AIToday
Video GenerationAI Coding AssistantsZenn AI/MLPublished: Oct 9, 2026, 22:00 JST

One ffmpeg contact sheet cuts AI video QA to 1–2 minutes

One ffmpeg contact sheet cuts AI video QA to 1–2 minutes

An engineer ran a single ffmpeg command with fps=1/5 and tile=8x16 on an 8-minute 24-second 1080p clip; it finished in about 8 seconds and output one 2560×2880px PNG (3.6MB).

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Context & Analysis

The problem the engineer describes started with producing videos using AI: reviewing two 10-minute clips by watching them end to end took 20 minutes each time. Skipping that review risks shipping visible defects that automated checks do not catch, such as a background staying completely black for several minutes. The fix relies on ffmpeg's contact sheet, which arranges frames sampled at fixed intervals into a grid so a whole video becomes one image.

The engineer notes that burning timestamps into each frame via the drawtext filter is not always possible, because that filter is only available in ffmpeg builds that include libfreetype. The Homebrew build on hand (ffmpeg 8.1.1) did not list drawtext in the output of ffmpeg -filters, so the timestamp is instead back-calculated from the tile's position in the grid. One caveat is that if the grid has more cells than the video has sampled frames, the leftover cells are filled with black, which can look like the black-screen defect; the advice is to tell the AI that trailing black cells are padding.

The approach is deliberately split into two layers. Deterministic checks such as blackdetect, freezedetect, and matching actual duration against audio timing catch only abnormalities that can be defined in advance. The contact sheet plus AI acts as the visual layer, which can catch abnormalities that were never put into words but tolerates the coarseness of the sampling interval — a gap the engineer says must be closed by the deterministic layer.

FAQ
How long does it take to build the contact sheet?
On an 8-minute 24-second 1080p video, the ffmpeg command finished in about 8 seconds and produced one 2560×2880px PNG file of 3.6MB.
What kind of problems does the AI spot in the sheet?
It flags stretches where the background stays solid or black, overlapping or cut-off subtitles and captions, and abrupt changes between frames, with approximate timestamps.
How are timestamps calculated if they are not burned into the frames?
Since tiles fill left-to-right, top-to-bottom, the time is (row-1) × columns + (column-1), multiplied by the sampling interval — so with 8 columns and 5-second spacing, row 3 column 4 equals 95 seconds.

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleOpenAI halts Dark Clark, Iranian fake-byline influence ops