Can You Un-Pixelate an Image? What Really Works
You cannot truly un-pixelate an image: the detail is gone. But pixelated text can often be guessed, so hide anything sensitive with a solid block.
Open the free Pixelate Image tool
What pixelation does to a picture
Pixelation divides an area into a grid of square cells and paints each cell with a single colour, usually the average of the pixels it covers. Everything that made the original area recognisable, such as the edges of letters, the shape of an eye or the curve of a number, is collapsed into that one colour per cell.
That averaging throws information away. A cell that started as 16 by 16 pixels held 256 separate colour values; after pixelation it holds one. Many different originals produce exactly the same cell, so there is no formula that can run the process backwards and return the picture you started with.
This is why every honest answer to the question "can I un-pixelate this?" starts with no. What is possible is narrower and more interesting: someone can sometimes guess what was there, and for certain kinds of content that guess can be uncomfortably good.
Why some unpixelate tools seem to work
Apps that promise to unpixelate or enhance a photo use machine-learning models trained on millions of faces and scenes. They do not recover the hidden detail. They paint new detail that is statistically plausible for an image of that kind. The output can look sharp and convincing while showing a face that never existed, which is why an enhanced image should never be treated as evidence of what the original showed.
Text is a different case, and it is the one that matters for privacy. Screenshots use a small set of characters drawn in a known font at a known size. An attacker can pixelate each candidate letter in the same way, compare the result with the blocks in your screenshot and keep the closest match. Security researchers at Bishop Fox built a tool that does this, and the open-source Depix project recovers text from pixelated screenshots under similar conditions.
Academic work points the same way. In a study titled Defeating Image Obfuscation with Deep Learning, McPherson, Shokri and Shmatikov trained neural networks to identify faces and to recognise objects and handwritten digits in images that had been pixelated or blurred. The lesson is not that every pixelated photo can be read, but that pixelation is a visual effect, not a security control.
Which censor effect to use for what
Match the effect to what would happen if the hidden content were recovered. For decoration or background clutter, any effect is fine. For anything that identifies a person or an account, choose an effect that keeps none of the original pixels.
| What you are hiding | Mosaic or soft blur | Solid block or random scramble |
|---|---|---|
| Account numbers, emails, addresses, passwords in a screenshot | Unsafe: text can be guessed | Use this |
| Names in a chat or document screenshot | Unsafe: short words are easy to guess | Use this |
| Licence plates | Risky with small blocks | Use this |
| A face in a public photo you are sharing | Acceptable for casual viewers if the blocks are large | Safest choice |
| Messy background or a brand logo | Fine | Not needed |
| A retro low-resolution look | Use this | Not suitable |
Step by step: censor a screenshot safely with the free tool
The Pixelate Image tool on ToolkitPoint works in your browser. You mark the areas yourself; it does not detect faces or text automatically, so take a moment to cover everything sensitive.
- Open the tool and pick your screenshot or photo with "Choose an image". A preview appears below the settings.
- Drag on the preview to draw a rectangle over each piece of text, or an ellipse over a face. Draw slightly larger than the content so no edge of a letter or chin stays visible.
- In "Censor effect", choose "Solid block" for text and account details, or "Random scramble" if you want the area to look less stark. Both keep none of the original pixels inside the region.
- For faces in photos where you only need to stop casual recognition, "Mosaic" with a large block size is also an option. Read the warning the tool shows if your setting is light.
- Press "Show at actual size" and zoom in on each region. If a region missed something, use "Undo last region" and draw it again.
- Choose PNG as the output format, press "Create censored image(s)" and download the file. PNG keeps the solid areas clean, while JPEG adds compression noise around block edges.
Settings that leave less to guess
- Block size relative to the content: a mosaic block that is smaller than a letter leaves the letter's rough shape behind. If you must use mosaic on text, make each block larger than a full character.
- Cover the whole field, not just the value: hide the label as well when the label plus the length of the hidden text would give the answer away.
- Prefer a solid block for anything typed: it is the only effect where there is nothing left to compare against.
- Check the export, not the preview: open the downloaded file at full size before you post it.
- Mind the metadata: the tool redraws exported images on a canvas, so camera details and GPS position from the original are not copied into the result.
- Crop when you can: if the sensitive part is near an edge, cropping it out removes it completely and leaves no shape to analyse.
If you need the original detail back
If someone sent you a pixelated image and you need the original, the only reliable route is to ask them for it. An enhancement app will give you a new picture that looks sharper, not the real one, and it may confidently show the wrong face or the wrong number.
For your own photos, the practical fix is a habit: keep the untouched original in a private folder and only censor a copy for sharing. Censoring is then a one-way step you can repeat whenever you need a fresh version, with different regions or a stronger effect.
Common mistakes when hiding information in images
- Using a light blur on text because it looks neat. Blur keeps the outline of every character, which makes guessing easier, not harder.
- Censoring one screenshot but sharing an uncensored version elsewhere, for example in an earlier message, a cloud album or a document that contains the same image.
- Missing reflections and repeats: the same name may also appear in a window title, a tab, a notification or a reflection in glasses.
- Redacting a PDF by drawing a black box over it in an image editor and then exporting the PDF with the original text layer still inside. For PDFs, use a redaction tool that removes the text itself.
- Assuming automatic detection: this tool, like many simple editors, only changes the areas you mark.