Photo clearer, 2×, 4K

Upscale and enhance an image

Make a small picture 2× or 4× bigger and sharper — or clearer at the same size.

1What to do

What to do Improve quality Same size, cleaner Enlarge 2×, 4× or a ready size

2×: the most natural for photos.

2Photo

Choose the photo

or drop it here, or paste with Ctrl V

JPEG, PNG, WebP or iPhone HEIC · one at a time, as often as you like

Nothing is uploaded

Your photos stay on this device.

The AI runs in your own browser: the model is downloaded once and your photo is never uploaded. That is also why there is no limit: it costs us nothing per photo.

Free
No sign-up
No watermark
No daily limit

A picture that is too small — an old phone photo, a logo someone sent you at 200 pixels, a screenshot that went through a chat app — looks soft and blocky as soon as you make it bigger. This page enlarges it two or four times with a small AI model that redraws the edges instead of just stretching them, so the result is sharper and cleaner than any ordinary enlargement. Or it keeps the size and just improves the quality, taking out the noise and the blocks that messengers leave behind. On a phone, the result goes straight to WhatsApp, Telegram or your mail with one tap.

It is free and unlimited: no sign-up, no watermark, no credits and no daily cap. That is not generosity. The model runs in your own browser, on your own device, so your picture is never uploaded and processing it costs us nothing.

How it works

The model is Real-ESRGAN, an open-source upscaler published by researchers at Tencent ARC in 2021 and still one of the most widely used — in its compact “general” version, which was trained on ordinary photos with the kind of damage real pictures have: blur, noise, and the blocks left by JPEG compression. It looks at every pixel’s neighbourhood and predicts what the missing finer pixels most likely are. Where a plain enlargement turns one pixel into a smooth smear of four or sixteen, the model draws an edge where it sees one, and removes the noise and the compression blocks on the way.

The first time you choose a picture, the page downloads the model — 2.4 MB, about the size of one phone photo — and your browser keeps it, so the next time it starts at once. The picture is processed in tiles of about two hundred pixels, which is why a large one shows steady progress instead of freezing the page. Where your browser has WebGPU (current Chrome and Edge, Safari on iOS and macOS 26, recent Android phones) the graphics card does the work: on an ordinary laptop with Intel graphics, a 1-megapixel picture takes about eight seconds. Where it does not, your processor does it instead, which is several times slower; the page tells you when that is the case.

What the result looks like

These are real results from this page’s model, cropped so that you can see the pixels. On the left is an ordinary enlargement, the way any picture viewer would show it; on the right is what the tool gives you.

A 2009 phone photo enlarged twice the ordinary way: soft, with colour noise — A 640×480 photo from a 2009 phone camera, enlarged to 1280×960. The noise and the JPEG blocks are gone and the thread has crisp edges; the skin is smoother than in life.

The same photo enlarged twice by the AI model: cleaner edges, noise gone — A 640×480 photo from a 2009 phone camera, enlarged to 1280×960. The noise and the JPEG blocks are gone and the thread has crisp edges; the skin is smoother than in life.

A small, compressed screenshot enlarged four times the ordinary way: blurry text — A 160×80 screenshot saved at low JPEG quality. The logo and the large text come out sharp. Look at the two smallest lines, though: “small tools” became “smoll tools” and “Ref.” became “Rof.” — the model guessed letters it could not see.

The same screenshot enlarged by the AI model: the large text is sharp, the smallest text is distorted — A 160×80 screenshot saved at low JPEG quality. The logo and the large text come out sharp. Look at the two smallest lines, though: “small tools” became “smoll tools” and “Ref.” became “Rof.” — the model guessed letters it could not see.

That second example is the most important thing to know about any AI upscaler, this one included. It sharpens what is there; it cannot recover what is not. Text you could already read comes out crisper. Text too small to read comes out as confident-looking letters that may be wrong. Never use an upscaled picture to read a number, a name or a date that you could not read before — not on a receipt, not on a document, not on a number plate.

A woven bracelet enlarged four times the ordinary way: blurry threads — A woven bracelet, 400 pixels wide, enlarged to 1600. The threads come back; up close, some of them are the model’s invention rather than the camera’s.

The same bracelet enlarged four times by the AI model: individual threads visible — A woven bracelet, 400 pixels wide, enlarged to 1600. The threads come back; up close, some of them are the model’s invention rather than the camera’s.

2× or 4×?

2× for photos. Doubling the size asks the model to invent much less, and the result looks natural: a 1000×750 photo becomes 2000×1500, enough for a large screen or a 15×20 cm print. 4× for drawings, logos, icons, screenshots and very small pictures, where edges are clean lines and there is little fine texture to get wrong — a 300-pixel logo becomes 1200 pixels, large enough for a slide or a printed flyer.

You can try both: after a result, the page offers the other factor in one click, and the comparison under the result lets you check either one at actual pixels before you download.

Improve quality without enlarging

Sometimes the size is fine and the photo is just bad: grainy, blotchy, with the square blocks a messenger leaves after compressing it twice. Choose Improve quality and the photo keeps its exact size. The model still works at four times the detail, then the result is averaged back down, so what you get is the same pixels, cleaned. It works on photos of any ordinary size, up to about 16 megapixels.

A 2009 phone photo at its own size: colour noise and JPEG blocks — The same 2009 phone photo, shown twice as large so the pixels are visible. The size did not change; the colour noise and the compression blocks did.

The same photo improved at the same size: the noise and blocks are gone — The same 2009 phone photo, shown twice as large so the pixels are visible. The size did not change; the colour noise and the compression blocks did.

Be realistic about it: it does the most for photos from messengers, old phones and screenshots. A sharp photo from a good camera barely changes, and a blurred one does not become sharp.

Ready sizes: 4K, logo, print

Often you do not want “twice as big” but a particular size. Under Enlarge, open Ready sizes; each runs the same AI enlargement and then brings the result to that exact size.

4K — exactly 3840×2160 pixels (2160×3840 for an upright picture), for a TV, a monitor or a wallpaper. 2K is 2560×1440.
Logo — 4× and always saved as PNG, so the edges stay crisp and a transparent background stays transparent, even when the logo you have is a JPEG.
Print 4×6″ (1800×1200, the 10×15 cm photo-lab print) and Print A4 (3508×2480) — at 300 dpi, the resolution print shops ask for. The 300 dpi is written into the file, so a photo lab or a word processor sizes it correctly on paper without asking.

When the shape of your picture differs from the size — a 4:3 photo made 16:9 for 4K — choose Fill the frame to trim the edges that stick out, the way a photo lab does, or Whole picture to keep everything and add plain margins. If your picture is very small for the size you chose (a 480-pixel photo for 4K needs eight times its size, and the model gives four), the last step is an ordinary enlargement and the page tells you so; if it is already larger than the size, it is simply reduced, with no AI at all.

What it cannot do

A badly blurred photo stays blurred. Light softness is fixed; motion blur and a photo out of focus are not.
Faces are not rebuilt. A face that is a few pixels across in a group photo can come out odd. Faces at a reasonable size get sharper, with smoother skin.
It does not restore old prints. Scratches, folds and faded colour stay; this is an enlarger, not a restorer.
Fine textures can look painted at 4× — grass, hair, knitwear. Use 2× for those.
The largest result is 16.7 megapixels (4096×4096), the most a browser on an iPhone can hold. So 4× works on pictures up to about 1 megapixel (1024×1024) and 2× up to about 4 megapixels; the ready sizes all fit under it. A larger photo is already large enough for almost anything, and the page says so.

What happens to your picture

Nothing leaves your device. The model is downloaded to your browser, and the picture is enlarged there; there is no upload, no queue and no copy on a server to be deleted later. Once the model has loaded you can switch off your internet connection and the page keeps working. The result has no watermark and carries no camera or location data: it is saved as a new file from the pixels alone, as a JPEG — or as a PNG if your picture was a PNG, transparency included.

Questions people ask

Is it really free and unlimited?

Yes. Sites that upscale on their own servers pay for every picture, which is why they give you three a day or ask for an account. Here your device does the work, so there is nothing to ration: enlarge as many pictures as you like, one after another.

Is this AI?

Yes — a neural network trained to enlarge pictures, not a chatbot. It does not understand what is in the picture and it does not add objects; it only predicts finer pixels from the pixels around them.

Can I upscale an image to 4K?

Yes: choose Enlarge, then Ready sizes and 4K, and the result is exactly 3840×2160. A Full HD picture (1920×1080) needs only the AI’s 2× to get there; a smaller one gets 4×. 8K (7680×4320) is not offered: at 33 megapixels it is twice what a browser can save.

How do I make a picture 300 dpi for printing?

Dpi alone changes nothing: what matters is enough pixels for the paper size. Choose Print 10×15 or Print A4 and the result has exactly the pixels 300 dpi needs at that size, with 300 dpi written into the file.

Can I enlarge an image without AI?

Yes. After an enlargement, switch the result to Without AI: it becomes an ordinary smooth (bicubic) enlargement, or hard-edged nearest-neighbour for pixel art, and that is what you download. Use it when exact pixels matter more than sharpness.

Does the AI change my photo?

A little, and the comparison shows where. Skin gets smoother, fine textures can look painted at 4×, and tiny text can come out as different letters. It adds no objects and changes no colours.

How do I send the result to WhatsApp or Telegram?

On a phone, press Send under the result and pick the app. Messengers shrink photos again, so to keep the full quality send it as a file (a document in WhatsApp). On a computer, Copy puts it on the clipboard to paste straight into a chat.

Does it work on a phone?

Yes, on iPhones and Android phones alike. It is slower than on a computer — expect tens of seconds for a large picture — and the page shows how long is left.

Why is it slow on my computer?

Your browser has no WebGPU, so the processor does the work instead of the graphics card. Updating the browser, or trying Chrome or Edge, usually brings it back to seconds.

Can I use the result commercially?

The result is your picture, enlarged; this page claims nothing in it. Whether you may use the picture itself depends on who owns it. The model, Real-ESRGAN, is published under the BSD licence, which allows commercial use.

Will it fix a blurry photo?

A slightly soft one, often. A photo that is badly out of focus or smeared by movement, no: the detail was never recorded, and no enlarger can bring back what the camera did not capture.