AI Text Remover: Remove Text From an Image

iWhat an AI text remover does

The photo is fine and the lettering on it is not. An orange date burned into the corner of a print from a shoebox. A phone number somebody typed across a listing shot. A caption strip along the bottom of a screenshot. The logo you put in the corner of your own pictures three years ago, before the name of your shop changed. A text remover takes those characters off and puts back the surface they were sitting on.

The first screen of this page is an editor, not a form. You upload the frame, it opens across the full width, and from then on you work on the picture itself: press once on the lettering, look at the outline the model draws around it, and apply the edit. There is no box asking you to type the words out — the click has already said which lettering you mean.

What comes back is your photograph with one area rebuilt, not a new picture of the same scene. The next two sections are about exactly what that means, because it is the part people get burned on.

★Click the lettering instead of painting over it

Nearly every tool that erases text from photos opens the same way: choose a brush, set its size, and paint over the words yourself, carefully, along their edges. That step exists because the software has no idea what it is looking at — it needs you to mark the pixels for it. On an object that is tedious. On type it is worse, because type is mostly thin strokes and holes: the gap inside an O, the tail of a Q, the space between a date and its slashes, the faint drop shadow behind white letters on a bright sky.

Here the click does that work. Tap the text and the model reads the frame, works out where the characters end and the surface begins, and outlines the whole line for you. The hint under the frame says as much: Tap an object — the model outlines it. It is the gesture your phone has already trained you into — point at the thing, not around it — and it runs in a browser on any laptop or phone.

The brush has not gone anywhere. It has simply stopped being step one, and it is there for the correction that takes a second or two:

Zoom in before you commit. A text remover is only as good as the outline it starts from, and the classic mistake is a selection that stops at the baseline: the body of the word goes, and the descenders of the g and the y stay behind in the picture. Nothing touches the photo while you do any of this — until you press the button, you are only drawing a selection.

What ends up where the text was

There are two cheap ways to make writing disappear, and everyone has seen both. One is to smear the area until the letters are unreadable, which leaves the soft grey blob that gives the edit away from across the room. The other is to stamp a rectangle of neighboring pixels over it, which leaves a patch of wall that is visibly a copy of the wall beside it. The blurred patch is the thing people are actually afraid of when they come to a page like this, and it is worth saying plainly that it is not the method here.

The model works from the other direction. It reads what surrounds the selection and continues it through the gap, so a sky keeps its gradient, a brick wall keeps its courses and its mortar lines, a lawn keeps the direction it was mowed in, a knitted sweater keeps its stitch, a wooden table keeps its grain. Semi-transparent lettering takes its own soft shadow along with it when that shadow falls inside the selection.

Now the honest half. What appears in the gap is a plausible continuation of the surroundings, not a recovery of what was underneath. Nothing ever recorded the pixels the characters covered, so they cannot be read back — they are drawn from the parts of the picture that were recorded. Over sky, water, asphalt, grass, a painted wall or a plain shirt, the difference is invisible. Over a face, over fine print, over a patterned rug, it is a guess, and a guess is worth looking at closely before you send the picture anywhere.

One phrase is worth untangling here, because it gets typed two different ways. When people ask to remove text from an image without the background, they almost always mean the text gone and the background left alone — that is this page, and a clearer way to put it is without affecting the background. If what you actually want is the background itself gone, leaving the subject on a transparent cutout, that is the background remover, and a different scene behind the subject is change background.

1How to remove text from a photo in three steps

Three steps, and the middle one is a single press.

The first word matters more than the rest of the sentence. An instruction that opens with Remove, Delete or Erase is treated as a clearing job, and the model is told to fill the selection with the surface behind it and put nothing new there. Start the sentence any other way and it will try to change what you selected instead of taking it out — which is a useful thing to do, but it is a different tool. Beyond that first word you do not need to be clever: "Remove it" is a complete instruction.

Remove text from a screenshot

A screenshot is the one situation where the surface under the writing is not a photograph at all. It is a flat panel: a solid header, a white card, a grey chat bubble, a block of brand color. For a model whose whole job is to continue what surrounds the gap, that is about as easy as the task ever gets, which is why screenshots come out cleaner than most people expect.

What people are usually taking off is a name at the top of a conversation, a phone number or an email address in a listing, an address bar with a personal link in it, a promotional banner across a game screen, a timestamp under a message. Click the line, say Remove, and the panel closes up behind it.

Where it gets harder is text that is inside other text. A single line lifted out of a paragraph leaves a gap surrounded by writing on every side, and the model is continuing a surface that is itself made of characters — which is the one condition under which it may put something letter-shaped back. Short, isolated, on its own patch of background: good. A sentence in the middle of a wall of print: expect to look at it, and to run it a second time with a tighter selection.

And one caution worth more than any tip. If you are stripping a name or a number out of a screenshot before it goes somewhere public, check the result at full size. This is an image edit, not a redaction tool: it changes what the picture shows, and where you need the information to be provably and permanently gone, a solid box drawn over it is the blunter and safer instrument.

Remove a caption from an image

Captions are a bigger job than dates for one simple reason: they are big. A headline set across the top of a picture, a two-line strip along the bottom, subtitles burned into a frame you grabbed from a clip, the impact-font line on a meme somebody sent you. The characters are large, and large characters cross several surfaces at once — sky, then a roofline, then somebody's shoulder, then a face.

The way through it is not one enormous selection. Take a line at a time, starting with whichever one sits over the plainest background. Each pass becomes its own entry in Versions, so the second removal can be undone without losing the first, and the third can be undone without losing either. On a caption, two careful runs beat one greedy selection nearly every time.

Before you start, check whether the caption is actually on the picture. A lot of captions sit on their own band of solid color added below or above the frame, and if that is what you have, cropping the band away is faster, free of guesswork, and keeps the original sharpness. Editing is for text that is genuinely lying on top of the photograph.

Remove a signature from a photo you own

Handwriting behaves differently from printed type. A signature is thin, angled, often semi-transparent, and it usually runs across a corner where the background is doing something busy. That makes it a middling case: harder than a block of date digits, easier than a caption in a slab font. It works, and it more often wants a second pass with the selection widened over the tail of the last letter.

The versions of this that belong here are the ones where the picture is yours: your own signature in the corner of your own work, a photograph of a piece you made and signed and now want a clean copy of for a portfolio, a studio mark on a photo you own outright. Same routine as everything else on this page — click the writing, say Remove, check the corner at full size.

There is a version that does not belong here, and it is worth being blunt about. A signature that is doing a job — on a contract, a form, an authorization, an identity document, anything somebody else is going to rely on — is not decoration, and taking it out of a photo of that document is forgery rather than editing. Do not use this for that. Whose picture it is in the first place is the subject of the rights question in the FAQ below.

Difficult surfaces: text on texture, on a pattern, on a face

How a run goes depends far less on how much text there is than on what is underneath it. The model is rebuilding a surface, so the question is always the same: how predictable was that surface, and how much of it is still visible around the edges of the selection?

The alphabet itself makes no difference. Nothing here is reading the words or understanding the language — it is clearing a set of shapes off a surface, which is why Latin letters, Cyrillic, Greek, Arabic, kanji, numerals and symbols are all the same kind of job. What changes the odds is size and contrast: crisp characters on a quiet background, in an original that is big enough for them to be more than a few pixels across.

Formats, free edits and limits

The uploader takes JPG, PNG and WEBP, up to 10 MB per file, and says so on the panel: JPG, PNG, WEBP — up to 10 MB. You will select the object right on the photo. HEIC and HEIF straight off an iPhone are not accepted — set the camera to "Most Compatible", or share the picture to yourself once, which converts it to JPG on the way.

There is no minimum resolution, and no maximum on the pixel dimensions of what you send. Large files are fitted to a working size of about 2048 px on the long side when they upload, so a frame straight out of a modern camera comes back a little smaller than it went in; anything already below that keeps its own size. The crop, the framing and the aspect ratio are never changed either way. If the picture you end up with is smaller than you need it, the image upscaler is the next stop.

There is also a pace limit of up to ten edits a minute on the account, which ordinary work never comes close to.

Free edits and what they cover

Your first edit is free once you are signed in, and no plan is needed to make it — press the button while it still reads Try it free. A new account gets two free generations in total, not two a day and not two a month, and they are shared with the other tools on the site. After that, edits run on a paid plan or a pack, and the terms are on the pricing page. Saving is open on this tool from the beginning: Download frame hands you the edited picture from your very first run, with no purchase in between, so the free edit is a file you keep rather than a preview you look at.

Do I need an account?

Yes, and here you are asked for it early: pressing the upload zone as a guest opens the sign-in window straight away, before the editor takes a photo at all. That is not a formality — your frame is uploaded so the model can work on it, and every pass is stored as a version against your account. It is also what lets you come back to an edit tomorrow, put it next to Original, and open it from another device.