Image Upscaler

Plenty of sites promise "AI upscaling" that makes a small photo look like it was shot on professional gear. This tool does something more honest: it enlarges your image using real, decades-old, mathematically well-understood resampling algorithms — the same family used in professional image-editing software long before anyone called anything "AI" — and shows you exactly how that compares to the scaling your browser already does for free.

Two genuinely different algorithms are implemented, not one algorithm with two labels. Lanczos resampling uses a windowed sinc kernel — sinc(x) multiplied by sinc(x/a) within a support radius of a=3 — which acts as a near-ideal low-pass filter and tends to produce the sharpest, highest-contrast edges of the two. Bicubic resampling uses the classic Keys (1981) cubic convolution kernel with a=-0.5 (equivalent to a Catmull-Rom spline), a piecewise cubic polynomial with a support radius of 2 that produces smoother, softer results with less ringing near hard edges. Both are applied as genuine separable convolutions: every row is resampled horizontally first, then every column is resampled vertically, with kernel weights computed per output pixel and normalized so they always sum to 1.

For comparison, the tool also renders the same enlargement using the browser's own default drawImage scaling (imageSmoothingEnabled = true, default quality) — the fast bilinear-family method every canvas-based tool falls back to if it doesn't implement anything better. Placing it next to your chosen Lanczos or bicubic result makes the value of real resampling visible: sharper edges and less blur at the same target resolution, from the same source pixels.

What this tool will not do is invent detail that was never captured. There is no neural network guessing what a blurry patch of hair or a distant license plate "should" look like — every output pixel is a mathematically defined weighted combination of real source pixels. Pick 2x, 3x or 4x, choose Lanczos or bicubic, and download the result as a PNG at full resolution. Everything runs locally with the browser's own Canvas 2D and ImageData APIs — your image is never uploaded anywhere.