Enhance and enlarge images using advanced Lanczos interpolation.
Improve image quality while scaling up with minimal artifacts.
You can select multiple files at once
Enlarging an image means inventing pixels that were never captured, and how those pixels are guessed makes all the difference. The browser’s default enlargement blends each new pixel from its immediate neighbours, which produces the soft, slightly gummy look everyone recognises as “blown up”.
This tool offers several interpolation algorithms of increasing sophistication, up to an edge-directed method that detects the direction of edges in the source and samples along them rather than across them — which is what keeps a diagonal line looking like a line instead of a staircase. An unsharp mask can be applied afterwards to restore the crispness that any enlargement costs. Scaling runs from 1.5× to 4×.
Every algorithm runs in your browser, including the heavier ones. Nothing is uploaded, so processing time depends on your machine rather than on a queue.
| Algorithm | Method | Speed |
|---|---|---|
| AI-Level / Ultra / Edge-Directed | Edge-directed interpolation with anti-aliasing and denoising | Slowest |
| Lanczos | Windowed sinc filter over a wide pixel neighbourhood | Slow |
| Bicubic | Cubic interpolation over a 4×4 neighbourhood | Fast |
| Bilinear | Linear blend of the four nearest pixels | Fastest |
One point of honesty: the top three options share the same edge-directed implementation, and none of them is a neural network. “AI-Level” describes the intended quality tier, not a generative model — no detail is hallucinated, and nothing that was not in the source is invented. What you get is the best guess classical interpolation can make, which for line art, logos and screenshots is genuinely excellent, and for a blurry photograph is still a blurry photograph, only larger.
| Source | Outcome at 2× |
|---|---|
| Logos and line art | Excellent — edges stay clean |
| Screenshots of interfaces | Good — text remains readable |
| Sharp, well-lit photographs | Good — noticeably better than default scaling |
| Small JPEGs saved from the web | Poor — compression blocks are enlarged too |
| Out-of-focus or motion-blurred shots | Poor — blur cannot be undone |
No. Despite the name of the top quality preset, all the algorithms here are classical interpolation — sophisticated, edge-aware interpolation, but interpolation. No detail is invented that was not in the source image.
The highest quality setting for anything that matters, since the difference is most visible on edges. Drop to Bicubic if a large batch is taking too long; Bilinear only when speed is the sole concern.
2× is reliable for most content. 3× and 4× work well for logos and line art but show their limits on photographs, where they mostly produce a larger version of the same softness.
Enlargement magnifies everything, including JPEG compression blocks, sensor noise and focus errors. A small image saved from a website has all three. Starting from the highest-quality original you can find matters more than any setting here.
It applies an unsharp mask that increases local contrast at edges, restoring some of the crispness enlargement costs. Light suits most images; Strong can introduce visible halos around high-contrast edges.
No. All the processing happens inside your browser, which is why large images at high scale factors take time — the work is being done by your own machine.
