Pixelate and blur faces, license plates and text in a photo, then save at the original size
Pixelate & Blur Image is a tool for choosing the parts of a photo that must not be recognizable and covering them yourself. Drag across the photo to draw a rectangle or an ellipse, or paint with the brush, and that spot is immediately turned into pixelation, blur or a solid fill, whichever you picked. Each hidden area has its own method and strength, so in one photo you can pixelate a face and cover a phone number with black.
Everything is processed on a canvas inside your browser, and the photo is never uploaded to a server. The saved file has the same size as the original; only a very large photo whose pixel count is over 4,096 × 4,096 is scaled down, to stay within what phone browsers can handle. The saved file is created fresh on the canvas, so the photo data of the original, such as location, camera and capture time, is not carried into it.
When to use it
- Posting a parcel photo when selling second-hand or leaving a review: a shipping label carries the recipient's name, phone number, address and tracking number all at once. The sample photo that opens first is exactly this case: the name is covered with a fill and the barcodes with pixelation.
- Sharing a screenshot of a chat, an order or a payment: the other person's name and profile picture, an account number and an order number are scattered across the screen. Paint along the lines with the brush, or draw several rectangles to cover them in one go.
- Publishing a group or event photo: cover only the faces of people who have not agreed, or of children, with ellipses. An ellipse that follows the shape of a face touches less of the background than a rectangle.
- Posting road, parking lot or dashcam photos: other cars' license plates and passersby's faces are caught in the frame. Why a fill is the safe choice for plates, which are made of letters and numbers, is explained below.
- Sending a photo of a document or ID: leave only what the other side needs to check, and cover the rest, such as the Social Security number, the address or the portrait. You can also open several photos and get them as one ZIP.
How to use it
- Open a photo: press Choose photo, drag a photo onto the work area, or paste a copied image. Up to 12 photos can be open at once, and a photo strip appears when there are 2 or more. If a JPG contains location data, a tag under its name tells you.
- Pick a tool: the rectangle suits lines of text and license plates, the ellipse suits faces, and the brush suits slanted lines or irregular shapes. When you pick the brush, a brush size slider appears.
- Pick the method and strength: for pixelation set the block size, for blur set the blur radius, and for fill choose black, white or the surrounding color. If a hidden area is selected, changing a setting changes only that area; if nothing is selected, the setting applies to what you draw next.
- Draw and adjust: drag across the photo to draw. A single click creates a rectangle one sixth of the short side at that spot. Drag a drawn area to move it, and drag any of its four corners to resize it. On the keyboard, Enter makes a new rectangle, the arrow keys move it, + and − resize it, Delete removes it and Ctrl+Z undoes.
- Check at actual size: the photo on the stage is shrunk to fit the screen, so it looks denser than it really is. The check window below shows the selected area at actual size (100%), together with a verdict and a block ruler that tells you how much is hidden.
- Save: choose PNG or JPG and press Save. With several photos, Save all (ZIP) gives you everything at once, and Copy or Share lets you paste or send the result directly.
Pixelate, blur and fill leave different amounts of information
All three methods look like "hidden", but the amount of information left in the saved file differs. Pixelation keeps an average color for every block, and blur keeps the flow of outline and brightness. Only fill replaces the original pixels with a single color and leaves nothing. The color strip in the list and the verdict window shows this difference as green (fill), yellow (pixelation with few blocks) and red (blur, or pixelation with many blocks).
| Method | What it does | What stays in the saved file | Where it fits |
|---|---|---|---|
| Fill | Replaces the original pixels with black, white or the average surrounding color | Only the shape and size | Names, phone numbers, addresses, account numbers, license plates, barcodes, 2D codes |
| Pixelate | Replaces the pixels inside each block of a set size with one average color | An average color per block (more remains when blocks are small) | Faces and human figures (with only a few blocks across the short side) |
| Blur | Mixes each pixel with its neighbors so it spreads out softly | Outline, and the flow of brightness and color | Softening a background so it looks tidy |
Research has confirmed why text should not be hidden with pixelation or blur. Hill et al. (2016) recovered text from documents hidden with pixelation and blur using hidden Markov models, and concluded that although both methods are widely used, they are not suitable for hiding text. They recommended that, for text in images, a solid bar is the best choice in all cases that matter. There is also a public tool, Depix, that recovers text from pixelated screenshots. Depix works by knowing the font and text size and comparing against samples pixelated in the same way.
A fill in an image is safe because it differs from a black box in a PDF. In a PDF, a box laid over text can leave the text itself in the file, but this tool builds and saves a new image in which the pixel values of the filled area have been replaced, so the original pixels are not in the file.
Judge pixelation by the number of blocks across the short side
With the same block size, a large photo is hidden weakly and a small photo strongly. So besides the block size (px), this tool shows how many blocks fit across the short side of the hidden area. If a face is 300px wide and the blocks are 50px, that is 300 ÷ 50 = 6 blocks; with 20px blocks it is 15 blocks. For an area painted with the brush, the brush size counts as the short side.
The reference points of the block ruler come from the experiments of McPherson et al. (2016). In that study, neural networks were trained on pixelated face photos to tell who a pixelated face belonged to. The FaceScrub data (530 people) was pixelated after fitting faces to 224 × 224px, and the AT&T data (40 people) used 92 × 112px images with the face in the centre as they were, so for AT&T the block count is per image width and slightly lower for the face alone.
| Data (candidates) | Face image width | Block | Blocks across | Identified | Random guess |
|---|---|---|---|---|---|
| FaceScrub (530 people) | 224px | Original | - | 75.49% | 0.19% |
| FaceScrub (530 people) | 224px | 8 × 8 | 28 | 65.25% | 0.19% |
| FaceScrub (530 people) | 224px | 16 × 16 | 14 | 57.56% | 0.19% |
| AT&T (40 people) | 92px | 8 × 8 | 11.5 | 95.00% | 2.5% |
| AT&T (40 people) | 92px | 16 × 16 | 5.75 | 96.25% | 2.5% |
| AT&T (40 people), YouTube face blur | 92px | Blur | - | 57.75% | 2.5% |
Two things can be read from the table. First, even pixelated to 14 blocks across, more than half of the 530 people were identified, so with more than 14 blocks much of the face outline remains. This tool marks that case red (much of the shape remains). Second, with only 40 candidates, 96% were identified even when reduced to a little under 6 blocks. If the people who will see the photo already know the face, no block size will stop them from recognizing the clothes, the hairstyle or the people next to it, so for such photos a fill is the safer choice. The study measured the share identified by a trained machine, not by people, and that is not the same as the share people identify.
How this tool hides an area
- Pixelate: with block size B, the pixel at horizontal position x falls in the block numbered x ÷ B rounded down. The grid starts at the top-left corner of the photo, so block edges stay aligned even when hidden areas overlap or touch. The red, green and blue of the pixels in a block are averaged with opacity as the weight, so edges do not bleed dark even in a semi-transparent PNG.
- Blur: a Gaussian blur is approximated with three box blurs. The box width follows Kovesi (2010), √(12σ²/3 + 1), where σ is the blur radius you see on screen. The W3C filter specification also states that three box blurs approximate a Gaussian to within about 3% for a standard deviation of 2 or more. When the blur radius is 4px or more, the area is shrunk to 1/f (f = ⌊σ ÷ 2⌋, at most 8), blurred with σ ÷ f, and enlarged again; σ after shrinking is still 2 or more, so the same approximation condition holds. Safari does not support the built-in canvas blur filter by default, so the tool does the calculation itself to give the same result.
- Fill: the area is filled with black, white or its own average color. The surrounding color is for covering something less conspicuously.
- Overlaps: areas are applied in list order on top of the result of the earlier ones. Even if you draw a wide pixelated area over a name you have already filled, the pixelation is computed from the filled pixels, so the name does not show again.
Pixelation block number = ⌊x ÷ B⌋, block color = Σ(color × opacity) ÷ Σ opacity, blocks across the short side = short side ÷ B
Example: with 25px blocks, the pixel at x = 318 is in block ⌊318 ÷ 25⌋ = 12, which runs from x = 300 to 324. The contact box in the sample photo is 64px tall, so it has 64 ÷ 25 = 2.56 blocks across the short side. With a blur radius σ = 12px the formula gives box widths of 23, 23 and 25px, but since σ is 4px or more the area is scaled to 1/6 and blurred with σ = 2 (boxes of 3, 3 and 5px).
The preview on the stage applies the same calculation to a copy shrunk to fit the screen, while the check window and the saved file are calculated again at the original size. Block edges land in the same place in terms of original pixels, so what you saw in the preview is what is saved.
Photo data (location, device, date) is not kept in the saved file
Smartphone photos carry photo data called Exif. The Exif standard (CIPA DC-008) puts location in a GPS information block (tag number 34853), and a phone with location recording on writes the latitude and longitude of where the photo was taken there. Even if you hide the face and the address, if this data remains, the location of the home or school where the photo was taken can travel with it.
When you open a JPG, this tool reads its Exif and, if there is location data, a camera or a capture date, shows them as tags under the name. The saved file is created fresh on the canvas, so this data is not carried over. Even with no hidden areas, the saved file has no photo data, so you can also use this tool to save a photo again just to remove its location. Note that if you send or upload the original file separately, its data goes along unchanged.
Save formats and size
| Format | Compression | Transparent areas | Suited to |
|---|---|---|---|
| PNG | Lossless | Kept as they are | Screenshots, images with a lot of text, transparent backgrounds (block edges stay crisp) |
| JPG | Lossy (quality 0.92) | Filled with white | Photos taken with a camera (smaller files than PNG) |
The initial format follows the photo you open: JPG for a JPG photo, PNG otherwise. If you change it yourself, the format you chose is used from then on. The saved size is the same as the original. MDN notes that iOS devices limit the canvas to 4,096 × 4,096px, so a photo over 16,777,216 pixels is scaled down to fit within that while keeping its aspect ratio, then edited and saved at that size, and you are told so when you open the photo. Files over 30 MB, with a side over 16,384px or with more than 50 million pixels are not opened.
Common mistakes
- Choosing blocks from the small preview alone: when a large photo is fitted to the screen, the blocks look denser than they really are, and it feels as if enough is hidden. Look at the 100% view in the check window and at the number of blocks across the short side.
- Hiding letters and numbers with blur: blur looks clean, but the outline of the characters remains. Use a fill for names, phone numbers and license plates.
- Missing the edges: a stroke or two at the end of a line of text, the first character of a plate, or an ear and the hair next to a face often stick out of the rectangle. Draw a little more generously than the area you need to hide.
- Missing reflections and other spots: check faces reflected in windows and mirrors, the screen of a monitor in the background, name badges, and a second label on the side of a parcel.
- Sending the original file along: sending a photo as a "file" in a messenger, or uploading the original to the cloud, passes on the photo from before it was hidden, with its photo data. Send only the new file you saved.
Limits and cautions
- It does not find faces or text automatically. You must draw the areas to hide yourself, and before saving you should look for anything you missed using Show original and the check window.
- There is research that tries to estimate the original from pixelation and blur. The reference points of the block ruler come from the face identification results of one study, and hiding to below those numbers is no guarantee that no one can recognize the person.
- HEIC photos open only in Safari 17 and later, and Chrome, Edge and Firefox cannot open them. Animated GIFs are handled as their first frame only.
- Calculation uses the browser canvas defaults (sRGB, 8 bits per channel). A photo with a wide color gamut may lose a little of its most saturated colors when saved.
- When you close the page, the opened photos and the record of hidden areas are gone. This tool never changes the original photo, so it stays on your device as it was.
Basis: on 2026-10-20 the relevant sentences and tables of the papers, standards and documents below were checked. The numbers in the study table and the reference points of the block ruler are the same as in this tool's calculation.
Sources
- 01McPherson, R.; Shokri, R.; Shmatikov, V. (2016). Defeating Image Obfuscation with Deep LearningarXiv 1609.00408, section 5.2 photo sizes (FaceScrub 224 × 224, AT&T 92 × 112), table 2 pixelation and table 3 YouTube blur accuracy
- 02Hill, S.; Zhou, Z.; Saul, L.; Shacham, H. (2016). On the (In)effectiveness of Mosaicing and Blurring as Tools for Document RedactionProceedings on Privacy Enhancing Technologies 2016(4), 403–417, recommends a solid bar for hiding text
- 03Depix — proof-of-concept tool that recovers text from pixelated screenshotsScreenshots pixelated with a linear box filter, compared against samples in the same font
- 04CIPA DC-008 — Exif 2.32 (2019)GPS Info IFD Pointer tag 34853 (0x8825), orientation tag 274
- 05MDN — HTMLCanvasElement.toBlob()Default type image/png; the quality value applies only to lossy formats such as image/jpeg and image/webp
- 06MDN — <canvas> maximum sizeiOS devices limit the canvas to 4,096 × 4,096px; drawing does not work beyond that
- 07W3C — Filter Effects Module Level 1, feGaussianBlurFor a standard deviation of 2 or more, three box blurs approximate a Gaussian to within about 3%
- 08Kovesi, P. (2010). Fast Almost-Gaussian FilteringDICTA 2010, 121–125, ideal width √(12σ²/n + 1) for n averaging filters
- 09Can I use — HEIF/HEIC image formatSupported only in Safari 17 and later (macOS and iOS); not in Chrome, Edge or Firefox
Frequently asked questions
Is my photo uploaded to a server?
No. The photo is read and processed only on a canvas inside the browser where this page is open, and it is not sent to any server. When you close the page, the opened photos and the record of hidden areas are all gone.
Should I use pixelate, blur or fill?
Use fill for text, numbers and barcodes. A study on document redaction (Hill et al., 2016) recovered text hidden with pixelation and blur and recommended a solid bar. For faces, use large pixelation with only a few blocks across the short side, or fill; blur is for tidying a background softly.
Can a pixelated photo be restored?
The saved file does not contain the original pixels of the hidden areas, so this tool cannot restore them. However, there is research that estimates the original from the remaining block colors and outlines in pixelation and blur. In one result, 57.6% of faces pixelated to 14 blocks across were identified out of 530 candidates, so use fill if something must stay hidden. Before you close the page, undo takes you back up to 60 steps.
What block size should I use for pixelation?
Look at how many blocks fit across the short side of the hidden area rather than at the pixel value. If a face is 300px wide and the blocks are 50px, that is 6 blocks. In the study, a machine identified 57.6% of faces with 14 blocks out of 530 and 96.3% of faces with about 6 blocks out of 40, so for photos seen by people who know the subject, make the blocks larger or use fill. The block ruler shows the block count of the area you have selected.
Does saving make the photo smaller?
It saves at the same size as the original. Only a photo over 16,777,216 pixels (4,096 × 4,096) is scaled down, keeping its aspect ratio, to fit the canvas limit of phone browsers, and you are told the reduced size when you open the photo and when you save.
Is the location data removed too?
Yes. The saved file is created fresh on the canvas, so the photo data (Exif) of the original, such as location (GPS), camera and capture time, is not included. When you open a JPG, a tag first shows whether the original has location data.
Can I process several photos at once?
You can open up to 12 photos at a time, switch between them on the photo strip and hide areas in each separately. Press Save all (ZIP) to get the results in one archive. Photos with no hidden areas are included too, as files without photo data.
Can I use it on a phone?
Yes. Drag over the photo with your finger to draw, and a single tap creates a rectangle there. Dragging outside the photo moves the view. An iPhone HEIC photo opens in Safari but may not open in other browsers such as Chrome, so convert it to JPG first in that case.