# Remove text from an image

URL: https://dreambench.ai/tools/remove-text-from-image

A preset of the [AI object remover](https://dreambench.ai/tools/ai-object-remover): the same engine, set up for this one job.

Upload a photo or a design with words printed over it and the words come off: a caption, a date stamp, a sticker, a line of subtitles. The picture behind them is filled in from what surrounds it, and everything else is kept close to the original.

## Can AI remove text from an image?

Yes. It erases words printed over a picture, captions, date stamps, stickers and subtitles, and rebuilds what was under them from the surrounding picture. Leave the box empty to clear every piece of text at once. It is for scans, screenshots and your own designs. It is part of every DreamBench plan, from $9.99 a month.

## How to remove text from an image

1. Upload the image: A photo with a date stamp, a screenshot with a caption, a design with a sticker or a line of text you no longer want.
2. Say what goes, or leave it blank: Left empty, the run removes every caption, sticker and date stamp printed on the picture. The note is a prompt: say which text goes and which stays, with where it sits, to remove only that: the white caption along the bottom. Then press Remove it. The first time, create a free account and choose a plan.
3. Download the clean image: The text is gone and the picture continues through where it was, ready to download full size.

**Choices.**
- What to remove: an optional note (up to 200 characters)

## Remove a date stamp from an old photo

The orange date burned into the corner of a print or an early digital photo sits over a small area of sky, grass or wall, so the fill is usually hard to find. Scan the print in even light first, and restore it before or after if it is damaged.

## Remove captions and subtitles from a screenshot

A frame saved from a video or a meme with words across it, cleared so the picture can be reused or captioned again. Words over a plain area come off cleanest; words over a face or small print leave the model the most to guess.

## Remove text and keep the background

Everything outside the text is instructed to stay as it was, the framing, the colors, the people, and comes back close to the original. The area under the letters is rebuilt, not recovered, because the model never saw it, so check that patch before you use the picture.

## Keep some text, remove the rest

Say what stays as well as what goes: remove all text except the shop name above the door. The box takes up to 200 characters, enough to name both. Check the text you kept at full size, as small lettering can shift when the picture is drawn again.

## Replace the text with your own

Clear the old words here, then set new text over the clean picture in Canva, PowerPoint or any design app. The patch where the text was is filled in, so your new words can sit anywhere without old letters showing through.

## Works with

- What it takes: JPEG, PNG, WebP or HEIC, up to 40 MB, straight from a phone or a camera. An iPhone's HEIC photo becomes a JPEG on your device as you choose it, in any browser. A picture larger than 2048 pixels on its long edge is scaled to 2048 before the run.
- What you get: A JPEG in your photo's proportions.
- Canva and PowerPoint: Place the clean picture in a design and set new text over it.
- Instagram and messaging apps: Repost a photo without its old caption, in the same proportions.
- Family albums and archives: Old prints and scans without their date stamps, ready to print again.

## Which model runs it, and where it fails

The Text remover runs on [Nano Banana 2](https://dreambench.ai/models/nano-banana-2). Nano Banana 2, also called Gemini 3.1 Flash Image, is an image model from Google DeepMind. It takes your photo and one instruction and draws the picture again with the change made. Google says it can still get spelling, small faces and fine detail wrong, so check them in the result. Every image it makes carries SynthID, Google's invisible watermark that marks it as AI-generated. A run the model declines under its content rules is refunded.

- Set up for text: One click removes all the printed text in the picture: captions, subtitles, stickers, date stamps and your own watermark. No brush, no mask, no selecting letters one by one.
- Filled from what surrounds it: The area the text covered is rebuilt from the surfaces, colors and light around it, so sky stays sky and a wall stays a wall. Text over a plain area comes off cleanest.
- Signs come off blank: Lettering that is part of the scene, a shop sign or a poster, can be removed too, but the board stays and is left blank. If the whole sign should go, name the sign rather than the words on it.

**Where it fails.**
- A face, small print or a fine pattern under a caption is rebuilt from its neighbors, not recovered: it is a guess.
- Large blocks of text covering much of the picture leave little to rebuild from, and the fill can look soft.
- The whole picture is drawn again, so fine detail elsewhere can drift slightly; compare before and after.

## Text remover FAQ

- **How do I remove text from an image?** Upload the image and press Remove it. With the box left empty, every caption, sticker and date stamp on the picture is erased and the picture behind it is filled in. To remove one piece only, name it and where it sits. Make the edit with any DreamBench plan and download it full size.
- **Can I remove a date stamp from a photo?** Yes. A date stamp sits in a corner over a small area, so the model fills it from the sky, grass or wall around it and the join is usually hard to find. Scan or photograph the print in even light, upload it, and run with the box empty; every stamp and caption on it goes in one pass.
- **Can it remove a shop sign or lettering that is part of the scene?** Yes, but only the lettering goes: the board stays, with its own surface given back, so a shop front ends up with a blank sign. The same is true of a poster or a license plate. If the whole sign should go, name the sign and where it is rather than the words on it.
- **Is it legal to remove a watermark or copyright text from an image?** Removing a watermark doesn't change who owns an image, and removing another person's watermark or credit from their work can infringe their copyright; in the United States, stripping copyright management information is covered by 17 U.S.C. § 1202. It is fine on images you own or have permission to edit, such as your own caption or date stamp. A creator's handle on their clip is a credit, not a caption.
- **How do I remove text from an image without deleting the background?** To remove text from an image without deleting the background, run an AI text remover: it erases the caption, date stamp or sticker and rebuilds the area under the letters from what surrounds it, so sky stays sky and a wall stays a wall. Plain surfaces come back cleanest; check text over a face or small print.
- **Why is there a soft patch where the text was?** The area under the letters is rebuilt from its neighbors. Over sky, a wall or grass that leaves no trace; over a face, small print or a fine pattern the model has to guess, and the guess can look soft. Run it again naming the text and where it sits, and compare at full size.
- **Will removing text affect the quality of the image?** Removing text with an AI tool affects quality slightly: the picture is drawn again at the image model's own size, its 2K tier over about 1.5 megapixels and its 1K tier below, so a large upload can come back with fewer pixels, and fine detail away from the text can drift a little. For a large print, enlarge the clean image with the image upscaler afterward.
- **Can I remove text from a PNG and keep it a PNG?** No. The result is a JPEG in the picture's proportions, so a transparent PNG comes back on a flat background and the transparency is lost. For a logo or a graphic that must stay transparent, remove the text in a layered editor instead; for a photo saved as PNG, the JPEG is fine.
- **Is there a way to remove text from an image on iPhone and Android?** Yes. This page works in a phone browser, with nothing to install: upload the picture and leave the box empty to clear every caption. Clean Up in Photos on iPhones with Apple Intelligence and Magic Eraser in Google Photos do it with a brush, by painting over the words, and Canva's Magic Eraser does the same in its editor.
- **How is this different from Photoshop's Content-Aware Fill?** Content-Aware Fill fills a selection you draw around the letters with texture from nearby, and only that selection changes; large or busy text needs the Healing Brush afterward. This tool finds the text itself, with the box left empty or from a description, and redraws the picture with the text gone, so nothing is selected by hand but the rest of the picture is a close re-render.
- **Can it remove text from a video?** No. It cleans one still image at a time. Subtitles burned into a video have to be removed frame by frame, which this tool doesn't do; a single frame saved as a picture can be cleaned like any image.
- **Why are there odd letters where the text was?** Odd letters where the text was are drawn, not recovered: the model fills the gap from what surrounds it, and next to other lettering or a fine pattern it can draw letter-like shapes, since the image model can get spelling and fine detail wrong. Run it again naming the text and where it sits, and say what the surface should be: leave the wall plain.
- **Should I use the text remover or the AI object remover?** Use the text remover for words printed over a picture: with the box left empty it clears every caption, sticker and date stamp, and fills in the picture behind them. The AI object remover is the same tool without that preset, so it needs a description; pick it for a thing rather than words, such as a sign board with its post.
- **Can I remove text from an image without paying?** No. Every run needs a DreamBench plan; you create an account on the way, and the button shows the price in credits before you press it. The tools come with every plan, alongside the video models, and a run that fails is refunded.
- **Can I remove text from several images at once?** No. The tool takes one image per run, and there is no batch upload. Run it again for the next one, with the box empty or its own note: each run is priced separately, and the button shows the price before you press it.
- **Does the clean image have a watermark of its own?** No visible one. A DreamBench tool adds no watermark to an image it returns, on any plan. The image model that fills the gap embeds SynthID, Google's invisible mark that identifies a picture as made or edited by AI; it changes nothing you can see, and the download is the clean image full size.
- **How much does the Text remover cost?** A run starts at 7 credits from your DreamBench plan, depending on the photo, and once you have a plan the button shows each run's price before you press it. Starter, $9.99 a month, includes 1,100 credits a month, at least 122 runs of the Text remover, plus AI video. A run that fails is refunded.
- **Do I need an account to use the Text remover?** Yes: a free account and a DreamBench plan. The tools are part of every plan, from Starter at $9.99 a month.
- **Which image files does the Text remover take?** A photo up to 40 megabytes: JPEG, PNG, WebP, HEIC or another format your browser opens. A photo straight from a phone is fine, an iPhone's HEIC included: the page scales a large picture down to 2048 pixels on its long edge before the run, so the result is never larger than the page says.
- **Can I use what the Text remover makes commercially?** Yes. Under the DreamBench Terms you may use what a tool makes for personal and commercial purposes, including advertising, monetized content and work for clients, with no separate commercial-use upgrade. The Terms don't guarantee copyright in a result or clear another person's rights in your upload, so only use photos you have the rights to.
- **Is my photo used to train AI?** No. DreamBench doesn't use your prompts, uploads or videos to train or fine-tune AI models, and any future training program would be opt-in.
- **What happens to my uploaded photo?** It stays in your browser until you make a result. Then it is used to make that result and nothing else, and removed from the tool's storage after seven days along with the result. A result you keep in your library stays as long as you keep it.

Runs on Nano Banana 2. DreamBench is an independent platform; its tools come with every plan (https://dreambench.ai/pricing). Copy last checked 2026-10-22.
