Image Emboss
Apply a Laplacian-of-Gaussian-style emboss filter.
Overview
The Image Emboss tool applies a directional relief filter that gives an uploaded picture the look of being pressed into a sheet of metal or stone. Highlights and shadows fall on opposite sides of every edge, leaving a neutral grey across flat regions. Upload a photo and the tool returns a re-encoded copy with the embossed effect baked in.
It is a quick way to create stylised header artwork, embossed logos, faux-metal UI elements or coin-style portraits without launching a full graphics editor. The effect also works well on text screenshots to produce engraved-style social cards.
How it works
Embossing is a convolution against a directional kernel. A canonical "top-left lit" kernel looks like [[-2, -1, 0], [-1, 1, 1], [0, 1, 2]]. The negative weights on one diagonal and the positive weights on the other amplify pixel transitions in that direction; flat regions where neighbouring pixels are similar produce values clustered around the midpoint.
After convolution the result is biased by 128 so that flat regions render as mid-grey. A Laplacian-of-Gaussian variant blurs first to suppress noise and then convolves with a centre-symmetric kernel, producing slightly softer relief that resembles wax or clay rather than crisp metal stamping. The output is desaturated by default so the lighting reads as physical.
Examples
Source: portrait.jpg
Result: face outlined as a low-relief carving — cheek bones, nose
bridge and jawline catch the light.
Source: logo.png
Result: glyph appears stamped into a steel plate, mid-grey background.
Source: leaf.jpg
Result: veins and outline lit from the top-left; the rest neutral.
FAQ
Can I change the light direction?
The kernel sets the lighting angle. Rotating the kernel by 90 degrees swaps highlights and shadows along the perpendicular axis; the bundled preset uses top-left lighting.
Why does my output look flat?
Low-contrast inputs leave little gradient for the kernel to amplify. Try boosting contrast first or pick a higher-contrast source image.
Does it work on colour photos?
The filter is applied per-channel and then desaturated. For a colour emboss skip the desaturation step and run the kernel on each RGB channel independently.
Is it the same as a height-map?
Conceptually similar — the output's brightness approximates a surface normal projected onto the light direction. Game-engine workflows can repurpose the result as a quick fake normal map for low-effort relief.
What's the difference from edge detection?
Edge detection treats both sides of an edge symmetrically. Emboss is directional: one side lights up, the other darkens, producing a 3-D illusion rather than a contour map.