Pencil Sketch
The classic sketch recipe, done properly: convert to grey, invert, blur, then colour-dodge the grey against the blurred inverse. Where the two nearly cancel the result blows out to white, so flat areas disappear and only the places where brightness changes — the edges — are left as dark strokes. It looks like pencil because pencil is also just edges.
Drop images here
PNG, JPG, WebP — multiple files allowed
The radius is the pencil
Everything about the character of the drawing comes from the blur radius, because the radius decides what counts as an edge. A small radius — 2 to 4 — only cancels detail within a few pixels, so fine texture survives and you get a tight, technical, etching-like result. A large radius, 20 and up, cancels everything except the largest tonal transitions, giving soft charcoal shading with broad strokes and almost no fine detail.
Scale matters here. A radius of 8 on a 600-pixel image is a bold sketch; the same 8 on a 4000-pixel image barely registers, because 8 pixels is nothing at that size. Roughly, aim for about 1% of the image's long edge as a starting point and adjust from there.
Contrast, colour and inversion
Colour-dodge tends to produce a pale result with most values crowded near white. The contrast control pushes values away from the midpoint to fix that — 130–180% is usually where the strokes start to look like graphite rather than a faded photocopy. Past 250 you are down to pure black lines on pure white, which is closer to line art than sketch.
Coloured pencil keeps each pixel's hue and takes only its brightness from the sketch, so a red jacket stays red but is drawn rather than photographed. It works well on images with a few strong colours and turns muddy on busy ones. Invert gives you white strokes on black, which is the chalk-on-blackboard look and reads much better than you would expect for architecture and machinery.
FAQ
My sketch is almost entirely white.
The radius is too small for the image size, so nearly everything is cancelling. Increase the radius, then raise the contrast. Low-contrast source photographs also give thin results — a picture with clear light and shadow sketches far better than a flat, evenly lit one.
What kind of photo works best?
Strong directional light, clear subject, uncluttered background. Portraits, buildings and vehicles do well. Foliage and crowds produce a mess of strokes with no legible subject, because everything is an edge.
How is this different from Edge detect?
Edge detect gives you a mathematical gradient magnitude — bright lines on black, uniform in weight. This produces a tonal drawing with varying stroke darkness and white paper, which is why it reads as hand-drawn rather than machine-processed.
Does it keep transparency?
The alpha channel is preserved, but a transparent background will look odd because the sketch's "paper" is white pixels rather than nothing. Flatten onto white first with Background for a predictable result.