AI Image Upscale

What AI upscaling actually does to a photo

Every digital image is a grid of coloured squares. When you enlarge it, the software must invent the pixels that sit between the existing ones. The cheapest way to do that is nearest neighbour, which copies the closest pixel and produces blocks. Bilinear and bicubic averaging are smoother but still soft. Lanczos3 samples a wider window with a sinc-like weighting function, which keeps edges and fine texture noticeably tighter, and that is why it is the default resampling kernel for professional image pipelines.

Resampling alone is never enough at 4x, because the output has sixteen times the pixels of the source and only a sixteenth of the real information per pixel. That is why the second half of the pipeline matters: a scale-aware unsharp mask. At 2x the sharpening radius stays small, so grain does not turn into noise. At 4x the radius widens slightly to recover the structural lines that resampling softened, without producing the white halos that a naive sharpen filter creates.

How to upscale an image in three steps

1. 1Upload the largest original you haveDrag a photo, screenshot, logo export or scan into the upload box, or browse for it. JPG, PNG and WEBP are supported up to 8 MB. Start from the biggest file you have, because every upscaler can only work with what is inside the file.
2. 2Choose 2x or 4x and a sharpening levelThe output size updates live above the button, so you can see exactly what you will get before spending a second on it. Auto sharpening suits photographs, Strong suits screenshots, logos and line art, and Off is for images you plan to sharpen yourself in an editor.
3. 3Download the resultThe processed file is delivered from our CDN at full size with no watermark. PNG keeps transparency and crisp text, JPG stays much smaller for photographic content.

2x or 4x: which one to pick

Output is capped at 4096 pixels on the longest edge. Above that, file size grows faster than visible quality, and most platforms downscale your upload anyway.

Where an image upscaler earns its keep

Marketplaces are the most common case. Amazon, Flipkart, Etsy and most catalogues want a minimum pixel width, and a supplier photo that arrives at 800 px has to be enlarged before the listing goes live. A 2x upscale with Auto sharpening, followed by a white background clean-up, is usually enough to pass review and look sharp in the grid.

Print is the second case. A 600 px portrait cannot go on an A4 sheet at 300 dots per inch without help, and while no upscaler can create true missing detail, doubling the file and printing at a lower density gives a result that reads fine at normal viewing distance. Flyers, menus, posters and small format packaging all live in this zone.

The third case is the one people forget: screenshots and documents. A screenshot of a dashboard, a scanned receipt, a graph from a PDF, or a logo exported at the wrong size all benefit disproportionately, because they are made of flat colour and hard edges, which is exactly what a good resampling kernel handles best.

Getting the cleanest result

• Work from the original file, never from a screenshot of a screenshot: each re-encode throws away real detail.
• Keep sharpening on Auto for photographs and switch to Strong only for text, logos and UI captures.
• Export PNG when the picture has transparency, tiny type or flat colour, and JPG when it is a photograph.
• If the source is grainy, denoise it slightly before upscaling, otherwise the grain is enlarged along with everything else.
• For a product set, upscale every image at the same setting so the catalogue looks consistent.
• If the picture is damaged, restore it first and upscale second, then colorize if you want colour.

What an upscaler cannot do

It cannot recover detail that was never captured. A face that is completely out of focus, motion blur across a whole frame, or an image compressed down to a few kilobytes has lost its information, and no resampling kernel can bring it back honestly. Tools that claim otherwise are generating a plausible new face, which is a different product and a risk for anything official.

It also will not fix composition, lighting or colour. If a photo is badly exposed, the correct order is exposure and colour first, then upscale. Our enhancer handles the lighting and texture side; this page handles the pixel count.

Privacy, limits and cost

The upscaler is free and requires no account, which is possible because the work is plain image mathematics rather than a paid AI model call. Images are processed in memory and the finished file is written to our CDN so you can download it; we do not build a gallery of your uploads and we do not ask for an email address.

Limits exist for reliability, not monetisation: 8 MB per upload, 30 megapixels of source, 4096 pixels on the longest edge, and a per-network rate limit that keeps the endpoint from being used as a free conversion service. If you need a bigger output for a billboard or a large print, upscale in two passes at 2x rather than one pass at 4x, and keep the intermediate file.

Free image upscaler vs paid upscalers

Paid tools usually add one of three things: a trained super-resolution model that hallucinates plausible texture, batch processing, or an API. The first genuinely helps on faces and foliage at 4x, and it is the direction we are adding next. The second and third matter to studios, not to a seller fixing three photos before a listing goes live.

For the everyday job on this page, a well-tuned resampling and sharpening pipeline is faster, deterministic, and does not alter what is in the photograph. That is the right trade for product photos, screenshots, documents and anything where the original content has to stay exactly as it is.

Upscaling for print: the DPI maths that actually matters

Print resolution is expressed as dots per inch, and the number that matters is pixels along the short edge divided by the print size. A 1200 by 1600 pixel file printed at 6 by 8 inches gives 200 DPI, which is fine for a poster seen from a metre away but soft for a brochure held in the hand. Printing the same file at 4 by 6 inches gives 300 DPI, which is the usual standard for close viewing.

That arithmetic is why doubling beats quadrupling for most print jobs. If your source is 1200 pixels on the short edge and the target is a 6 inch print at 300 DPI, you need 1800 pixels, and a 2x upscale gets you there with room to crop. A 4x upscale produces 4800 pixels, which the printer downsamples anyway while the file size triples.

A 200 DPI print viewed from arm length looks sharper than a 300 DPI print with upscaling artefacts. When in doubt, print slightly smaller rather than upscaling harder.

Batch work and repeatable settings

If you are processing a catalogue, consistency matters more than any single file. Pick one magnification and one sharpening level for the whole set, then apply it to every image, because a grid where half the photos are crisp and half are soft reads as carelessness even when each photo is individually fine.

The order that works for a product set: enhance the lighting on the originals, remove the background if the marketplace requires it, expand or crop to the required aspect ratio, then upscale as the last step before upload. Upscaling earlier means every later operation works on a larger file for no benefit, and re-encoding a large file twice costs quality.

Keep going with the rest of the toolkit