Cleanmark

Does video watermark repair keep chapter markers?

No: our downloaded repair did not retain the original chapter markers. The MP4 contained three verified chapter names and time ranges; its saved WebM contained none. Keep the original if your delivery needs named navigation points, even when the repaired picture looks usable.

Open the local video repair tool

Tested 12 November 2026 with new footage and metadata we own. This is an actual input-and-download check in this browser workflow, not a promise about every player, chapter format or video editor.

We checked real chapter data before repairing

We made a four-second, 480 by 270 video with a moving gold circle, a frame counter, a generated 550 Hz tone and DEMO on a plain patch. The picture says “owned navigation metadata test”; those painted words are not chapters. The chapter evidence comes from extracting the encoded container metadata.

The chaptered input measured 80,927 bytes. Its extracted timebase was 1/1000, with Opening reference starting at 0 and ending at 1250, Middle reference at 1250–2750, and Closing reference at 2750–4000. Converting those ticks within that declared timebase gives the boundaries in the table. We checked the actual encoded MP4 rather than accepting the preparation command as proof.

Actual extracted input chapters — 12 November 2026
NameStartEndSaved result
Opening reference0.000 s1.250 sAbsent
Middle reference1.250 s2.750 sAbsent
Closing reference2.750 s4.000 sAbsent

Our no-chapter control measured 80,134 bytes. It contained no chapter entries, but its decoded video and audio hashes matched the chaptered input. Both inputs decoded to 96 video frames, and we verified every binary frame counter from 0 through 95. Each original audio stream decoded to 192,512 mono samples at 48,000 Hz. The control therefore changes chapter metadata without substituting different picture or sound content.

Same repair box, two real saved files

We selected each original file in the unchanged production tool. Each loaded as 480 by 270 pixels and 4.000 video seconds. Zone 1 was X 294, Y 95, width 118 and height 42 pixels, active from zero to four seconds. Zone 2 stayed off. The rectangle enclosed DEMO without covering the counter or moving reference.

Both runs reached the ready message. We downloaded each actual WebM and independently extracted its container metadata. Neither extraction contained a chapter entry. This is evidence about the saved files, not an inference from whether the input preview displayed a chapters menu. We did not inject a navigation control, modify the repair tool or replace the missing metadata with letters drawn on the output picture.

Actual downloads — 12 November 2026
Input caseInput / saved chaptersSaved bytesDecoded video framesDecoded audio
Named chapters3 / 0124,793723.96 s, stereo
No-chapter control0 / 0128,382773.96 s, stereo
Actual source and downloaded frames: DEMO cleared in the inspected pictures, while extraction found three original chapters and none in the saved chaptered result
Each row matches its own source counter. The upper row is counter 24; the lower row is counter 23. The diagram reports extracted metadata separately because a still picture cannot demonstrate saved navigation.

The picture result does not rescue the missing chapters

For the chaptered run, source counter 24 appears at saved frame index 18 and presentation time 1.390 seconds. In the control, counter 24 was not present, so we retained the nearest available counter, 23, at saved frame index 18 and 0.971 seconds. We matched each picture to that encoded source counter instead of pretending equal playback time or equal output-frame index identifies the same source moment.

DEMO is absent in these inspected output frames. That observation is limited to the shown moments; it is not a claim that hidden original detail was restored or that every frame is flawless. The input had 96 frames, while the saved files had 72 and 77, so neither output is described as an unchanged remux or an exact timeline copy.

Both complete downloaded files decoded successfully. Each contained a stereo Opus stream with nonzero signal and 190,080 decoded samples per channel at 48,000 Hz: 190080/48000 = 3.96 seconds. That measures decoded sound length, not the whole container timeline or perceptual synchronization. Hearing a tone is separate from retaining chapter names, and a valid downloadable video can still fail a navigation requirement.

The chaptered input had three saved metadata entries; its output had zero. The no-chapter control also had zero. This pair establishes the loss for our chaptered recording. It does not identify the internal component responsible or predict every browser and container combination.

Inspect the same files

What should I keep when chapters matter?

Keep the authorized original and its chapter names and boundaries before trying a repair. Inspect the downloaded file in your intended delivery workflow, including the navigation you need. If named chapters are required, this tested export does not qualify; use a workflow whose actual saved chapters you have verified. Do not accept a cleaner thumbnail as a substitute for that check.

Embedded subtitle cues and separate selectable soundtracks are different container features. The late-audio-start test checks sound and picture timing. None of those checks proves chapter retention, and this experiment does not establish the preservation or removal of invisible provenance.