No sign-up · no watermark · unlimited · runs on your device
or drop it here
Checking what this device can do…
More resolution and cleaner edges.
Turns blocks and stair-steps back into gradients.
Takes out the grain, leaves the edges.
Tightens soft edges, keeps your dimensions.
2× or 4×, photos and illustrations.
2×, 3× or 4× on your own GPU, frame by frame.
Crisper edges, without the tell-tale halo.
Tightens soft edges, frame by frame.
Cleans up compressed, soft footage.
Find the file you have. The second one is the case that changes which method you should use.
BeforeAfter
BeforeAfter
BeforeAfterPNG, JPEG or WebP, one file or a dozen. The page checks what your device can allocate before you choose a scale, so you find out about limits up front rather than half way through a 4× job.
Hard letter edges mean the default is right. Edges that look photographed through glass mean Restore & Enlarge (2×) is right — that choice is worth up to 2.5 dB. Then work out the pixels your print size needs at 300 DPI and pick the smallest scale that clears it.
The file comes back as a PNG with a pHYs header declaring 300 DPI, so InDesign, Illustrator, Word, Preview and print RIPs place it at the right physical size. The settings panel also offers 72, 150, 350 and 600. No watermark, no sign-up, no export cap.
Because DPI is not a property of the picture. It is one number in the file header, and rewriting it is a nine-byte edit.
DPI says how many pixels to put in each printed inch — nine bytes in the file, changeable without ever decoding the picture. Which is what the tools ranking for this search do, and the one ranked first says so in its own words: “the actual quality (resolution) of the image is not affected”.
Metadata only · 1200 × 1904 · the pixels you started with
This page, 2× · 2400 × 3808 · 34.4 dB from the original
A real 2400 px scan · the master this test was built from
The 300 DPI written here is real — macOS and ImageMagick both read it back out of the file. The check that costs nothing is the other one: note the pixel dimensions before and after. If they did not change, neither did anything else.
Is your source crisp or soft? Measured over the same fourteen documents twice, the ranking inverts — so there is no single best method here.
| Method | Pick it when◆ | Crisp scans | Soft scans | Cost |
|---|---|---|---|---|
| Balanced (default) | You are not sure, or the source already looks sharp | Best free option | Second-best free option | No download · 65 ms |
| Restore & Enlarge (2×) | The scan looks soft before you run it | Ninth — do not use it here | First on 7 of 10, and ahead of the default on all 10 | 924 KB · 130 ms |
| Unpixelate (Trained) | Screen captures, or anything heavily compressed | Top two, wins 5 of 10 | Top two, beats the default 10–0 | 4.9 MB · 851 ms |
| Photo (Grainy Source) | You need four times the size in one step | Middle of the 4× field | Best of the 4× field | 5.9 MB · 514 ms |
Balanced, on a soft scan · no download · 30.6 dB
Restore & Enlarge (2×) · 924 KB · 33.0 dB
The original scan · 1920 px master
Does sharpening ruin text? It is the standard warning, and it is why this page used to default to a method that did not sharpen at all. Measured over the same fourteen sources, the sharpened default was closer to the original on 19 of the 20 scored images. Ringing means overshooting the original’s own edge energy, and on crisp paper the default stayed at 75% to 97% of it; on soft scans nothing came close. The worry there is not sharpening too much, it is that nothing sharpens enough.
Multiply the print size in inches by the DPI. Check your source against this before choosing a scale — it decides whether you need 2× or 4×.
| Print size | At 300 DPI◆ | At 150 DPI | Typical use |
|---|---|---|---|
| Business card · 3.5 × 2 in | 1050 × 600 | 525 × 300 | Held at reading distance |
| 4 × 6 in photo | 1200 × 1800 | 600 × 900 | Photo prints, postcards |
| 5 × 7 in | 1500 × 2100 | 750 × 1050 | Framed prints, invitations |
| 8 × 10 in | 2400 × 3000 | 1200 × 1500 | Portraits, certificates |
| A4 · 8.3 × 11.7 in | 2480 × 3508 | 1240 × 1754 | Documents, flyers, CVs |
| A3 · 11.7 × 16.5 in | 3508 × 4961 | 1754 × 2481 | Posters, plans |
| A1 poster · 23.4 × 33.1 in | 7016 × 9933 | 3508 × 4967 | Viewed from a metre away |
Two practical readings of that table. First, the 150 DPI column is not a compromise for anything large: a banner or an exhibition panel is looked at from a distance where 300 dots per inch are invisible, and insisting on them makes an enormous file for no visible gain. Second, if your source already sits in the 300 DPI column, nothing needs rebuilding — set the DPI at 1×, keep the pixels, and you are done.
If it does not, work out the shortfall before picking a scale. A 1200 × 1800 photo needs 2× to clear A4 comfortably. A 620 × 880 phone screenshot needs 4× — and 4× of an already large scan can exhaust the graphics driver mid-job, in which case the same job usually succeeds on a second attempt.
Two things have to happen, in this order: the image needs more pixels, and then the file needs to declare 300 DPI. Drop a file above and both happen on your own graphics card — the picture is rebuilt at 2×, 3× or 4× its size, and the output PNG carries a real pHYs header saying 300 dots per inch.
Doing only the second half is what a DPI converter normally means. It takes a fraction of a second, costs nothing to run, and returns a file that prints exactly as it did before. Nothing about it is untrue; it just does not solve the problem people arrive with.
4× gives you the pixel count, not the stroke quality — no method measured came close to the original's own edge definition at that factor.
Setting the number does nothing. Adding pixels does something, and how much depends on what you start from.
The biggest win measured, and the one nobody expects. On ten blurred documents the Restore & Enlarge (2×) method beat plain rebuilding every single time, by 0.3 to 2.5 dB — larger than the gap between any two methods on crisp material. It costs a 924 KB download and about twice the time.
Synthetic type on flat backgrounds rebuilds beautifully, and the trained option pulled 3.5 dB ahead of the default on both crisp desktop captures tested. If you are putting a screenshot into a printed document, this is the case the tool handles best.
If your pixel count already clears the table above, all you need is the header. Rebuilding a file that does not need it costs a little sharpness for nothing, so pick 1× and just set the DPI.
It is safe and does not hurt, but changing method barely matters: across a typed 1945 report the top six methods sat inside 0.85 dB of one another. Whatever the paper holds, you get — a careful rebuild will not spoil it and will not invent more of it.
You will get the pixel count a print shop asks for. You will not get the strokes: measured against a real high-resolution scan, no 4× method held more than 81% of the original's edge energy on crisp sources or more than 55% on soft ones, dropping to 24% on typewritten paper. Usable, not crisp — prefer 2× whenever the arithmetic allows.
The lowest scores across everything tested: a 1730 engraved map came back at 16.4 dB against 36.8 for a typed letter. At half resolution, a hundred engraved lines per inch have already merged into one another, and rebuilding merges them more confidently. No method here sees through that.
Drop the file in above, pick a scale, then download. The result is a PNG with more pixels and a pHYs header declaring 300 DPI, so design software and printers read it as a 300 DPI image and place it at the right physical size. There is no account, no credit counter and no watermark, and the file is never uploaded — the resampling runs on your own graphics card.
On its own, no — and this is the thing almost every DPI converter quietly avoids saying, with one honourable exception that says it outright. DPI is a number stored in the file header that tells a printer how large to print each pixel. Changing it from 72 to 300 without changing the pixels does not add any detail; it instructs the printer to pack the same pixels into a smaller area, so the print comes out smaller rather than better. What genuinely helps is having more pixels. This page rebuilds the image first and then writes the DPI.
300 DPI is the default here and the standard for commercial printing — it is what print shops usually ask for. The other four values sit next to it in the settings panel. 150 DPI is fine for large items viewed from a distance — posters, banners, exhibition panels. 72 DPI is the web default and means nothing on screen, since browsers work in pixels and ignore the header entirely. 350 DPI is what Japanese print shops ask for on colour originals where the English-speaking trade asks for 300. 600 DPI is for fine art reproduction and line art, and only pays off if the source has the detail to support it.
Multiply the print size in inches by the DPI. An A4 page is roughly 8.3 × 11.7 inches, so at 300 DPI you need about 2480 × 3508 pixels. A 4 × 6 inch photo needs 1200 × 1800. The table further up the page covers the common sizes. If a print shop still says the resolution is too low after you have run the file, they are dividing pixels by print size rather than reading the header — either raise the scale or reduce the print size, which is free and always works.
Completely, and this is the single most useful thing on this page. Every method was measured twice over the same fourteen documents: once on crisp scans and once on the same scans softened by a lens-blur pass. The rankings invert. On crisp sources the default is the best of the free options and the restoration method is ninth; on soft ones that same restoration method beats the default on all ten, by up to 2.5 dB. It is listed as Restore & Enlarge (2×). If your scan looks soft before you run it, switch to it — that is a bigger difference than any other choice here.
2× if you can get away with it. Measured against a real high-resolution scan of the same page, the default holds 75% to 97% of the original's edge energy at 2× on paper. At 4× nothing does: edge energy landed between 59% and 81% on crisp sources and between 29% and 55% on soft ones, dropping as low as 24% on typewritten paper. 4× gets you the pixel count a print shop asks for; it does not get you the strokes.
Because what this page delivers is a file with a real pHYs header in it, and those nine bytes only exist in a PNG. PNG is also lossless, so a freshly rebuilt image is not immediately degraded again by JPEG compression. Every still comes back as a PNG whatever went in, and the download is named .png so nothing downstream is misled about what it is holding. If your print shop needs a JPEG or TIFF, convert afterwards — the pixels are already there by then.
No. The resampling runs on your GPU and the header is rewritten in your browser. Load the page, disconnect from the internet, and it still works. To be fair to the alternatives: some metadata-only DPI tools are also purely local. The difference between them and this page is not where the work happens, it is whether any work happens to the pixels at all.
No, and the reason is worth knowing rather than being a gap. The DPI is written into the file on your own machine, in the step right after the image is encoded — a model running on a server hands back its own bytes, and those do not carry it. A file that came back from the cloud would print at whatever size its header happened to claim, which is the exact problem this page exists to fix. If you want cloud models for the picture itself, the photo-enhancing page has them.
Partly, and more than expected before it was measured. On softened scans the Restore & Enlarge (2×) method gained between 0.3 and 2.5 dB over plain rebuilding on every test image. What it cannot do is undo a focus miss or motion blur, because that information was never recorded. If softness rather than size is the whole problem, the blurry picture page is the better starting point.