Detail, not stretched pixels
An AI model fills the new pixels with likely detail — sharper outlines and cleaner text — instead of smearing the pixels that are already there.
Make a small picture bigger without it going soft
oppure trascina le immagini in un punto qualsiasi di quest'area, o incollale conCtrl V
Image tools · free, no account
Drop in up to 10 JPG, PNG or WebP images, pick a factor, and every card shows the size it will come out at before anything is sent. The first 10 a day are free with no account, and there is no watermark. The upscaling model runs on our server, the page says so first, and nothing is kept afterwards.
Make small or soft pictures bigger and sharper for print, slides or the web.
An AI model fills the new pixels with likely detail — sharper outlines and cleaner text — instead of smearing the pixels that are already there.
Double or quadruple the resolution. Every card shows the exact output size before you press the button.
Tick Save at 300 DPI and JPG or PNG results print at the size shown on each card.
It cannot recover what was never captured: a face ten pixels wide will not become a portrait. The model draws something plausible, not something true.
Upscale without an account up to 10 times a day. After that, log in and each image costs 2 credits up to 2×, 4 beyond.
The image is upscaled on our server and returned in the same request; it is not saved or attached to an account.
| If the image is for | Pick |
|---|---|
| A sharper version for the web or a slide | 2× |
| An old phone photo, a thumbnail, a small logo | 4× |
| Whichever reaches the print size, with Save at 300 DPI ticked | |
| A picture that is already large | Neither — above about 10.7 megapixels (a typical 12 MP phone photo) there is too little room under the 4096 × 4096 cap, so it is skipped |
This page offers 2× or 4× and says plainly that it cannot recover what the camera never caught. Both come from the model itself; here is where each is written down.
The model enlarges in two fixed steps, 2× and 4×, and the result is then fitted to the exact size we ask for. That is why those are the only two choices, and why the size printed on each card is the size you download.
“When upscale=generate is specified, the AI model runs a single pass at the nearest supported scale (2x or 4x), then adjusts to the exact target dimensions.”The model underneath is ESRGAN, and its authors are open about how this kind of model works: the new detail is drawn to look right, not recovered from the photo, and it can come with odd artefacts. It sharpens a soft picture well; it cannot rebuild a face that was only a few pixels wide.
“However, the hallucinated details are often accompanied with unpleasant artifacts.”Six jobs with a size or resolution floor to clear, and how to get over it. Every rule links to the provider's own help page.
Printful asks for print files of at least 150 DPI, sized to match the actual print dimensions. Its own example: a 5 × 5 inch image at 100 DPI, scaled up to 10 × 10 inches, drops to 50 DPI and prints blurry.
How it helps
Choose 4× and a design that sat at 75 DPI at print size reaches 300 DPI at that size. Tick Save at 300 DPI for printing and each card shows the inches it will print at, before anything is sent.
Udemy requires course images of at least 750 × 422 pixels that are clear and non-pixelated, and says course image violations are one of the most common reasons a course is rejected on first submission to quality review.
How it helps
A 640 × 360 screenshot or photo falls short. Choose 2× and the card shows → 1280 × 720 before anything is sent, with outlines and edges redrawn by the model rather than stretched into blur.
Blurb's BookWright shows a low-resolution warning when a photo placed on the page falls below 250 PPI, and Blurb can't offer replacements for quality issues caused by low-resolution images.
How it helps
Upscale 2× and the same photo can fill a page twice as wide at the same PPI. Tick Save at 300 DPI for printing to see on each card how many inches it covers at 300, before placing it.
Mpix lists 2000 × 2500 pixels as the optimal resolution for an 8×10" print and 800 × 1000 as the minimum. An older digital photo or a small scan often lands somewhere between the two.
How it helps
A 1000 × 1250 photo at 2× comes back at 2000 × 2500, the optimal size on that chart. The model adds plausible detail, so it sharpens a soft photo but cannot bring back a face that was never in focus.
Kickstarter suggests scaling images to 700px or higher before adding them to a project description, and notes that every upload is compressed, so what backers see has a lower resolution than the original file.
How it helps
A 350 px sketch or diagram reaches 700 px at 2×; pick 4× for 1400 px and some headroom against that compression. Up to 10 images go in one batch, and JPG, PNG and WebP keep their format.
Eventbrite recommends event images of at least 2160 × 1080px at 2:1, as a JPEG or PNG no larger than 10 MB. Artwork made for an earlier listing or a social post is often much smaller.
How it helps
A 1080 × 540 banner at 2× comes back at exactly 2160 × 1080, still JPG or still PNG. If the file then passes 10 MB, Compress it from the result screen before uploading.
Drop in up to 10 JPG, PNG or WebP images, pick 2× or 4×, and run it. Each card shows the size it will come out at before anything is sent. The results come back in the same format you put in, in one ZIP.
The first 10 a day are free with no account and no watermark. After that, log in and each image costs credits: 2 for up to 2×, 4 beyond. The price is on the panel before you press the button.
Ordinary resizing stretches the pixels you already have, so edges go soft and text blurs. An AI upscaler predicts the detail that would be there at the larger size — sharper edges, cleaner lines. It cannot recover information that was never in the picture: a face too small to recognise will not become a portrait.
Results are capped at about 4096 × 4096 pixels. Past that, an iPhone browser cannot hold the image in memory, and the other image tools here could not open it. A 4× that would cross the cap becomes the largest factor that fits, and the card shows the real size before you run it. An image that can't be enlarged by at least 1.25× is skipped rather than enlarged by a few percent — that is anything above about 10.7 megapixels, which includes a typical 12 MP phone photo.
It is uploaded — the model runs on our server, and the page says so before you start. The result is sent straight back in the same request; nothing is stored.
Yes. A PNG goes in and a PNG comes out, transparency included. JPG and WebP also keep their own format.
Yes — tick Save at 300 DPI. The upscaled JPG or PNG is marked 300 DPI, and each card shows the size it prints at, so a 2400 × 1800 result reads 8 × 6 in. WebP has no DPI field; convert it to JPG or PNG first if it is going to a printer.