ON YOUR DEVICE · FREE

Remove Text from Image

Words baked into the picture, painted over and rebuilt

How this compares with Photoroom

  1. 1

    Why the text cannot just be deleted

    Once words are flattened into a picture they are the picture. A screenshot, a meme, a photograph of a sign: in all three the letters are pixels like every other pixel, and whatever was behind them is gone from the file.

  2. 2

    Paint over the letters

    Trace them rather than boxing them in. Text is thin, so a tight stroke leaves most of the background untouched and gives the repair the most to work from. A rectangle over a line of text throws away everything between the letters, which is the part that was fine.

  3. 3

    What it is good at, and what it is not

    Text on a flat wall, a sky, a blurred background or a solid colour comes out clean. Text over a face, over other text, or over something with a straight edge running through it is where the rebuild has to guess, and it will.

FAQ

How do I remove text from an image?
Drop the picture in, paint over the words, and press the button. It runs in your browser, so there is no upload, no account and no watermark on what comes out.
Can I edit the text instead of removing it?
Not here, and not anywhere, from a flattened picture. The letters are pixels rather than characters, so there is nothing to select or retype. What this does is rebuild what was behind them, which leaves you a clean background to put your own text on.
Is this the same as the watermark tool?
The same engine, and a different problem. A stock watermark is translucent and spread thinly across the whole picture, so a lot of what was behind it still shows through. Baked in text is opaque and has hard edges, which is easier to paint accurately and harder to reconstruct underneath.
Why did it leave a smudge where the text was?
Because it had nothing to copy from. A line of text crossing a face or a patterned surface asks the repair to invent detail that was never in the file. It does best where the background is repetitive or plain, and worst where it is unique.
Is my picture uploaded?
No. The repair model runs on your machine, and no request carries the picture anywhere. The model itself is downloaded once and your browser keeps it.
Will it work on subtitles burned into a video?
One frame at a time only, since this takes a picture rather than a video. Burned in subtitles change every few seconds, so a video would need every frame painted, which is not what this is for.

The rest of this converter

More tools like this one