Upscale a PNG
Enlarge an image when you need more pixels than you have — a logo for print, a screenshot for a slide. This resamples in repeated doubling steps with high-quality interpolation and can sharpen afterwards to recover some apparent crispness. Runs in your browser.
Drop images here
Multiple files allowed
What upscaling can and cannot do
Interpolation invents pixels by averaging the ones around them. It cannot recover detail that was never captured — a blurry 200×200 photo becomes a blurry 800×800 photo, larger but no more informative. What it genuinely helps with is output that must meet a pixel requirement: a print shop's minimum dimensions, a template that wants 2000 px, an asset that will be viewed at a distance.
The AI upscalers you may have seen do something different: a trained model hallucinates plausible detail, which looks impressive on faces and text and can quietly invent things that weren't there. That needs a large model and a server. Nothing here uploads your image, so nothing here does that — this is honest resampling, and it will look softer than a marketing comparison screenshot.
Why stepped doubling, and when to sharpen
Scaling straight from 1× to 8× in one draw makes each output pixel sample a thin slice of the original and the result goes flat. Doubling repeatedly — 1×→2×→4×→8× — resamples from a progressively denser source at each stage and preserves noticeably more edge definition. Odd factors double as far as they can, then finish with one fractional pass.
The sharpen pass applies a mild convolution that restores apparent crispness at the cost of slight halos around high-contrast edges. Leave it on for photos and screenshots. Turn it off for artwork with large flat areas, where it can make gradients look ringed.
One important exception: pixel art, sprites and tiny icons should not come here. Smooth interpolation destroys exactly what makes them work. Use Pixel-art upscale, which duplicates pixels with no blending and keeps every edge razor-sharp.
FAQ
Why does 8× look soft?
Because there are 64 output pixels for every input pixel and 63 of them are interpolated. Beyond about 4× the returns fall off sharply — if you need much more, you need a bigger original.
Is there a size limit?
Your browser's canvas limits apply. An 8× upscale of a 4000×3000 photo is 32000×24000 pixels, which exceeds what browsers allow and will fail. Work in stages, or scale a crop.
Is transparency preserved?
Yes. The alpha channel is resampled along with the colours, so soft edges stay soft.
Will the file be much larger?
Yes — 4× the dimensions is 16× the pixels. Run the output through Compress or Resize to file size if that's a problem.
Can I upscale to an exact width instead of a factor?
Yes — Resize takes exact dimensions and will enlarge as well as shrink.