Frequently asked questions

Yes. The model generates replacement pixels for the masked area rather than applying a blur or scaling the frame. Results depend on the background: flat walls and simple textures reconstruct almost invisibly, while faces, foliage and fast motion may show artefacts you should check before publishing.

Manual selection defines a fixed region rather than a motion-tracked mask, so text that stays in one place works best. For captions that change every few seconds but keep the same position, select the caption band once and the whole clip is processed.

MP4, MOV, WebM and MKV up to 100 MB and 10 minutes per job. Input resolution up to 2K; output is capped at 1080p.

Roughly 5 to 20 times the clip length depending on the mode. A 30-second clip usually finishes in a few minutes. You can close the tab — the job continues and appears in your dashboard.

Turn subtitles off in your player. If the text stays on screen, it is burned into the picture. Another quick check: open the file in a subtitle editor — if no track is listed, there is nothing to disable.

Yes. The output keeps the original resolution, frame rate and duration, so any subtitle file timed to the original clip stays in sync.

The default model handles English and Chinese well. A newer version adds Spanish, Korean, Thai, Indonesian, Vietnamese, Japanese, Portuguese, German and French, at the cost of longer processing time.

Semi-transparent text is harder to mask cleanly. Select the caption band manually and check the preview; if the result is soft, try the higher-quality erase mode.

Softsubs are a separate text track the player renders on top of the video. Hardsubs are rendered into the frames at export time. Softsubs can be toggled or replaced instantly; hardsubs must be removed pixel by pixel.

Every new account gets one free run, enough for a 15-second clip. Longer videos use credits — one credit is one minute of standard-quality erase.

Yes. Only the picture is modified; the audio stream is copied through untouched.

The output is re-encoded, so there is a small generational loss. Resolution and frame rate are preserved, and the erased area is reconstructed rather than blurred.

Less than cropping and upscaling. The frame keeps its original resolution, and only the masked area is regenerated. There is still one re-encode, so keep the source file if you plan more edits.

Partially. Manual selection defines a fixed rectangle, so choose one that covers the full range of movement. Text that travels outside that box will be missed.

Include the box in your selection, with a small margin. The model removes the whole element rather than trying to separate text from its background.

Yes — those are ordinary MP4 files with burned-in text. Vertical video is supported; keep the selection inside the frame and the output stays vertical.

No. That is a copyright and licence issue regardless of the tool used, and it is explicitly prohibited in our acceptable use policy.

A moving watermark needs a fixed selection box that covers its full travel path, which means a large area gets reconstructed. Quality drops as the selected area grows.

On simple backgrounds, usually yes. On detailed or fast-moving backgrounds, expect some softness in the repaired area. Always review in motion before publishing.

Files are deleted seven days after upload. You can delete a job earlier from your dashboard, which removes both the source and the result immediately.

Yes, as long as the file is digital. Capture or digitise the tape first, then upload the resulting MP4 or MOV. Heavy analogue noise makes reconstruction harder, so check a short segment first.

No. Duration, frame rate and audio are preserved.

No — automatic detection looks for captions in the lower half of the frame. Use manual area selection for corner timestamps.

Pro and Business plans support batch submission of up to five clips. Each clip is charged by its own duration.

It can remove a text watermark you select. The animated TikTok logo is a moving graphic element — it is harder to reconstruct and results vary, so test a short segment first.

Yes. The output keeps the input resolution and aspect ratio, up to 1080p.

Yes — Reels exports are the same kind of file. Use the same workflow; only the caption position may differ.

Yes. You own the footage, so removing your own captions or handle is fine. Removing another creator’s watermark is not.

If the text is at the very edge and you can afford to lose 5% of the frame, crop. If you only need to hide information, blur. If you need the shot to look untouched, reconstruct.

Every new account gets one free run of up to 15 seconds. Beyond that, credits are sold in packs or included with a subscription.

No. The audio stream is copied through unchanged.

Ten minutes and 100 MB per job. For longer files, split the clip or contact us about the API.

You can remove a text or handle watermark that you select. Note that removing another creator’s watermark to repost their work is a copyright issue and against Instagram’s terms — this tool is for footage you own.

Yes. The output keeps the input resolution and aspect ratio, capped at 1080p. Nothing is cropped, letterboxed or rescaled.

Manual selection defines a fixed region, so captions that jump position need a box large enough to cover every position they occupy. If they drift across the whole frame, run the clip in segments or use automatic mode.

Use Your activity → Download your information, or export the original from your editor if you still have the project. Re-downloading a published Reel gives you a re-compressed copy, which is usable but softer.

No. Removing a watermark or handle from another creator’s video to repost it is copyright infringement and violates YouTube’s terms. Only process footage you own or have a licence to edit.

There is nothing to remove — the player UI is drawn by YouTube at playback time and is not part of the file.

Chapters live in the description, not the video. If a timestamp is burned into the picture itself, select it manually and it is treated like any other text.

Input up to 2K is accepted; output is capped at 1080p. For 4K sources, downscale first or contact us about the API for batch work.

Technically the tool will attempt it, but doing so is almost always copyright infringement. We do not support using the service to strip broadcaster idents from content you do not own.

Partly. A semi-transparent mark is blended with the background rather than replacing it, which makes the mask boundary harder to detect. Use manual selection with a slightly larger box and review the result in motion.

Up to five boxes per job. Keep the total covered area small — under about 4% of the frame — for the most convincing reconstruction.

No. The selection is a fixed region in frame coordinates, so it follows the frame, not the scene. If the mark stays put on screen, it stays covered.

No. For privacy redaction, blur is correct and reconstruction would be a mistake — you do not want to invent a plausible face. Reconstruction is for removing text that is merely in the way.

On flat backgrounds and moderate motion, usually not. On faces, foliage, water or heavy grain, some softness in the repaired area is normal. Always review in motion before publishing — a still frame hides flicker.

Yes. Process the clip here first, then apply your blur or mosaic in any editor. Doing it in that order avoids the blur being interpreted as part of the background.

It can, but with less information to work from. A tripod shot with a static background behind the text is the hardest case; a slow push-in or handheld drift gives the model far more to reconstruct from.

Yes. You get an MP4 back at the original resolution and frame rate, with the audio stream copied through.

Yes, MOV is accepted. iPhone recordings are often HEVC-encoded — those work, but converting to H.264 first can shorten processing time.

Usually slightly. The video is re-encoded around the repaired frames, so the bitrate may differ from the original even though resolution and duration are identical.

Not in the web app. Batch processing is available through the API on the Studio plan — contact us with your expected volume.

No. It runs in Safari. Nothing is installed, and no access to your photo library beyond the single clip you pick.

Tap Download, then choose Save Video — the clip lands in Photos next to the original. You can also re-download it later from your dashboard.

Yes, and the larger screen makes precise selection easier. The workflow is identical.

Yes — Chrome on Android behaves the same way. The upload picker opens your gallery instead of Photos.