Video generation

Same endpoint as images, two more fields: how long, and how sharp. Video is priced per second, so those two fields are the invoice.

Endpoint

POST /api/generate

Request body

Field	Type	Required	Notes
`kind`	string	yes	Use `"video"`.
`prompt`	string	yes	Checked by the safety guard before anything is priced or run.
`model`	string	yes	One of the ids below.
`seconds`	number	no	Clamped server-side to the model range. The clamped value is what gets priced and what gets run.
`quality`	string	no	The resolution: `"480p"`, `"720p"` or `"1080p"`. Defaults per model.
`aspect`	string	no	`"16:9"` or `"9:16"`, snapped to what the model supports.

Resolution goes in quality. There is no resolution field. Send one and it is ignored without complaint — you get the model default and the price that goes with it.

Short clips can come back longer than you asked for. Some models only produce fixed durations, and a request below their shortest clip is served at that shortest clip. A three-second request on a model whose minimum is six seconds returns roughly six seconds of video. Ask for a duration the model actually offers if the exact length matters to you.

Models, durations and cost per second

Read straight from the pricing table the billing code uses. Total credits = seconds × the rate for the resolution you chose.

Model id	Seconds	Default	Credits per second
`wan`	5–10	5s · 720p	720p 10.11080p 15.1
`kling`	5–10	5s · 720p	480p 8.5720p 8.51080p 13
`kling-3-pro`	3–15	10s · 720p	480p 28.1720p 35.11080p 56.1
`seedance`	4–15	5s · 720p	480p 9.6720p 20.61080p 51.1
`seedance-2-fast`	3–12	5s · 720p	480p 8720p 16.61080p 51.1
`seedance-2-5`	4–30	5s · 720p	480p 24720p 51
`seedance-lite`	3–12	5s · 720p	480p 2720p 31080p 6
`sora`	4–20	5s · 720p	480p 7720p 121080p 67
`veo`	4–8	5s · 720p	480p 15.1720p 181080p 56
`veo31`	4–8	6s · 720p	480p 40.1720p 40.11080p 40.1
`hailuo`	6–10	6s · 720p	480p 5720p 51080p 8.2
`ltx`	6–10	6s · 1080p	1080p 6.1
`kie-wan-2-5`	5–10	5s · 720p	480p 6.1720p 6.11080p 10.1
`kie-hailuo-02`	6–10	6s · 720p	480p 3720p 51080p 8
`kie-seedance-2`	3–15	5s · 720p	480p 12720p 20.61080p 51.1
`kie-seedance-2-fast`	3–12	5s · 720p	480p 10720p 16.61080p 51.1
`kie-seedance-2-mini`	3–12	5s · 720p	480p 6720p 10.31080p 20.6
`kie-seedance-lite`	3–10	5s · 720p	480p 4720p 61080p 11
`kie-kling-3-turbo`	5–10	5s · 1080p	480p 9.1720p 9.11080p 11.3
`kie-wan-2-6`	5–10	5s · 720p	480p 7.1720p 7.11080p 10.5
`kling-25-turbo`	5–10	5s · 720p	480p 8.6720p 8.61080p 8.6

Example

curl https://katama.ai/api/generate \
  -H "Content-Type: application/json" \
  -d '{
    "kind": "video",
    "prompt": "a red paper crane slowly rotating on a grey table",
    "model": "ltx",
    "seconds": 6,
    "quality": "1080p"
  }'

Holds: what happens to your credits

Video takes time, so the credits are not spent up front — they are held. While the run is in flight the balance shows the hold, and creditsCharged is still 0. When the video lands, the hold settles into a charge. If the run fails, the hold is released and nothing is charged.

# during the run
{ "balance": 2787, "holdsActive": 13 }

# after it succeeds
{ "balance": 2787, "holdsActive": 0 }   →  creditsCharged: 13

Polling

The first response usually carries status: "queued" or "running" with an empty outputUrl. Poll the generation until status is "succeeded" or "failed". A short clip typically settles in well under a minute; longer ones take proportionally longer.

Next: Image generation · Plans & credits