Subtitles split into two completely different problems, and the first thing to do is work out which one you have. One takes seconds with zero quality loss. The other needs real reconstruction.
Check this first
Open the video in a player that exposes subtitle tracks — VLC is the usual choice — and look for a subtitle menu.
If the subtitles disappear when you switch them off, they’re soft-coded. They’re stored as a separate track and rendered by the player at playback time. The pixels of your video were never altered.
If they stay on screen regardless, they’re hardcoded — burned into the frames when the video was exported. They’re part of the picture now.
This single check determines everything that follows, and it takes about ten seconds.
Soft-coded: trivial
Strip the subtitle track from the container. Tools that remux a video file can do this without touching the video stream at all — no re-encoding, no quality loss, near-instant even on a long film.
There’s genuinely nothing else to do. If a tool wants to run AI processing on soft-coded subtitles, it hasn’t checked what it’s dealing with.
Hardcoded: reconstruction
Now the subtitles are pixels, and the area has to be regenerated from its surroundings.
Good news: subtitles are among the easiest text to remove. Unlike creator captions, they’re well-behaved:
- Consistent position — a fixed band, nearly always near the bottom
- Consistent size — same font, same scale throughout
- No animation — they appear and disappear rather than moving
- High contrast with an outline, which makes them easy to isolate
So instead of tracking anything, you define one region covering the subtitle band and apply it across the whole clip. Every subtitle in the video is handled at once.
Sizing the band
Two things to get right:
Height. Subtitles are sometimes one line and sometimes two. Size the region for the tallest subtitle in the clip, or the second line survives on those where it appears.
Outline and shadow. Subtitles carry a stroke or drop shadow so they stay readable over any footage, and it extends past the letterforms. Include it, plus margin.
The case that never works
Subtitles over other text.
Most often this is a second subtitle track — a video already subtitled in one language, then subtitled again in another, with the two overlapping. It also happens over signage, screens or documents in the shot.
Text cannot be reconstructed. Inpainting predicts how texture continues, and letterforms aren’t texture — there’s exactly one correct arrangement of pixels for a particular word, and no way to derive it from context. You’ll get something that looks like writing and reads as nonsense.
No tool solves this. If the underlying subtitles matter, that information is gone. See AI text remover.
Cropping is often the right answer
Genuinely underrated here, because subtitles sit near the bottom edge.
Cropping the subtitle band out is lossless for everything that remains — no reconstruction, no artefacts, no shimmer, no failure modes. You lose some height and the framing shifts.
For a 16:9 video destined for a vertical feed that’ll be cropped anyway, this is frequently the best option. Worth trying before reaching for AI.
Comparing the options
| Situation | Method | Quality |
|---|---|---|
| Soft-coded subtitles | Strip the track | Perfect, no re-encode |
| Hardcoded, bottom band, croppable | Crop | Perfect for what remains |
| Hardcoded, need full frame | AI reconstruction | Very good |
| Hardcoded over a face | AI reconstruction | Risky |
| Hardcoded over another subtitle | Nothing works | — |
| You have the project file | Re-export without the layer | Perfect |
Steps for hardcoded subtitles
- Confirm they’re hardcoded with the player check.
- Consider cropping if the framing allows.
- Define a region covering the full subtitle band, sized for two-line subtitles, including strokes.
- Apply across the whole clip rather than per subtitle.
- Review full screen in motion — check for shimmer, and for subtitle edges left at the top or bottom of the region.
- Export at maximum bitrate.
On iPhone
MarkOff handles a subtitle band as a single region applied across the clip, which suits this case well — one selection covers every subtitle in the video.
Video is processed on-device, with no blur step. A blurred band where subtitles were is arguably worse than the subtitles themselves; see remove watermark without blur.
Worth considering
Subtitles are frequently someone’s work. Translation and accessibility captioning take real effort, and fan-subtitled content in particular often represents unpaid hours.
Removing subtitles from your own video, or stripping a language you don’t need from something you own, is uncomplicated. Removing someone’s translation to repost the clip removes their contribution.
There’s also an accessibility dimension worth a thought: subtitles are how a lot of people watch video at all.