Check a 30-second ad in 15 stills: video inspect fps 0.5 costs nothing

To review a 30-second ad as a contact sheet, call Sume's video inspect with frames set to fps 0.5. That samples a still every two seconds, mid-bin, so a 30-second clip returns 15 stills as durable media.sume.com images. Probe and stills are free; only the optional transcript is billed, at $0.01 per audio minute.

What the call returns

Video inspect takes one clip in your workspace, up to 1,800 seconds, and returns a probe plus frames. The frames program accepts either at[] with 1 to 24 explicit timestamps or fps greater than 0 and at most 2, with a limit of 24 stills. Sending both is refused with video_inspect_frames_program_conflict.

The default frame edge is 768 pixels, and max_edge accepts 64 to 2,160. For legibility of on-screen price text, ask for a larger edge and png rather than jpeg. Default mode is sync and waits up to 30 seconds, then returns 202 if the work is not done.

What to look for

Check the first still for the hook, the middle for the product in frame, and the last for the end card and legal line. Scan every still for artifacts such as extra fingers, garbled text and logo drift. The 8-still default is fine for a quick look, but 15 catches a bad 2-second segment that 8 would skip.

Use seek fast for a quick look: it moves each still to the keyframe at or before its instant, which can be earlier by up to one group of pictures, while precise decodes the exact instant. Use precise when you will cite the timestamp.

Adding the words

Set transcribe true to run Sume STT on the clip's audio at $0.01 per audio minute. If you do not send duration_seconds Sume reserves 1 minute, and the maximum hint is 600 seconds. A 30-second ad bills a fraction of a minute at most one cent, so the full review of 15 stills plus transcript costs at most about one cent. Frames and transcript together let a reviewer match every spoken claim to what is on screen.

A practical note

For a team review, put the stills in a shared doc with the timestamp under each one. The response gives each frame as t, url, width and height, so the timestamp is already in the data. Reviewers can then reply with a time rather than a vague point in the video, and an editor can trim or re-generate exactly that two-second span. If a defect sits between two sampled stills, rerun with at[] and the exact seconds, up to 24 values, instead of raising fps. This keeps the check free while still finding the problem.