balance while it runs, and a job whose hold is more than your available balance is refused before anything starts, with nothing reserved. The balance a job needs is its quote plus that bit held on top.
Quoted before, charged after
Send the request with dry_run and you get the quote with no job created and nothing charged. When you run it for real, the hold is replaced by the final charge, which follows the usage the generation service reports at the rate published above, so it can land a little either side of the quote.
Reference video length: free on some models, billed on others
On the token-billed models (Seedance 2.5) attaching a reference video roughly doubles the tokens counted per output second, but its length is never billed: a ten-second render costs the same with a three-second reference as with a thirty-second one. On the per-second reference-to-video models (Seedance 2.0 Mini and Seedance 2.0 Fast) it is the opposite: you are billed for reference seconds plus output seconds, subject to the floor listed on those rows. A long reference genuinely costs you more there, and the quote will say so.
Whole cents, and a floor
Charges are whole US cents, rounded at the quote. Any job the generation service actually started is charged at least $0.10, so a render that fails halfway costs that minimum rather than its full quote. A job that never started, or whose output could not be delivered, is not charged at all and its hold is released.
The whole integration
One POST per job. Keep dry_run on while you are wiring it up: the request is validated and quoted, no job is created and nothing is charged. Drop it to submit, then poll the job or take a signed webhook.
The dry run answers with estimate_cents and the token or second count behind it, so you can reconcile it against the rates above before you spend anything. Test keys begin sk_test_ and are allowed dry runs only.
Not open to everyone yet
We are issuing keys by hand while the preview settles, so tell us what you want to build and roughly how much you expect to render. Rates on this page are the rates you get; there is no negotiated tier hiding behind the form we have not built.