Media API: images, video, speech, music

The same sk-up-… key and the same base_url that serve text also make images, video, speech and music. There are four routes:

Every model and its toman price is on the models page. If you would rather not write code, the studio does all of this in the browser.

Automatic model picking

Leave model empty, or send one of these ids, and the best-suited model is chosen for the request:

A cheap language model reads the request and picks from our routing table. That decision is billed on its own (usually a fraction of a toman) and reported as uttapen.classifier_cost_toman; the reason is in uttapen.auto. If the classifier fails for any reason, fixed rules decide instead — a request with an attached image goes to an editing model, for example — and the failed classification costs nothing. Name a model yourself and that model always runs.

Billing

• Before a run an amount is held: video at duration × the per-second price of the chosen resolution and sound, images at the per-image price, speech by character count, music at the per-track price.
• After the run only the real cost the model maker reported is charged and the rest of the hold returns. Speech is the exception: its makers report no cost, so it is charged at the model's published price × characters.
• Failed runs are free. An error, a content-policy refusal, a video that misses its deadline or a file that never arrives all release the hold in full.
• Synchronous answers carry X-Uttapen-Cost-Toman and X-Uttapen-Balance-Toman; for video, cost_toman on the job fills in once it completes.

Files and retention

Outputs are stored on our own servers in Iran. Every url in an answer is a signed link on https://api.uttapen.ir/v1/files/… that opens without a key for 24 hours; fetch the job again for a fresh one:

curl https://api.uttapen.ir/v1/uttapen/media/<id> -H "Authorization: Bearer sk-up-..."

Files are kept for 30 days and then deleted, so download what you need. Your recent jobs: GET /v1/uttapen/media?kind=video&limit=20. The text of API requests is never stored.

Content policy

Sexual content, anything harmful involving children, graphic violence, and images or voices made to look like real people are not generated. Such a request is refused before anything is charged, with 400 content_policy_violation. If the model maker itself refuses a request, you get the same code and nothing is charged.

Errors and limits

Limits: 4 images per request, a 4,000-character prompt (5,000 for speech), video up to 20 seconds, and at most 4 reference images of up to 10 MB each (an https URL or data:image/png|jpeg|webp;base64). The general codes are on Errors.