How to Remove Text from Image

A practical guide to taking text off a photo — the one-click browser route, the manual Photoshop route, and the phone route — with the caveats nobody lists.

Guide · updated 3 November 2026 · 6 min read

There are three honest ways to take words off a photo: let a model find them, paint over them by hand, or crop them out. Each one has a place. This guide walks through all three, says where each falls over, and shows the route most people end up taking when all they want is the text gone.

It is a guide, not a tool page. The tool itself lives on the PixPurge home page; everything below is about how to use it well, and when not to.

The fastest way: let the model find the text

Auto-detection is not magic and it is not always right, but for a photo where the text is simply in the way, nothing is quicker. The sequence is three steps.

  1. 1

    Upload the largest version you have

    JPG, PNG, WebP or HEIC, up to 10MB. A 1200px WhatsApp forward gives the model 1200px to work with; the original gives it far more room to rebuild what was behind the letters.

  2. 2

    Press Remove Text and let it read the frame

    It picks out text blocks — captions, timestamps, price tags, UI labels — and clears them in one pass. If you want to keep some of that text, skip the one-click button and mark only what should go with the brush.

  3. 3

    Check the spot at 100% before you download

    Most results need no second look. Where the text sat over busy detail, zoom in: a faint ghost of the letters means one more pass on that patch.

Open the text remover

10 free credits on a new account · 720p output on the free tier

If you would rather do it by hand: Photoshop and friends

Photoshop, Affinity Photo, GIMP and Pixelmator all ship the same three tools, and they all work on the same principle: you tell the software which pixels are wrong, and it copies or invents replacements from the ones around them.

  • Content-Aware Fill — select the text, run the fill, and Photoshop rebuilds the area. Best when the background is repetitive: brick, sand, water, a plain wall.
  • Spot Healing Brush — quick dabs over small marks. Good for thin type on simple backgrounds, poor over faces and fine patterns, where it smears.
  • Clone Stamp — you choose the source pixels yourself. The most control and the slowest route: every stamp is a decision.

The honest trade-off: manual work gives you control over every pixel and takes minutes per image, not seconds. On a single photo where the text sits over someone's face, that control is worth it. On twenty product shots with a caption in the same corner, it is an afternoon.

Cropping is the other manual option, and the one people forget. If the text sits at an edge and the composition can take it, cutting 40 pixels off the bottom beats any repair.

Removing text from a picture on your phone

The browser route is the short one: open the tool in Safari or Chrome, tap the upload box, pick the photo from your camera roll, and press Remove Text. Same account, same credits, no app install — and the processing happens on the server, so an older handset is not a handicap.

Phone-only apps exist too. Snapseed's Healing tool and Photoshop Express' spot removal are the usual pair, and both are manual brush work: fine for a word on a wall, tedious for a full caption over a patterned shirt. iOS also has Live Text, which is the opposite operation — it reads text out of a picture rather than removing it, useful when you want the words rather than the pixels.

One phone-specific trap: screenshots. A screenshot of a chat or a post is a clean, sharp, untextured image, which makes removal easy — and also makes a bad fill more obvious, because there is no grain to hide behind. Check those at 100%.

Watermark, caption or date stamp: same problem, different jobs

People say "text" and mean three or four different things. The technique is similar; the rules around it are not.

A watermark or a stock mark

Usually not yours to remove unless you bought the licence or shot the photo. The mechanics are easy; the permission is the hard part. Details on the watermark remover page.

A caption or a social overlay

Text baked into the picture by whoever posted it — meme lines, quote cards, "repost" credits. Often the easiest case, and the one where attribution questions matter. See remove caption from image.

A date stamp

Orange digits from a 1990s camera, white overlays from an early digicam, or a "Shot on" line. Family-archive work rather than commercial work. See remove date stamp from photo.

A brand logo

Your own product, your own photo, or a supplier's image you are licensed to edit — that is a different job again. See remove logo from image.

An AI badge

The corner mark an image generator adds. Visible badge: removable. Hidden provenance metadata: not. The AI watermark remover page explains where the line sits.

Five things that make the result look clean

  1. Work on a copy. Keep the original file untouched; a rebuild you do not like is not reversible.
  2. Do not crop first. Removing the text and keeping the full frame gives the model context at the edges. Crop afterwards if you want to.
  3. Mark a little wide, not tight. With the brush, give the letters two or three pixels of margin. A mark that clips the edge of a glyph leaves half a letter behind.
  4. Judge at 100%, on a real screen. A patch that looks invisible on a phone at arm's length can be obvious on a 24-inch monitor — and the other way round.
  5. Expect a second pass on texture. Grass, hair, knitted wool and fine print are where every method, ours included, produces a plausible rather than a true fill.

Try it on the photo that has been annoying you

One upload, one button, one download. Ten free credits come with a new account.

Remove text from an image
FAQ

FAQ About Removing Text from Images

Yes — that is the main reason people look for a browser tool. You upload the picture, press one button, and a model finds the text and rebuilds what was behind it. No install, no subscription to an editor, and it works on a phone.
Screenshots are the easiest case because the text is sharp and the background is flat. Scans are usually fine too, though a low-resolution scan of a printed page can leave a slightly soft patch where the words were. Scan at 300 dpi or more if you have the choice.
On plain or gently graded areas you will not see the join. On busy texture — foliage, hair, patterned fabric — the fill is invented rather than recovered, so a close look at 100% shows it. That is true of every method in this guide, manual ones included.
JPG, PNG, WebP and HEIC up to 10MB. Images smaller than 512px on a side are scaled up before processing, and output runs at 720p on free credits or 1K and 2K on the paid plans.
One image per run — there is no batch queue yet. For a set of twenty, plan on twenty uploads; each takes a few seconds once the file is in.