Unblur Image — Sharpen Soft Photos
Recover a photo that missed focus. Instead of a plain sharpen filter, this runs a real deconvolution that estimates what the image looked like before the blur — on your device, in a few seconds.
Drop photos here, paste with Ctrl+V, or browse
JPEG, PNG or WebP up to 60 MB · select several for a batch · processed on this device, nothing uploaded
Fine-tune
AI deblur (optional, your own key)
Gemini redraws the photo rather than filtering it, so it can recover more from heavy blur — and can also change small details. Nothing is billed to us; your key goes straight to Google.
Get a free key at ai.google.dev. Stored only in this browser, sent only to Google. Takes 10–30 s.
How it works
Drop the blurry photo
JPEG, PNG or WebP. Decoded in the tab; nothing is uploaded.
Choose a strength
Light for slight softness, Medium for a typical missed focus, Strong for heavier blur (watch for halos).
Compare and download
Drag the slider, inspect at 100%, then save as PNG or JPG.
How unblurring works here
A blurred photo is the sharp scene convolved with a blur kernel. Richardson–Lucy deconvolution turns that around: it starts from your photo, blurs its current estimate, compares that with the real photo and corrects the estimate — 4, 7 or 10 iterations depending on the strength you pick. The blur is modelled as a small Gaussian, which fits typical missed focus and slight camera shake. A thresholded unsharp mask finishes the job so edges pop without lifting noise in flat areas, and 15% of the original is blended back to keep ringing under control.
This is genuinely different from the "sharpen" slider in a phone editor, which only exaggerates edges that are already there. Deconvolution can recover texture that a plain sharpen cannot, but it cannot invent what was never captured.
What it can and cannot fix
It works well on soft focus, slight front/back focus and light hand-shake. It will not fully rescue heavy motion blur (long streaks), a strongly out-of-focus subject, or a 100 px thumbnail that was never sharp. For those, the AI deblur below redraws the image with your own Gemini key and often produces a cleaner result — at the cost of possibly changing small details.
Frequently asked questions
How does this unblur tool work?
It runs Richardson–Lucy deconvolution: the blur is modelled as a small Gaussian kernel, and the tool iteratively estimates the sharper image that would have produced your photo. A thresholded unsharp mask then sharpens edges without amplifying noise in flat areas. It runs in your browser in a few seconds.
What kind of blur can it fix?
Mild to moderate softness: a missed focus, slight front or back focus, a little hand-shake. Heavy motion streaks or a strongly defocused subject cannot be fully recovered by any filter — try the AI deblur option for those.
Why do I see halos or ringing at Strong?
Deconvolution amplifies high frequencies, and if the real blur was smaller than the model assumes, edges overshoot. Step down to Medium, or use Light on photos that are only slightly soft.
Is there a size limit?
Photos are processed at up to about 2.2 megapixels so the deconvolution finishes in seconds. A larger photo is scaled down to that size first; you can run the result through the upscaler afterwards.
Is it free and private?
Yes. Instant mode is free, has no signup or watermark, and never uploads your photo. The optional AI deblur sends the photo to Google using your own Gemini key, so there is no cap from us either.
What does the AI deblur do differently?
It asks Gemini to redraw the photo sharp rather than filtering the existing pixels. That can recover much more from a heavy blur, but it may alter fine details such as text or facial features, so compare carefully before you use the result.