Check a 30-minute render before upload: 24 stills and a transcript

Video Inspect reads up to 1,800 seconds, returns up to 24 stills per call and a transcript at $0.01 per audio minute. A QA pass before a long upload.

5 min readSume
All posts

Short answer

Sume Video Inspect accepts a source up to 1,800 seconds, returns up to 24 stills per call (default 8 at max_edge 768) and can transcribe speech at $0.01 per audio minute. That matches the 30-minute Timeline render cap, so you can QA a full-length render before uploading it to a platform whose upload takes much longer than the check.

The transcript rate alone is $0.30 for 30 minutes; every inspect also bills its compute, so confirm the total in GET /v1/catalog. Spot-check stills at a few points and you will usually catch a missing segment or a bad cut before it ships.

What the check covers

Which problem each output catches, using the limits from the Video Inspect doc.

QA with Video Inspect (read 2026-10-03)
OutputLimitCatches
Stills24 per call; default 8 at max_edge 768Wrong clip order, black frames, clipped text
Seek modeprecise or fastFast for a quick look, precise for an exact frame
Transcript$0.01 per audio minute; duration_seconds hint max 600Missing or garbled narration
Source lengthUp to 1800 secondsFull 30-minute render in one call

A practical pass

Take eight stills evenly spaced across the render for a first look, then more near any transition you changed. Read them at phone size. For a 30-minute video, 24 stills is one still about every 75 seconds, which is enough to see that sections appear in the right order.

For the audio, note that the duration_seconds hint for speech-to-text has a maximum of 600. The hint is only what Sume reserves for the transcript; it is not stated as a limit on transcript length, so check the returned text covers the whole render, and compare it with the script. A skipped paragraph shows up as a gap you can see.

  • Check the first and last second; most mistakes are at the edges.
  • Scan for black frames near transitions.
  • Compare the transcript with the script for cuts and repeats.
  • Only then start the platform upload, which can take much longer than the check.

Why before the upload

Platform uploads are slow and sometimes irreversible. TikTok's guide lets API uploads run up to 10 minutes and 4 GB with sequential chunks, so a bad file wastes the transfer. A check that costs cents is cheaper than re-uploading a gigabyte.

The check is not a guarantee of platform acceptance; it tells you what the video contains, not whether the platform will approve it. For length questions on TikTok, see the 60-minute post.

What it will not tell you

Inspect shows what is in the file, not whether a platform will accept it. It does not test codec profile, bitrate or container structure; use a probe tool for that. It also does not judge quality or taste. The goal is narrow: confirm the right content is there, in the right order, with the right words.

If a stage fails, fix it at the source. Re-render only the segment that changed; with Timeline, a change to one slot means a new render of the whole timeline, so for long programs split into chapters and re-render the chapter that changed.

Sources

Related posts

More in Media tools

All Media tools posts

Written by Sume