No. 05 — AI Tools · Cloud HDPro · on our GPU

Blurry photos, made clear.

Two photographs can look equally soft on your screen and be nothing alike underneath. One was blurred by the camera; the other was blurred by everything that happened to the file afterwards. Work out which you have, then let our server rebuild the faces.

Pro · Cloud HDDeleted on completionUp to 25 MB · 24 MP
Restore a blurry photoCloud HD · our GPU

How it works

  1. Pick the blurry photo of a person — JPEG, PNG or WebP, up to 25 MB and 24 megapixels.
  2. Our GPU restores the faces and checks each one against your original. If the result drifts too far from the person in your photo, the run is refused rather than handing back an invented face — and you are not charged.
  3. Download the restored PNG at the same resolution you uploaded. Enlarging is a separate, on-device step afterwards.

This page is the photo restorer, tuned to diagnosing where the blur actually came from.

All Image Tools

AI Tools

Solutions by use case

Restore Blurry PhotosPro, on our server

There are two different blurry photos, and only one of them is the camera’s fault. Capture blur happens at the moment of exposure: the lens focused on the wrong plane, the subject moved, the hand shook in low light, the digital zoom cropped into a sensor that had nothing left to give. The picture is soft but the underlying structure of the scene is intact — spread out, not deleted. File blur happens afterwards: the image was resized down and later blown back up, saved at a punishing JPEG quality, screenshotted, or run through a messaging app that re-encoded it twice. That kind of softness is information that has been quantised away and replaced with blocks. Zoom to 100 per cent on a hard edge and you can tell them apart in about five seconds.

The distinction decides what restoration can honestly return. RestoreFormer++ does not sharpen a face; it rebuilds one, using what it has learned human faces look like, and then blends the rebuild back into your photograph. That works on both kinds of blur, but it works better on capture blur, because the model still has a faithful if soft record of the underlying face to anchor to. On a heavily compressed file it is also fighting artefacts that are not features — ringing beside an eyelid, blocking across a cheek. Either way the identity check has the final word: if the rebuilt face no longer reads as the person in your upload, the run is refused rather than returned, and a refused run is not charged. See the main restorer for how that check works, and restoring faces in old photos for what happens when a frame holds several.

Blur is not the same problem as size, and mixing them up wastes a run. Restoration hands back a PNG at exactly the resolution you gave it. If your photo is both soft and small, do the restoration first and the enlargement second: the upscaler runs on your own device and enlarges faithfully, which means it will enlarge blur just as faithfully as anything else. And if the blurry photo has no person in it at all, this tool will decline it — send that one to the AI photo enhancer, which is the same pipeline with the face requirement lifted, or to unblur image for the general case.

What happens to the file, in one paragraph and no euphemisms. This is Cloud HD, the Pro-only, opt-in part of MiniPx that runs on our GPU because the model is too heavy to ship to a browser. Your photo is encrypted in transit, uploaded to our server for that run and deleted when the run finishes, and in all cases within 24 hours. Processing happens in the United States, on Cloudflare R2 storage and a Modal GPU, and nothing you upload is ever used to train anything. The model that runs is named on the page, and it can change — you can always see which one processed your job. Every other MiniPx tool, including the upscaler and the compressor, runs entirely on your own device.

How it works

  1. Find the least-handled copy: Use the file straight off the camera or phone, not a screenshot, a chat forward or a download from a social feed.
  2. Diagnose the blur at 100 per cent: Smooth, directional softness is camera blur and restores well. Blocky, ringing softness is compression, and restores less.
  3. Restore the faces: Upload here, up to 25 MB and 24 megapixels. RestoreFormer++ rebuilds each face and identity-checks it against your original.
  4. Enlarge afterwards if you need to: Restoration returns the same resolution. Run the restored PNG through the on-device upscaler last if you want a bigger file.

Frequently asked questions

How do I tell whether the blur came from the camera or from the file?
Open the photo at 100 per cent and look at a hard edge, like a collar against a background. Camera blur is smooth: the edge fades gradually, and with camera shake it fades in one direction. File blur is not smooth — you see square blocks about eight pixels across, faint ripples beside strong edges, and stair-stepping. The two need different treatment.
Can you fix a photo that went blurry because it was saved small and blown back up?
Partly. Enlarging a small file does not restore the detail that resizing threw away, and re-saving as a low-quality JPEG on top of that quantises away more. Faces still restore, because the model rebuilds them rather than sharpening them, but the background stays as coarse as the file it came from. Always hunt for the original file first.
What if there is no person in my blurry photo?
This tool is built around faces and declines a photo with none, rather than returning it barely changed. For a blurry landscape, building or object, use the AI photo enhancer instead — the same restoration pipeline with the face requirement lifted, which cleans up blur and noise across the whole frame at the same output size.
Does restoring a blurry photo make the image bigger?
No. The result comes back as a PNG at exactly the resolution you uploaded — a 2400 by 1800 input returns 2400 by 1800. Blur and size are separate problems with separate tools. If you need a larger file as well, restore first and enlarge afterwards, because an upscaler faithfully enlarges whatever you give it, blur included.
Will a screenshot of a blurry photo restore as well as the original?
No. A screenshot is a re-photograph of a screen: it is capped at your display resolution, re-encoded, and often scaled. You lose real pixels and gain compression artefacts. Ask whoever sent the photo to send it again as a file attachment or at original quality, and restore that instead.
Where does a blurry photo actually go when I press restore?
It is uploaded to our server for that run and deleted when the run finishes, and in all cases within 24 hours. It travels encrypted, is processed in the United States on Cloudflare R2 storage and a Modal GPU, and is never used to train anything. This is Cloud HD; every other MiniPx tool processes on your own device.