How to remove hardcoded subtitles from a video

A soft subtitle is a separate stream inside the video container. It carries text and timing, not pixels. Players can turn it on or off, and you can remove it without touching the picture.

A hardcoded subtitle — also called burned-in or open caption — was rendered into the frames before export. Every letter is part of the image. No player can hide it, and removing it means changing pixels.

A remux copies the video and audio streams into a new container and leaves the subtitle stream out. It takes seconds, does not re-encode, and does not touch a single pixel.

In FFmpeg, a stream copy looks like this: read the file, map only the video and audio streams, and write a new MP4. Desktop editors do the same thing when you delete a subtitle track and export without re-encoding. The FFmpeg documentation covers stream mapping, and HandBrake’s subtitle documentation shows the same route in a GUI.

• Check whether the file has a subtitle stream before you start
• Use stream copy, not a re-encode, to keep quality identical
• Keep the original file until you have checked the output
• If the words still show, the subtitle was hardcoded after all

There is no free, lossless way to remove text that lives in the pixels. Each route trades something: framing, background detail, time, or money. The four routes are compared in detail in remove text from video without blur.

Re-editing from source is always the cleanest result. It is also the route people skip, because the project file is often gone. Check your editor’s autosave, cloud sync, or the client who sent the clip before assuming it is unavailable.

Subtitles mode follows the caption band and removes the lines it tracks. It expects the words to stay in roughly the same lower region. If captions jump between the top, middle, and bottom of the frame, Select area with a set of boxes is more predictable.

Ready to try it? The remove subtitles from video page covers the modes, the 200 MB and 10-minute limits, and the free monthly seconds.

• Trim the clip to the part you need, then upload an MP4 or MOV up to 200 MB and 10 minutes
• Choose Subtitles if the caption stays in the lower band, or Select area if it moves
• Render the uploaded clip — the live engine processes it as a whole, so trim first
• Run the job, then check the hardest five seconds before downloading
• Compare the audio and frame size against the source

Background rebuilding is an estimate. The model has to invent plausible detail where the text used to be. That works best on simple or slowly moving backgrounds and worst when detail must be preserved exactly.

• Captions crossing a face, hands, or a product being handled
• Fine patterns — hair, text on clothing, railings, foliage
• Low-contrast letters on a busy background
• Heavy compression that smears the edges of the letters
• Captions that change position every few seconds

Do not start with the full clip. Trim the five seconds where the caption sits over the most detail into a separate test clip, upload only that, and review it at 100%. If those frames hold up, the rest of the clip will usually be easier. If they do not, you have saved the cost of a full render and learned which frames need manual work.

A short test clip also tells you whether Subtitles or Select area suits this clip, before you commit to a mode.

• Confirm the subtitles are hardcoded, not a soft track
• Check whether a clean source or project file still exists
• Trim to the part you actually need before uploading
• Test the hardest five seconds
• Review faces, hands, and fine patterns at 100%
• Confirm audio, duration, and frame size in the export
• Keep a local copy of the original