Drop in an image, pick a preset (Sobel, Gaussian, sharpen, emboss…) or hand-edit any cell, and watch the filtered output update in real time. Useful for understanding the kernels behind every CV pipeline, Photoshop filter, and CNN layer.
A convolution multiplies each pixel and its neighbors by the matching kernel cell, sums the products, divides by the divisor, and adds the bias. Identity (1 in the center, 0 elsewhere) leaves the image unchanged.
For each output pixel O(x,y) the tool computes (Σ K(i,j) · I(x+i, y+j)) / divisor + bias, then clamps to [0,255]. Each color channel is processed independently. Auto-normalize divides by the kernel sum (or by 1 if the sum is zero, common for edge detectors). Edge mode controls how pixels outside the image are sampled — clamp repeats the nearest edge, wrap tiles, mirror reflects, zero treats them as black. Large images are auto-downscaled to ≤ 600 px on the longest side to keep things interactive — download the PNG to keep the working copy.