Review before anything changes
Find proposes boxes. Click one on the image, or untick it in the list, and it is left alone; nothing is covered until you press Cover.
Private image tools
Find the text in a photo and cover it
Street signs, a name badge, a screen in the background, an address on a parcel: a photo often carries text you did not mean to share. This page finds the text regions for you, shows you every one, and covers the ones you keep.
It finds text; it never reads it. The model is a 2.3 MB text detector that runs on your device and returns boxes, not words. There is no recognition step on this page at all, so nothing your image says is ever extracted.
Result proof
A copy of your photo with the text you chose covered
01
Input
Any photo or screenshot
02
Local pass
A pinned PP-OCRv3 text detector marks where text is, you review the proposed boxes, and each box you keep is filled, blurred or pixelated at a size set by its own letters
03
Result
A PNG at the size the page worked at, which is your image's own size unless the page said it scaled it down, with the covered boxes changed and every other pixel as it was
How to verify it: The caption under the result says how many boxes were covered and with what. Open the result at full size and read across it: anything still legible is either a box you switched off or text the detector missed, and both are visible to you before you share it.
Detection, not recognition
It finds where text is and has no way to read it
A pinned PP-OCRv3 text detector marks the regions that look like writing and hands back boxes, nothing else. There is no recognition step on the page, so no word, number or name is ever extracted. Every box is shown before anything is covered.
2.3 MB
Pinned detector, first use only
Sized to the letters
Blur and pixelation scale with the height of the text in each box, so a caption and a shop sign are treated alike rather than by one fixed radius.
Solid box by default
The default replaces the pixels with a flat box, the one treatment that leaves nothing of the lettering in the file.
Know before you process: In our own check it missed crumpled print a few pixels tall and boxed a row of distant trees with high confidence. Read across the result at full size before you share it.
What people use this for
Screenshots
Names, email addresses and account numbers in a screenshot are crisp, upright text. On the screenshot and the contact card in our check, every line was boxed.
Street and travel photos
House numbers, street signs and shop names that place a photo somewhere you would rather not say.
Badges, labels and parcels
Name badges at an event, a delivery label, a prescription box on the kitchen counter.
Documents on a desk
A letter or an invoice caught in the corner of a photo. Angled pages get generous boxes, which is the safe direction to be wrong in.
Limits worth knowing
- It can miss text. In our check a pile of crumpled receipts produced no boxes at all, and small faint marks on an angled invoice were missed. Always read across the result before sharing it.
- It can box things that are not text. A row of distant trees was boxed with high confidence in our check. Every box is shown for review first, so switch those off.
- Boxes are upright rectangles, so a slanted line gets a box larger than the words.
- Blur and pixelation are not removal. They keep an average of the original pixels, and pixelated text has been recovered by published techniques. Use Solid box for anything sensitive.
- It covers text and nothing else. Faces, plates, reflections, metadata and context can still identify a person or a place. Privacy Blur handles faces and plates; the Metadata Remover handles EXIF and GPS.
- An image larger than the working limit this device can hold safely is scaled down to that limit first, and the page says so when it happens.
How the local workflow runs
- Choose the image. It stays on your device.
- Press Find text. The first time, this downloads the 2.3 MB detector, plus the shared engine that runs it if this browser does not have it yet.
- Look at every red box. Click any that are not text, or untick them in the list.
- Choose Solid box unless you have a reason not to, then press Cover.
- Read across the result at full size, then download the PNG.
Frequently asked questions
Does it read my text?
No, and it cannot. The model is a text detector: it outputs a map of where text is likely to be and nothing else. There is no recognition step anywhere on this page, so no word, number or name is ever extracted, stored or sent. The boxes are the whole output.
What does it download?
One 2.3 MB model file, the PP-OCRv3 text detector published by OpenCV under the Apache 2.0 licence, fetched from a pinned version the first time you press Find text and then cached by your browser. The file is checked against a fixed fingerprint before it is used. The first run also fetches the shared ONNX Runtime engine that runs it, up to about 24 MB, unless this browser still holds a copy from another tool on this site or from an offline pre-download. Your image is never uploaded.
Which treatment should I use?
Solid box, for anything that matters. It replaces the pixels, so nothing of the lettering is left in the file. Blur and pixelation keep an average of the original pixels; they stop a casual reader, and published techniques have recovered pixelated text from screenshots. Both are sized to the letters in each box, so they are much stronger than a fixed blur, but they are still not the same as removing the text.
Why did it box something that is not text?
Because to a text detector, some patterns look like writing. In our own check a row of distant trees on a hillside was boxed with high confidence. That is why every box is shown before anything is applied: click it, or untick it in the list, and it is left alone.
Why did it miss some text?
In our check it missed print only a few pixels tall on crumpled paper, and small faint marks on a page photographed at an angle. Try Fine, which looks at a large photo in more detail. If it still misses something, Privacy Blur has the same Text option plus boxes you draw yourself.
Why are some boxes bigger than the words?
The boxes are upright rectangles. On a slanted line of text, such as a document photographed at an angle, the upright box that covers the line also covers some of the page around it. For redaction that is the safe direction to be wrong in.
Does my image leave my device?
No. The image is decoded, searched and covered in this tab. The only network requests are the one-time downloads of the detector and, if this browser does not have it yet, the engine that runs it, and neither carries anything of your image.
Related tools
Privacy Blur
Faces, licence plates and text in one review, plus boxes you draw yourself.
Metadata Remover
Strip the location and camera details a photo carries alongside its pixels.
Object Remove & Replace
Rebuild an area instead of visibly covering it.
Image Editor
Crop, adjust and export once the text is covered.
Ready to try it?
Nothing is uploaded, and every box is yours to review before anything changes.
Find and cover text →