Why does my trimmed product-ad shot keep extra frames or start at a nonzero timestamp?

Separate frame selection from the playback clock

A time-based trim reads frame timestamps. The start_frame and end_frame options instead count frames passing through that filter. The end boundary identifies the first frame to discard. With variable-rate material, a frame number is not interchangeable with a time calculated from a nominal frame-rate label.

The retained frames keep their timestamps. This matters when preparing separate shots for a concat filter, whose documented requirement is that segments start at zero. Apply rebasing after selecting the intended content, and inspect the audio path separately rather than assuming a video filter retimed narration too.

Executed boundary comparison

We generated a two-second testsrc2 sequence at 10 frames per second, with 20 frames. FFmpeg 7.1 showinfo logged each output frame before muxing. A trim from 0.4 to 1.2 seconds retained input frames 4 through 11: eight frames, with timestamps from 0.4 through 1.1 seconds. Adding setpts=PTS-STARTPTS kept those same input-frame checksums while changing the first timestamp to zero.

A single trim with start=0.4, start_frame=8, end=1.2 and end_frame=16 retained frames 4 through 15, not the narrower overlap a reviewer might expect. By contrast, chaining a 0.4–1.2 time trim with a 0.8–1.6 time trim retained frames 8 through 11. These measurements apply to this synthetic constant-rate input, not every file or command placement.

Choose the condition you actually mean

Illustrative exercise: shot A needs only the interval in which a product label is fully readable. Shot B must satisfy both a first timing window and a stricter approved-action window. For A, record a single time range and inspect its first and last retained frames. For B, express the intersection deliberately, such as chained time trims, rather than adding unrelated time and frame-count limits to one filter.

If a chosen range includes an unapproved product state, changing its starting PTS will not remove that state. Return to content selection first. Conversely, if the correct states survived but a following concat operation expects zero-based segments, preserve the selection and adjust its timestamp origin.

Accept the assembled file separately

Record the source identity, filter order, first and last approved frame, and timing policy. Treat the final container’s start time, duration, video cadence and audio synchronization as additional checks. Our null-output measurements deliberately exclude muxer or encoder behavior and do not establish a delivered MP4 duration.

Keep the original clip and the approved review range. If the final file has extra content, identify whether it entered during selection, rebasing or later output processing before paying to regenerate a product shot.

Inspect the actual measurement data

These data describe the synthetic checks scoped in the article, not real-product, publishing-platform or subjective-quality acceptance.

• synthetic-video-filter-measurements.csv · SHA256 2d8619cda22dc3ba1219061ade31f1132d6f67a9b6cf81efbe7986396221220a

Illustrative exercise — approved interval and clock origin

These are illustrative exercise rows. Save a copy and record your own observations; the example is not an executed acceptance test.

Download review worksheet (CSV)