How do I remove text from a photo for free?

Upload the image, brush over the words you want gone with a margin of a few pixels past the letterforms, and press Remove. The AI rebuilds the surface underneath in around 20 seconds and returns a watermark-free file. Guests get 2 removes a day and signed-in users 2 more.

How do I remove a date stamp from a photo?

Use a small brush, around 20 to 35 pixels, and paint over the digits including the colons or slashes between them. On scanned prints the orange bleeds into neighbouring pixels, so paint slightly wider than the numbers appear. Corner date stamps over sky, sand or a plain wall are one of the easiest cases there is.

Can I remove hardcoded subtitles from a video?

From a still frame, yes — screenshot the frame, upload it, and brush the subtitle line. For an actual moving video use the dedicated video subtitle remover, which processes the whole clip rather than one frame at a time.

Does the tool read the text before erasing it?

No. It is an eraser, not OCR. The model never decodes the words; it only rebuilds the surface they were sitting on. If you need the text extracted or translated, run an OCR tool first and clean the image afterwards.

Why is there a faint outline where the text was?

The mask was almost certainly too tight. Caption and meme text carries an outline or drop shadow that extends past the coloured fill, and any of it left unmasked traces the original lettering. Run a second pass over just the ghost.

How to Remove Text from a Photo: A Guide by Text Type

By the MagicRemover team · August 4, 2026 · 8 min read

In short

To remove text from a photo: upload it, paint over the writing with 3–5 pixels of margin past the letterforms, and press Remove. The detail that decides the result is the kind of text you are erasing — a flat meme caption comes off in one sweep with a 50–100 px brush, while a shop sign in the scene needs several passes and will come back invented rather than blank. Brush sizes and realistic expectations for six categories are below.

Open the text remover →

Most guides treat “remove text” as one job. It is not. Erasing the orange date a camera burned into a corner and erasing the lettering off a shop front are different problems with different techniques and very different success rates, and knowing which one you are looking at before you start saves a lot of wasted attempts.

The mechanical part is the same throughout: paint a mask, the model rebuilds the surface underneath. Everything that varies is in the brush and the expectations.

Before & after

Text erased from a photo, with the background behind it rebuilt — before

Text erased from a photo, with the background behind it rebuilt — after

Before / After

Six kinds of text, six techniques

Date and time stamps

Easy

Orange digits burned into a corner by a film camera

Brush: 20–35 px

Include the colons and slashes between the numbers. The colour bleeds into surrounding pixels on scanned prints, so paint a couple of pixels wider than the digits look.

Subtitles and captions

Easy

Hardcoded subtitles on a film still or a social video frame

Brush: 25–45 px, one stroke

Sweep the whole line in a single continuous stroke rather than word by word. Caption text usually has a dark outline for legibility — paint past the outline, not just the white fill.

Meme and overlay text

Easy

Bold Impact headlines, POV labels, sticker captions

Brush: 50–100 px

The easiest category, because meme text usually sits on a flat area of the image. Big brush, one sweep, done.

Watermark text

Medium

SAMPLE, an agency name, a diagonal © phrase

Brush: 40–90 px along the diagonal

Follow the angle of the text rather than making a rectangle around it — a tighter mask that follows the stroke gives the model more intact context to rebuild from. Only on images you own or have licensed.

Signs and labels in the scene

Hard

A shop sign, a name badge, a number plate

Brush: 30–70 px

This is real content in the photograph, not an overlay, so the model must invent a whole surface. Expect to work in passes and accept an invented sign face rather than a blank one.

Handwriting and document text

Medium

A note on a form, a signature, a price tag

Brush: 20–50 px

Clean on plain paper. A signature pressed into glossy stock physically deforms the surface, so the indentation can survive even after the ink is gone.

Four mistakes that account for most bad results

✗ Painting each word separately

Cover the whole line in one stroke. Word-by-word masking leaves the model guessing at the gaps between words, and those guesses meet as faint vertical seams.

✗ Hugging the letterforms exactly

Descenders on g, j, p, q and y drop below the baseline, and dots and accents sit above it. A mask that traces the visible body of the text leaves those behind as specks. Add 3–5 pixels all round.

✗ Ignoring the drop shadow

Most overlay text carries a shadow or stroke so it stays readable. Leave it and you get a legible ghost of the original wording. Paint past it.

✗ Zooming out to mask small text

A pale timestamp on a bright sky is easy to under-paint. Zoom the browser in with Cmd/Ctrl + and paint its known position confidently — the model cleans faint regions just as well once they are actually masked.

The background matters more than the text

Whatever category you are in, what actually decides the result is what the text was sitting on. Writing over sky, a painted wall, paper, skin or a smooth gradient reconstructs almost perfectly, because the model only has to continue something uniform. The same words over foliage, brickwork, a patterned rug or a second layer of text are much harder, and can leave a soft patch you will want to touch up.

If you have a choice of frames — several screenshots of the same scene, say — pick the one where the text falls on the plainest area. It makes more difference than any brush setting.

In short

Before & after

Six kinds of text, six techniques

Date and time stamps

Subtitles and captions

Meme and overlay text

Watermark text

Signs and labels in the scene

Handwriting and document text

Four mistakes that account for most bad results

The background matters more than the text

Questions about erasing text

Keep reading