Requests and media

Input media

Pass images, video and audio to a model by HTTPS URL.

Models that take media (an image to edit, a first frame, a clip to restyle, speech to lip-sync) take it as a public HTTPS URL in the request body.

Field	Used by
`image_urls`	Generate an image: reference or edit images, up to 16
`image_url`, `end_image_url`	Generate a video: first and last frame
`video_url`	Video-to-video models: the clip to edit
`audio_url`	Lip-sync models: the speech to say
`reference_image_urls`, `reference_video_urls`, `reference_audio_urls`	Video models that take references

Each URL must be downloadable without cookies or custom headers, and must stay valid until the job finishes. A signed URL from your own storage (S3, GCS, R2) works if it lasts longer than the job.

A model that does not take a field rejects it with 400 invalid_request. The model's row in its list says what it accepts (capabilities).

Direct upload

If the file is not online yet, upload it into Sume storage with your key, in three calls:

POST /v1/assets/upload-url with content_type and size_bytes (and optionally filename). The 201 returns data.asset.id and data.upload: a presigned url, its method (PUT), the headers to send with it, and expires_at.
Send the bytes to data.upload.url with that method and exactly those headers. Do not send your API key there.
POST /v1/assets/{id}/complete with the same size_bytes. The 200 returns the asset with status: "ready" and its url on media.sume.com.

Pass that url as image_url, video_url or audio_url like any other HTTPS URL. Uploading and completing are free. An upload URL that expires before you complete it must be created again.