Can Moving Captions Hide a Frozen Video Background? A 7-Second Test

A moving caption can make a frozen background look active to a full-frame freeze detector. In a local seven-second experiment, FFmpeg reported a freeze in the still clip, then reported no freeze events when only the caption moved. Checking the background separately recovered the freeze.

If you use an automatic check before exporting an AI video or a narrated slideshow, this matters: a clean detector log can describe the finished frame without telling you whether the main scene moves.

The same scene, three conditions

The inputs were synthetic clips made for this experiment on October 7, 2026. Each contained the same dark scene and a green square. There was no borrowed footage, generated customer experience, voice track or music. Each encoded clip lasted exactly seven seconds, at 640 × 360 pixels and 30 frames per second.

Clip A kept the scene and caption still. Clip B kept the same scene still but moved the caption horizontally. A and B used the same caption text, font, size and starting position. Clip C kept the caption still while the green square moved inside the scene.

All three used the same detector settings: freezedetect=n=0.001:d=2. The noise threshold was 0.001 and the minimum freeze duration was two seconds. Each clip was checked once as a complete frame and once with the top 80 pixels containing the caption excluded.

Seven-second inputFull-frame freeze startsBackground-only freeze starts
A: still scene, still caption11
B: still scene, moving caption01
C: moving square, still caption00

What the zero actually means

In B, the caption changed enough pixels for the complete frame to avoid a freeze report at these settings. The underlying square and scene stayed still. When the detector saw only the lower background area, it reported a freeze starting at zero seconds.

The still clips remained frozen until the file ended. Their logs recorded a freeze start, without a later freeze-end or freeze-duration event. A missing end event does not turn an ongoing freeze into zero seconds. The table therefore counts reported freeze starts; it does not pretend every event has a completed duration.

This is one controlled example, not a promise that any moving caption will hide any freeze. Caption size, speed, placement, encoding and the noise threshold can change the answer. The moving square is a control for this scene, not a measure of realistic character animation.

A useful check before adding captions

Keep the clean scene export as well as the captioned export. First inspect whether the person, product or action actually moves in the clean version. Then run the detector on that version and on the finished edit. If a caption, logo or animated border is the only moving part, a whole-frame result can be misleading.

For a file with this exact 640 × 360 layout, the two checks are:

ffmpeg -hide_banner -i B.mp4 -vf "freezedetect=n=0.001:d=2" -an -f null -
ffmpeg -hide_banner -i B.mp4 -vf "crop=640:280:0:80,freezedetect=n=0.001:d=2" -an -f null -

The crop dimensions above fit this experiment only. On a different video, choose an area that includes the relevant action and excludes the moving overlay. Cropping away the action would produce a different question.

A separate edit test reframed B into a 1080 × 1920 vertical clip using a blurred background. It preserved the seven-second duration and contained no audio. Reframing made a vertical export; it did not make the green square move.

What this test did not measure

The experiment checked pixel changes and selected frames. It did not evaluate a full spoken video, viewer enjoyment, retention, sales or the quality of an AI model. A freeze detector is one diagnostic alongside watching the actual scene. Before relying on a clean log, check the source before the overlays.

Sources checked

FFmpeg's official freezedetect documentation, checked October 7, 2026, explains the frame-change threshold and freeze event fields. All counts above come from the locally created clips, not from a claim that FFmpeg certifies video quality.

Disclosure: this article was prepared with AI assistance. The synthetic clips and detector commands were executed locally. This post contains no affiliate links.

Comments

Post a Comment

Popular posts from this blog

FFmpeg Shows −1.5 dBTP. Did You Actually Fix the Audio?

An output_tp of −1.50 in an FFmpeg log does not mean your original audio has been fixed. A command can show that number without saving any corrected audio. Before uploading, measure the file you actually exported. This follow-up to our AI song quality checklist tackles one practical mistake: confusing filter output statistics with a saved deliverable. You need FFmpeg installed and a local audio file. These commands process files on your computer; they do not upload your recording. One recording, two different stages On September 18, 2026, we measured one newly generated, roughly 2-minute-20-second MP3. Its integrated loudness was −11.41 LUFS and its true peak was +0.84 dBTP . The same analysis log showed output_tp: -1.50 . input_i and input_tp describe the file supplied to that command. output_i and output_tp describe the signal after the filter in that execution. They do not establish the measured properties of a separately encoded final file. We then reduced gain by...
Read more

VoxCPM2 vs ElevenLabs for a 30-Second Science Short: We Timed Both on the Same Script

Image
Disclosure: DigitalHarvest (NSST) is an independent affiliate of ElevenLabs. We may earn a commission if you subscribe through the referral link in this article. ElevenLabs has not sponsored, reviewed, or approved this article. This relationship does not change the numbers below, and the free local option wins on cost in our own test. We make 30-second science Shorts from public-domain NOAA ocean footage, and every one needs about six lines of narration. On October 4, 2026 we took one finished script and voiced the same six lines twice: once with VoxCPM2, an open-source model running on our own Apple Silicon Mac, and once with ElevenLabs on the paid Creator plan we already use. Then we ran both outputs through the same Whisper check our pipeline uses before anything is published. Here is what came out. The script we used The episode is "Why doesn't hot water on the seafloor boil?" (a small black smoker vent filmed at 2,947 m by NOAA Ocean Exploration). Six lines, 428...
Read more

Before You Upload an AI Song or Video: The Checks We Run (Free Commands)

The FFmpeg and file-hash commands below run locally. No purchase is needed to follow them. In September 2026 we published a 140-second "music video" for one of our AI songs. It was one still picture with a small moving waveform along the bottom. Our pre-upload checks had measured the length, the resolution and the loudness. Nobody had checked whether the picture actually moved. We deleted it from our channel the next morning. This post shows the checks we run now. Every command below runs on your own computer with ffmpeg , and every number comes from files we measured ourselves. 1. Loudness and true peak (free, one command) ffmpeg -hide_banner -nostats -i song.mp4 -af ebur128=peak=true -f null - Read the Summary block at the end: I: is the integrated loudness in LUFS. Peak: under "True peak" is the true peak. ffmpeg prints the unit as dBFS. A test file we deliberately pushed too loud read I: -8.4 LUFS and Peak: 3.5 . Our own working baseline is ...
Read more