Screenshots and text
Blurry text in an image: when it can be made easier to read
Blurry screenshot text can be easier to read when the letters are still identifiable and only surrounded by JPEG halos or shown too small. If you must guess a character, PhotoRescue cannot establish the original letter. Try a fresh PNG screenshot or a document export before processing the copy.
Tested settings and honest limits
We do not have a fitting real screenshot before-and-after from the production runs for this page. The text fixtures in the model evaluation were rendered test documents and screenshots. They are useful for detecting wrong letters, but would be misleading as a recovered real message, so we do not show them as one.
Our model checks found that the detail stage of the 4× upscaler could turn readable test text into sharp, incorrect words and numbers. Text protection reduced that failure, but still missed some dense small print. A sharper-looking answer is not evidence that a character is correct.
For a screenshot with JPEG halos, Enhance removes compression at 1× and protects the text it detects. For already legible but tiny text, Upscale 4× offers a larger viewing copy; small text stays as soft as in the source. These are suggested workflows, not a production screenshot result demonstrated here. Prefer a fresh PNG capture or an export from the source document, and compare every important character.
No model can establish the original letters when the pixels leave several possibilities. None of these tools reveals redactions, repairs torn paper, or replaces OCR. Keep the unprocessed source for records and evidence.
Read how we test and where our evidence stops. These are saved engine outputs, cropped for the web without retouching.
How it works
- 1
Do the honesty test
Zoom in and try to read the text. If you must guess letters, find a better source.
- 2
Clean, then enlarge
Enhance for halos and blocks, then Upscale 4× if the text is small.
- 3
Read it against the original
Check numbers, names and dates in both versions before you rely on the result.
Three kinds of blurry text
Small text: the letters are clean but only a few pixels tall, so they look jagged or fuzzy when you zoom in. This is common in screenshots of whole pages and in crops from larger images.
Compressed text: the letters are surrounded by halos, speckles and blocks from JPEG compression. This happens when a screenshot or a photo of a page is sent through a chat app or saved as a low-quality JPEG.
Out-of-focus text: the letters are soft because the camera focused somewhere else or moved. Mild softness can be reduced a little. Strong blur cannot be undone.
The honesty test: can you read it at all?
Before you upload anything, zoom into the text and read it slowly, word by word. If you can read it with effort, a cleaner and larger version will be easier to read. If you have to guess letters, the image does not contain enough information to decide between them, and no repaired version will tell you which letter is right.
Some AI tools show sharp, confident text anyway. They draw letter shapes that fit the blur, and the result can look perfectly readable while being wrong. A 3 that becomes an 8, or a name that gains a letter, is easy to miss. That is why PhotoRescue is not a text-recovery tool, and why it should not be used for documents, licence plates, serial numbers or anything evidential.
What to do when the text is still legible
For compressed text, run Enhance first. It removes halos and blocks at the same size, and text it detects is protected rather than redrawn.
For text that is legible but small, Upscale 4× makes the image four times wider and taller, which can make it easier to view. It does not sharpen letters beyond what the source holds: small text stays as soft as it was, because the model does not guess letters. If the text is both compressed and small, clean first and enlarge the cleaned file. Crop to the area with the text first if the image is large.
Neither model is a text-recovery tool. Both were built for photos of people, places and objects, and the text protection can occasionally miss very small or dense print, so read every important word against the original.
Always check the result against the original
Put the original and the result side by side and read the important parts in both. The result should say exactly what you could read, with effort, in the original. Treat numbers, names and dates with extra care. If the result seems to show something you could not read in the original, do not trust that part.
If you need the text as editable characters, run the cleaned image through an OCR tool afterwards. PhotoRescue itself does not convert images to text.
When to find a better source instead
If the text is unreadable, the only real fix is a better source: the original document, a new screenshot taken at a higher zoom level, the original file instead of a chat copy, or a new photo of the page. For a new photo, hold the phone parallel to the page, tap the text to focus and use good light. Text that someone deliberately blurred or pixelated cannot be recovered.
For screenshots in particular, our page on fixing compressed screenshots explains how to capture and share them so text stays sharp.
Free and paid
You get 3 free credits a day with no sign-up or card; free results carry a watermark. Paid plans remove it and raise the upload limit from 3 MB and 8 MP to 20 MB and 30 MP. Uploads and results are deleted within 24 hours.
What it won't do
- It cannot make unreadable text readable and does not reconstruct letters.
- It is not for documents, licence plates, serial numbers or evidential use where every character matters.
- It does not reveal text that was blurred or pixelated on purpose.
- It does not fix strong motion blur or badly missed focus.
- It does not convert images to editable text (OCR).
Questions
Can I make blurry text readable?
Only if you can already read it with effort. Then cleaning compression and enlarging the image makes it easier to read. If you have to guess letters, the information is gone.
Can AI unblur text that is completely unreadable?
Not reliably. When letters are unreadable, the image no longer holds the information to tell them apart. Tools that show crisp text there are drawing it, and it may be wrong.
How do I clean up text in a screenshot?
Use Enhance if the text has halos or blocks, and Upscale 4× if it is clean but small. Then read it against the original.
Can it read a licence plate or a serial number?
No. PhotoRescue is not a forensic tool and does not recover characters. Use the original source.
Is there a real recovered screenshot shown here?
No fitting real production screenshot pair is available. Rendered text tests exposed incorrect letters in the photo detail stage; they are not evidence that a real message can be recovered.