Inputs

A media request is a JSON body of named inputs. The names are shared across models of a modality, so prompt, width or duration_seconds mean the same thing on every model that accepts them. Which inputs a model accepts, which are required, and their ranges, enums and defaults are listed on that model's page under Models.

An input the modality does not define is refused with 400 invalid_request, and param names it:

{
  "error": {
    "code": "invalid_request",
    "message": "...",
    "param": "colour"
  }
}

Files are https URLsCopy Anchor Link

Inputs that carry media (image, images, mask, first_frame, audio, video, reference_images and the like) take an https URL, or a list of them. The file is fetched from that URL when the job runs.

• The URL must be absolute and use https. http, data: and other schemes are refused with 400 invalid_request.
• It must be reachable from the public internet without cookies or extra headers. A presigned link from your own storage works; give it an expiry long enough for the job to start.
• There is no upload endpoint. Put the file in your own storage first.
• The whole request body, URLs included, is capped at 25 MB.

curl https://api.opinference.com/v1/audio/transcriptions \
  -H "Authorization: Bearer $OP_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "cohere/transcribe",
    "audio": "https://example.com/interview.mp3",
    "language": "en"
  }'

Input namesCopy Anchor Link

Each table lists every name the modality defines. A model accepts a subset.

ImageCopy Anchor Link

VideoCopy Anchor Link

SpeechCopy Anchor Link

TranscriptionCopy Anchor Link

MusicCopy Anchor Link