To remove subtitles with automatic text detection, you upload the video and let the tool find the subtitle lines for you. There is no need to draw boxes or mark the subtitles frame by frame. A detection pass scans each frame for text, marks every subtitle line it finds, and an inpainting model then rebuilds the pixels behind those lines from neighboring frames. In Subtitle Remover this is Auto mode. It covers the lower half of the frame, where nearly all dialogue subtitles sit, and costs 1 credit per second of video.
This guide explains how automatic text detection works, what it can and can't find, and how to clean a clip with it. It is written for creators, editors and localization teams with burned-in subtitles they want gone.
TL;DR
- Automatic text detection finds subtitle text on its own. You upload, pick Auto, and start.
- It scans the lower 50% of the frame. Bottom subtitles, bilingual lines and burned-in captions are covered.
- Text higher up, logos and watermarks need Box select, where you draw the region yourself.
- Auto costs 1 credit per second, Box select 3. Files up to 1 GB and 2K, clean MP4 output up to 1080p.
What is automatic text detection?
Automatic text detection is the step that locates text in an image before anything is done to it. It is the first half of optical character recognition (OCR): before a system can read a word, it has to find where the word is. For subtitle removal, finding is all that matters. The tool never needs to know what the subtitle says, only which pixels belong to it.
In a video this happens on every frame. Subtitle lines appear, change and disappear several times a minute, so the detector follows them through time and builds a mask that matches the text exactly while it is on screen. Frames with no subtitle are left untouched.
This only matters for burned-in (hard-coded) subtitles, which are part of the picture. If your player can switch the subtitles off, they are a soft track, and you can delete that track without any AI at all. Our guide on how to remove text from video explains the 10-second check.
How Auto mode removes subtitles
Subtitle Remover's Auto mode runs in four stages:
- Scan. Each frame's lower half is searched for text. The upper half is left as it is, so signs, labels or titles up there are not touched.
- Detect. A region is drawn around every subtitle line, including two-line and bilingual subtitles, outlined lettering and subtitles on a colored background box.
- Look around. The detected regions become a mask. In the frames just before and after, the camera or the subject has usually moved, so part of the background the text was hiding is visible.
- Rebuild. Video inpainting fills in the masked pixels so they match the scene's texture, motion and lighting. You get a clean MP4, not a blurred strip or a black bar.
Because the mask follows the detected text, there is nothing to set up. That makes Auto the fastest way to clean an ordinary subtitled clip, and the cheapest.
Auto or Box select: where the text sits decides
Automatic detection is built for subtitles, so it looks where subtitles are. Use this table to choose:
| Text in your video | Mode | Why |
|---|---|---|
| Dialogue subtitles at the bottom | Auto | Inside the scan zone; detection finds every line |
| Bilingual or two-line subtitles | Auto | Both lines are found as long as they are in the lower half |
| Burned-in captions on a Reel, Short or TikTok, below the middle | Auto | Same as subtitles; see the caption remover for more |
| Titles, labels or captions in the top half | Box select | Outside the scan zone; draw a box over them |
| Logos, watermarks, timestamps | Box select | Not subtitle text; box them wherever they are. See remove watermark from video |
| Subtitles over a very busy background, where detail matters | Box select | Uses the finer reconstruction model described below |
There is one quality trade-off to know about. Auto is the cheapest and fastest mode, but on a very busy background (dense foliage, crowds, fast-moving water) the cleaned area can look slightly softer than the rest of the frame. Box select uses a finer model that rebuilds texture in more detail. If an Auto result looks soft in the moments that matter, run the clip again with a box drawn over the subtitle area.
Step by step: remove subtitles with automatic text detection
Setting up takes about a minute. Processing then runs in the background.
Step 1: Upload the video
Open the Studio and add an MP4, MOV, MKV or WebM file up to 1 GB. The source can be up to 2K (2560×1440). The duration and resolution are read in your browser before anything is uploaded, so you see the credit estimate straight away.
Step 2: Keep Auto selected
Auto is the mode for ordinary subtitles. There is nothing to draw. If you skimmed the clip and the subtitles move into the top half at some point, or there is also a logo you want gone, switch to Box select instead.
Step 3: Check the estimate and start
Auto is billed at 1 credit per second of the video's actual length, never rounded up to a full minute. A 45-second clip costs 45 credits. The credits are reserved when you start, and if the task fails they are refunded.
Step 4: Wait for processing
You don't need to keep the page open. A 30-second clip takes about 12 minutes, and a 2-minute video roughly 12–20 minutes. The result appears in your task list when it's ready.
Step 5: Review and download
Compare before and after. Pay attention to moments with fast motion or a face passing behind the subtitle area, then download the clean MP4 (up to 1080p). Source and output files are deleted automatically after 30 days, and your videos are never used for training.
Tips for a cleaner result
- Start from the best file you have. Detection and inpainting both work from the pixels. Heavy compression blurs the edges of the letters and leaves less background detail to rebuild from.
- Trim before you upload. If only one part of the video has subtitles, cut that part out. You pay per second, so a shorter file costs less.
- Check the halfway line. Subtitles that sit unusually high, such as on a vertical video with captions near the middle, may fall outside the scan zone. Use Box select for those.
- Treat Auto as the first pass. It handles most subtitled clips on its own. Keep Box select for the clips where the background is too busy or the text is somewhere else.
- Fix low-quality sources afterwards. If the original was already blurry, the AI Video Enhancer can sharpen the cleaned result.
Common mistakes
- Running AI on soft subtitles. If the subtitles can be switched off, delete the track. That is free and lossless.
- Expecting Auto to remove logos. Automatic detection looks for subtitle text in the lower half. A corner logo or watermark needs a box.
- Blurring subtitles before translation. A blurred strip under new subtitles looks like a mistake. Translators and dubbers need a clean frame, which is what inpainting gives you.
- Removing subtitles from footage you don't own. Removing subtitles does not give you the right to reuse someone else's video. Only process videos you own or are authorized to edit.
FAQ
Can AI remove subtitles automatically?
Yes. Automatic text detection finds the subtitle lines on each frame, and video inpainting rebuilds the background behind them. In Subtitle Remover you upload the video, keep Auto selected and start. You don't draw anything.
What does automatic text detection miss?
It only scans the lower half of the frame, so titles, labels and captions in the top half are left alone, as are logos and watermarks anywhere in the frame. For those, use Box select and draw a region over them.
Does automatic detection work with Chinese, Japanese or bilingual subtitles?
Yes. Detection finds where text is rather than reading it, so it handles any script: Chinese, Japanese, Korean, Russian, English and bilingual two-line subtitles, including outlined text and subtitles on a colored background box.
How much does it cost to remove subtitles with Auto?
1 credit per second of the video's actual length. A 60-second clip costs 60 credits in Auto and 180 in Box select. New accounts get 15 free credits, enough to test a 15-second clip. Plans and credit packs are on the pricing page.
Will the cleaned area be blurry?
On most clips it isn't: the area is rebuilt from neighboring frames, not blurred. On a very busy background, Auto's result can look slightly softer than the surrounding image. If that happens in an important shot, rerun it with Box select, which rebuilds texture in more detail.
Can I remove captions from TikTok, Reels or Shorts with Auto?
Yes, if the video is yours and the captions sit in the lower half of the frame. Captions placed near the middle of a vertical video may be above the scan zone; use Box select for those. The caption remover page covers social-video captions in more detail.
Conclusion
Automatic text detection takes the manual work out of subtitle removal. Upload the video, keep Auto selected, and the subtitle lines in the lower half of the frame are found and rebuilt for 1 credit per second. Switch to Box select for anything outside that zone, such as top titles, logos or watermarks, or for busy shots where you need the finest detail.
Ready to try it? Open the Subtitle Remover Studio, upload a clip, keep Auto, and download the clean MP4 once processing finishes.
