Upscale Photo for Printing — Free, Fast & Private
Printing is the one place where a photo has a physical size, and that changes everything. On a screen an image is just its pixel count; the display decides how big it looks. On paper the size is fixed by the paper, and the only question is how many of your pixels land in each inch of it. That ratio is the print density, quoted in pixels per inch — ppi in the file, dpi on the press, and near enough the same idea for this purpose. Everything that goes wrong with printed photos comes from a file whose pixel count does not support the size somebody chose, and almost every rescue attempt starts by getting the arithmetic straight rather than by opening a tool.
The numbers you need, for the sizes people actually order. At 300 ppi — the standard for anything you will hold — a 4x6 inch print needs 1200x1800 pixels, a 5x7 needs 1500x2100, a 6x8 needs 1800x2400, an 8x10 needs 2400x3000, an 11x14 needs 3300x4200, and a 12x18 needs 3600x5400. In the ISO sizes, A4 is 8.27 by 11.69 inches and therefore about 2480x3508 pixels, A3 is about 3508x4961, and A2 about 4961x7016. Halve the density to 150 ppi for a poster read from across a room and every one of those figures halves too, which is why a 2000-pixel file that fails as an 8x10 photographic print can succeed as a 13-inch poster. Our DPI checker reads what a given file already claims, and does the division for a size you type in.
Why relabelling the DPI field does nothing. Photo metadata carries a density tag, and it is entirely possible to open a 1000x1500 image, change that tag from 72 to 300, and save. Nothing about the picture changed: it still holds 1,500,000 pixels, and the tag has only redefined the physical size the printer will assume — from a 13-inch enlargement to a 5-inch one. The number moved; the resolution did not. That is the difference between a metadata edit and a resolution increase, and it is worth being blunt about, because a great deal of print advice on the web confuses the two. Real resolution increase means generating pixels that were not in the file, and the only honest ways to do that are to rescan or reshoot at higher resolution, or to reconstruct with a super-resolution model — which is what the on-device upscaler on this page does.
What the reconstruction gives a printed image. The free path here runs realesr-general-x4v3 in your browser — a 4.9 MB Real-ESRGAN model, checksum-verified, executed through WebGPU or WebAssembly on your own hardware, in 512px tiles so memory stays flat however large the picture grows. At 4x it turns a 900x1200 web export into 3600x4800, which prints as an 8x10 at 300 ppi with room to spare. Print is also the medium that rewards the model most: a press renders every soft edge exactly as soft as it is, so the difference between an interpolated enlargement and a reconstructed one is more visible on paper than it ever is on a screen. The free tier caps input at 1080px on the long edge and stamps a small corner watermark; Pro removes the watermark, makes on-device runs unlimited, and lifts the cap to 2048px at 4x and 4096px at 2x, both topping out at 8192px of output. Background on the engine itself is on the AI image upscaler page.
Cloud HD, and when a big print justifies it. Pro accounts get a switch beside the run button, off until you turn it on. Switched on, that one photo goes to our GPU and is rebuilt at 4x by HAT-L, a much larger model, with a JPEG pre-clean pass first — useful, because most sources destined for print have been through a messaging app at least once. Intake is 25 MB and 4 megapixels, output reaches up to 16× the pixels of the input — a 4 MP photo comes back at about 64 MP, and a full 4 megapixel source takes four to five minutes, which the tool states before you start. Each run spends one of the 10 HD runs in your month, inside the wider allowance of 150 runs on Pro Monthly or 200 on Pro Annual, and a failed run costs nothing. For a framed enlargement or a gallery print, that is a reasonable trade. For a batch of 6x4s, it is not.
Sequence, and what to do first when the photo is also old or soft. Crop to the paper ratio before upscaling with the cropper: a 3:2 photo on 8x10 paper loses a strip whichever way you turn it, and reconstructing that strip is wasted work. If the source is a faded print or a scan, repair before you enlarge — upscaling an old photo sets out that order, and the Cloud HD AI photo enhancer and enhance an old photo pages cover the repair step, which returns the same size rather than a larger one. For black-and-white originals you intend to colour before framing, colorize photo for printing is the matching page. Upscaling is always the last of those steps, because every earlier one is cheaper on a smaller file.
What happens to the file. The on-device path keeps the photo on your machine: the model is downloaded to your browser and every tile is processed there, so nothing image-shaped is transmitted, and once the model is cached the tool works with no connection at all. The exception, and the only one, is a Cloud HD run you switch on yourself as a Pro subscriber: that photo is uploaded encrypted, processed in the United States on Cloudflare R2 and a Modal GPU, deleted when the run finishes, and in all cases within 24 hours, and never used to train anything. Output is PNG either way, which is what you want going to a printer — lossless, no compression artifacts introduced on top of freshly built texture. Convert or shrink it afterwards only if the print shop asks for a JPEG, using the on-device converter.