Docs

Quickstart

One key, one request shape, every image and video model. Submit a job, poll it, get the file.

LiveEvery model on the models page runs on this endpoint. Pictures and video, one request shape; each model’s options and prices are in /data/models.json.

Base URL	`https://api.revana.si/v1`
Auth	`Authorization: Bearer <key>`
Submit	`POST /v1/generations`
Poll	`GET /v1/generations/{id}`
Models	`GET /v1/models` (no key): every model you can generate with right now
Format	JSON in, JSON out. Prices in USD; 100 credits = $1.
Machine-readable	`/openapi.json` · `/llms.txt`

Authentication

Create a key from your account, then send it as a bearer token on every request. Keep it on your server; never ship it in a browser or mobile app.

Submit a generation

POSThttps://api.revana.si/v1/generations

Send the model id and its input. The response comes back straight away with a generation id and status queued; the work happens in the background.

Field	Type	Description
`model`	string, required	A model id from the models page, e.g. `seedance-2.5`.
`input`	object, required	The prompt and settings; see input fields. Unknown fields are refused.
`max_cost_usd`	number, optional	The most you will pay for this generation, up to `100` (the default). Routes priced above it are skipped; if none is left the request answers `422 max_cost_exceeded`.
`max_wait_s`	integer, optional	Routes expected to take longer than this many seconds are skipped. Once it has passed no new route is tried, but one already running is allowed to finish (up to 40 minutes) and is charged if it delivers.
`Idempotency-Key`	header, optional	Up to 200 characters. Send the same key and body again (after a timeout or a `503`) and you get the same generation back, not a second one. The same key with a different body is `409 idempotency_conflict`.

Get the result

GEThttps://api.revana.si/v1/generations/{id}

Poll every few seconds until status is succeeded or failed, then download the files in output.urls with your key: GET /v1/generations/{id}/content for the first, and /content/2, /content/3… when a picture model returns several. Images usually take seconds; video takes from under a minute to several minutes depending on the model and length.

A finished generation:

Input fields

The same names are used across models. Which fields and values each model takes, and its defaults, are listed per model in /data/models.json (options, tasks). If a model doesn’t support a field or value, the request is rejected with 422 invalid_input before anything is charged.

Field	Type	Used for
`prompt`	string	What to generate. Required for text-to-image and text-to-video.
`image_url`	string (URL)	Start frame for image-to-video, or the image to edit for image-to-image.
`end_image_url`	string (URL)	Last frame, for models with first + last frame.
`image_urls`	string[]	Reference images, up to the model’s limit.
`resolution`	string	`720p`, `1080p`, `2K`... one of the model’s resolutions.
`duration`	number	Video length in seconds, one of the model’s durations.
`aspect_ratio`	string	`16:9`, `9:16`, `1:1`...
`audio`	boolean	Generate sound with the video, where the model supports it. May change the price.
`negative_prompt`	string	What to avoid, on models that take it.
`video_url`	string (URL)	A clip to transform, extend or upscale. Its length is measured and, on per-second models, billed.
`audio_url`	string (URL)	The voice or sound a talking-avatar or lip-sync model animates to.
`quality`	string	A quality tier on models priced by one (e.g. `low`, `medium`, `high`).
`seed`	integer	A seed, on models that take one.

Seedance 2.5 defaults to 720p, 5 seconds (4–30), 16:9 and no sound, and also takes:

Field	Type	Used for
`video_url`	string (URL)	A 4–30 s clip to edit. The result keeps its length, so leave out `duration`.
`video_urls`	string[]	Reference clips, up to 10 and 30 seconds in all. Their seconds are billed at the video-input rate.
`audio_urls`	string[]	Reference sound, up to 10 and 30 seconds in all.

image_url is an exact first frame, so it can’t be combined with reference media; use image_urls for a looser visual reference. end_image_url needs image_url. Media must be public https URLs that download directly.

Generation object

Field	Type	Description
`id`	string	Generation id (`gen_…`), used to poll.
`object`	string	Always `generation`.
`model`	string	The model you asked for.
`status`	string	See statuses.
`message`	string	Only while a result is being confirmed with a provider: what is happening.
`quoted_usd`	number	The price on the cheapest route, which is the one tried first.
`reserved_usd`	number	What is held from your balance: the dearest route that may run. The difference is released when it settles.
`output`	object \| null	When succeeded: `urls`, the result files (a picture model can return several), with `width` and `height`; video also has `duration_s` and `has_audio`. Download them promptly; files expire.
`cost_usd`	number \| null	What was charged. Set once the generation succeeds; `0` when it fails.
`attempts`	number	Routes tried, including fallbacks. Only the successful one is billed.
`error`	object \| null	When failed: `code` and `message`.
`created_at`, `completed_at`	string	ISO 8601 timestamps.

Statuses

Status	Meaning
`queued`	Accepted, waiting for a route.
`running`	Generating. If a route fails, the job stays `running` while it moves to the next one, and while a result is being confirmed (see `message`).
`succeeded`	Done. `output.urls` holds the files and `cost_usd` what you paid.
`failed`	Every route failed or the input was refused. Nothing is charged.

Billing

You pay from prepaid credits, 100 to the dollar, bought in packs from $10 to $5,000. Every price is a whole number of credits.
When you submit, reserved_usd is held from your balance. If the generation fails, all of it is released.
Only the route that succeeds is charged, at the price quoted for it when you submitted. Failed attempts on other routes cost nothing.
There is no monthly quota or per-minute cap: your balance and max_cost_usd are the limit.
Each model’s price is on its page and in the price table.

Errors

Errors use HTTP status codes and a JSON body: {"error": {"code": "...", "message": "..."}}.

HTTP	Code	When
400	`invalid_request`	The body isn’t JSON, is missing `model` or `input`, or has a field the API doesn’t know.
401	`unauthorized`	Missing or invalid API key.
402	`insufficient_credits`	Your available balance doesn’t cover `reserved_usd`. Top up and retry.
403	`account_disabled`	The account is paused. Contact support.
403	`generation_not_enabled`	Your key can read prices but can’t generate yet.
404	`not_found`	No generation with this id on your account.
409	`idempotency_conflict`	This `Idempotency-Key` was already used with a different body.
409	`not_ready`	Asked for the file before the generation succeeded.
413	`body_too_large`	The body is over 1 MB. Send media as URLs.
422	`invalid_input`	A field, value or combination the model doesn’t support, or media that can’t be read. Nothing is charged.
422	`max_cost_exceeded`	Every route costs more than `max_cost_usd`.
422	`model_not_available`	No model with this id on the API. Ids are on the models page.
429	`rate_limited`	Too many requests in the same second from one account (a guard against runaway loops, not a quota). Wait the `Retry-After` seconds.
429	`concurrency_limit`	The account already has its maximum of generations in progress. Retry when one finishes.
503	`restarting`, `busy`	A restart or a moment of load. Retry after `Retry-After` with the same `Idempotency-Key`.
503	`model_unavailable`	No route for this model is up right now. Retry shortly.
500	`server_error`	Something broke on our side. Safe to retry with the same `Idempotency-Key`.

Model list

Every model, its options and prices are published as JSON at /data/models.json, the same file this site renders from.