Remove TikTok Watermark

Same-file test: before and after

We ran the production tool again on 24 October 2026, selected the same owned six-second clip, set both repair zones, and downloaded a new result. This is a synthetic TEST overlay on a plain background, not a downloaded TikTok export. The comparison below uses frames extracted from that exact input and the newly saved output at one and four seconds.

The new WebM was 114,406 bytes and decoded successfully. Both extracted output frames measured 320 by 480 pixels, and decoded audio had a nonzero signal. The source and exported audio decoded to different durations, so keeping an audible track does not mean preserving exact timing or the original audio encoding. Play the saved file before using it.

Repeat this same-file comparison using the owned input clip and this run’s saved output. The coordinates and the earlier failed short-box test are recorded below. They explain why covering all the lettering and leaving a clean sampling border matters more than a tidy thumbnail.

What our saved-file checks found

We made two synthetic clips ourselves, then used the same file chooser, zone controls, export button and download link shown above. They were prepared on 23 October and tested on 24 October 2026. They are controlled fixtures with TEST lettering, not downloaded TikTok exports or proof that every TikTok layout works. The clean-background clip contains two fixed positions: one in the upper-left during the first half, and one in the lower-right during the second. It also carries an audio tone.

The complete source video has 150 frames at 25 frames per second, a 320 by 480 pixel frame, six seconds of video and an AAC audio stream. We set the first box at X 14, Y 24, width 80, with a time range from zero to three seconds. The second starts at X 224, Y 410, with the same width, and runs from three to six seconds. Those coordinates belong to this fixture. They are not default TikTok watermark locations and should not be copied onto an unrelated video.

The first saved result taught us something that a clean thumbnail hides. A box height of 36 pixels removed most of the white lettering, but its bottom sampling edge still touched the mark. The patch carried a bright vertical streak into the repaired area. Enlarging both boxes to 44 pixels included a clean border. We exported and downloaded again; frames extracted at one and four seconds showed the selected lettering gone on the flat background, while the central object and full frame remained. The first, imperfect output was retained rather than reported as a pass.

The successful downloaded WebM measured 116,809 bytes. Independent decoding found VP9 video at the same 320 by 480 dimensions and an Opus audio stream at 48,000 Hz. The audio decoded to six seconds and had nonzero signal energy: it was not merely a silent audio track attached to a video. The source had mono AAC, while the recording produced stereo Opus. That difference is why this page says the audio is routed and re-encoded; it does not promise a byte-identical soundtrack or the original container.

We also checked a harder background in the actual browser preview. A thin diagonal line ran through the selected mark. The patch removed the lettering but broke and smeared the line. A clean border is necessary, yet it does not make this method reconstruct a texture or a hidden edge. That case is a stop signal. Do not accept the output just because the logo text has become less readable.

The controls were checked for an invalid selection too. Setting the first box's X coordinate to zero produced an explicit border error and disabled export. The box must leave room for its samples inside the frame. Separately, the tool rejects overlapping repair zones and time ranges outside the clip. These checks stop invalid inputs; they do not judge whether a valid patch looks convincing. That decision still needs the original and the saved result side by side.

You can repeat the small-background check using our owned synthetic test clip. The saved test result is available for comparison. Start with the coordinates above and inspect both positions. Keep the original file, listen to the saved audio, and check the frame near the changeover as well as near the end. If a different mark moves through the scene, crosses a face or touches useful text, these fixed boxes and border estimates may be the wrong method.