Unblur an image

Upload the picture and press Upscale the image. The size is fixed at 4x, so there is nothing to choose: a super-resolution model rebuilds edges and texture as it enlarges the picture, which sharpens softness from small size and compression. Make it with any DreamBench plan and download it full size.

No. When the camera or the subject moves, each point of the picture is smeared into a streak, and the sharp version was never recorded. A super-resolution model sharpens edges that are there; it doesn't reverse the smear, so a shaken photo comes back larger with the same streaks. Retake it if you can.

A little out of focus, a little: edges come back crisper at the larger size. Badly out of focus, no: the detail the lens didn't resolve isn't in the file, and a model that doesn't redraw can't put it back. If it is still blurry after unblurring, the blur was recorded when it was taken; a tool that seems to fix it has drawn a new picture, not sharpened yours.

There is a way to unblur a photo if the softness comes from a small size or heavy compression: upload the original file and run it through a super-resolution upscaler, which sharpens edges as it adds pixels. Mild focus blur improves a little; strong motion blur stays, only larger. Keep the original either way.

You don't need one: this page works in a phone browser, with nothing to install, and takes a photo straight from your camera roll. The iPhone's Photos app has a Sharpness slider, which adds edge contrast but no pixels; this page enlarges the picture four times as it sharpens.

Upload the original screenshot or photo, as a PNG if you have one, and run it here. Lettering that is a little soft at a readable size comes back sharper. Text only a few pixels tall can come back clean but wrong, because the model doesn't read words, so check every letter.

No. Blurring or pixelating a face, a license plate or a line of text on purpose removes the detail, and no model can bring back what was removed. A tool that seems to do it is inventing a face or words, not revealing them. This page only sharpens what is still in the picture.

Sharpening raises the contrast along edges a picture already has, so it looks crisper without gaining detail, and too much gives halos. Unblurring with a super-resolution model rebuilds edges and texture while enlarging the picture, which recovers softness from a small size or compression. Neither reverses motion blur.

Partly. Messaging apps shrink and compress every photo, and that softness sharpens well here, at four times the size. Detail the app threw away can't be rebuilt, so ask the sender for the original by email, a shared drive or AirDrop if you can; it will always beat an unblurred copy.

They run the same model. This page and the 4K upscaler both take a picture 4x, the strongest pass, for small, soft or pixelated pictures; the photo enhancer takes it 2x, for photos already a good size. To keep the original size, shrink the sharper result back in any editor: it keeps the rebuilt edges.

Photoshop's Smart Sharpen adds edge contrast, and Lightroom's Super Resolution doubles the width and height; the iPhone's Photos app has a Sharpness slider, which adds edge contrast but no pixels, and Canva has an upscaler in its editor. This page enlarges the picture four times as it sharpens, with nothing redrawn, in the browser on a phone, tablet or computer, Android or iPhone, Mac or Windows.

No. It sharpens one still image at a time, and DreamBench has no video sharpener. A frame saved from a video can be unblurred like any picture, though a frame blurred by motion stays blurred.

Halos or blocky edges in the sharper image come from the original: heavy compression and noise are sharpened along with the picture, so a copy saved from a chat brings its blocks with it, four times larger. Start from the original file, or shrink an enlarged, blocky copy back to its small size first, so the edges come back smooth instead of the blocks getting sharper.

Unblurring changes none of them. The model adds pixels and rebuilds edges and texture; color, brightness and framing stay as they were, and face retouching is switched off. A PNG comes back as a PNG, a transparent background kept, and any other format as a JPEG, four times as wide and as tall, up to 4096 pixels on the long edge.

No. Every run needs a DreamBench plan; you create an account on the way, and the button shows the price in credits before you press it. The tools come with every plan, alongside the video models, and a run that fails is refunded.

No. The page takes one picture per run, and there is no batch upload. Run it again for the next one: each run is priced separately, and the button shows the price before you press it. Start with the sharpest original of each.

No. A DreamBench tool adds no watermark to an image it returns, on any plan, and the super-resolution model that runs this page embeds no mark of its own. The download is your picture, sharper and larger, in the format you uploaded, with nothing written or hidden in it.

Each run uses 7 credits from your DreamBench plan. Starter, $9.99 a month, includes 1,100 credits a month, about 157 runs of the Unblur image tool, plus AI video. A run that fails is refunded.

Yes: a free account and a DreamBench plan. The tools are part of every plan, from Starter at $9.99 a month.

A photo up to 40 megabytes: JPEG, PNG, WebP, HEIC or another format your browser opens. A photo straight from a phone is fine, an iPhone's HEIC included: the page scales a large picture down to 2048 pixels on its long edge before the run, so the result is never larger than the page says.

Yes. Under the DreamBench Terms you may use what a tool makes for personal and commercial purposes, including advertising, monetized content and work for clients, with no separate commercial-use upgrade. The Terms don't guarantee copyright in a result or clear another person's rights in your upload, so only use photos you have the rights to.

No. DreamBench doesn't use your prompts, uploads or videos to train or fine-tune AI models, and any future training program would be opt-in.

It stays in your browser until you make a result. Then it is used to make that result and nothing else, and removed from the tool's storage after seven days along with the result. A result you keep in your library stays as long as you keep it.