inspect media
Look at media, or ask a model about it: a library item, or a picture or video at a web address (urls) read without keeping it. An image returns the picture with its dimensions. A video returns frames: overview (one storyboard grid of the whole file, default), window (a grid of one span), frame (one moment), strip (one span as frames with the waveform and words under them), or shot_log (every segment in order with its span, frame, spoken lines and a written line each; cloud only, one video). spoken adds the sentences in the span where the file has a transcript; cuts: true adds every shot change and shot length once the video is indexed. Audio returns its transcript sentences. prompt, on one video or audio file, has a model watch or listen and answer the question (motion, sound, timing, when a thing happens), with the moments the answer rests on as seconds; cloud only, not for stills. Up to 12 refs or urls a call; one that cannot be read comes back as its own error. Cached. inspect_timeline reads the composite. Free to look; cuts on an unindexed video, shot_log and prompt use cloud processing.
Authorizations
An API key (es_live_…) from Settings, under API keys, or an OAuth access token.
Headers
Retry a paid dispatch safely: the same key returns the job already dispatched instead of charging again.
The workspace this one call acts in, when it is not the key's own.
Body
A library item ref. One of ref, refs or urls.
Up to 12 library item refs, read in one call. Every ref gets the same mode, window and at. One that cannot be read comes back as { ref, error } beside the rest rather than failing the call.
1 - 12 elementsUp to 12 web addresses of pictures or videos, read without keeping them: the media a pull returned, say. The same modes as refs; twelve in all with refs. One that cannot be read comes back as { url, error }.
1 - 12 elementsVideo only. overview = whole-video storyboard grid (default); window = zoom into a span; frame = one exact moment; strip = one span (up to 60 s) as frames, waveform and words in one picture.
overview, window, frame, strip, shot_log window and strip modes: {start, end} in seconds.
frame mode: timestamp in seconds.
x >= 0An image or a single video frame (frame mode) comes back ruled in tenths, 0 to 1 each way, so a point read off it is the x and y that edit_points on generate_image and generate_video take.
Video only: every shot change and shot length. Free once the video is indexed; otherwise the file is read through on cloud processing.
A question for a model to answer about one video or audio file: motion, sound, timing, how type enters, when a thing happens. The answer comes back with the moments it rests on as seconds. Not for stills; one ref or url with it.
1 - 4000
