Create Stencil from Image

An island is an enclosed shape that would physically fall out of a cut stencil, think of the inner circle of the letter O. A bridge is a small strip of material left in place to hold that island. When you cut a stencil from this tool’s output, scan it for closed shapes first: letters like O, A, R, and D always need bridges, and fine details may need thickening so the material does not tear. Edge-detect mode tends to produce fewer islands than threshold mode.

A silhouette is a single solid shape – the outer outline filled with one color. A stencil preserves internal detail as separate cutout regions: a portrait stencil has separate areas for eyes, nose, hair, etc. That’s why stencils work for spray-painting (paint passes through the cutouts to make the image).

Threshold mode with PNG output. The output is strictly 2-color (ink + paper); the cutter interprets ink as cut, paper as keep (or vice versa). For physical cutting, remember to add bridges manually in your vector editor – this tool doesn’t add them (single-pixel-thin connections will fall out during cutting).

Posterize with 3-5 tones. Each tone becomes a separate stencil layer (cut each level as a separate cut). For 3 tones: layer 1 is the darkest fill, layer 2 mid-gray fills only where mid-gray AND darker, layer 3 light fill only where light AND darker. You’ll likely want to separate them manually in Photoshop / GIMP after.

Sobel operator: for each pixel, compute horizontal and vertical gradient (per [[-1,0,1],[-2,0,2],[-1,0,1]] and its transpose), then magnitude = √(gx² + gy²). Edges with magnitude above threshold become ink; smooth regions become paper. Edge-thin stencils look like line drawings.

Compute the local mean luminance in a block-size × block-size window around each pixel (using an integral-image trick for O(1) per-pixel cost). If the pixel is darker than (local mean − offset), it’s ink. This handles uneven lighting much better than global threshold – a face that’s bright on one side and dark on the other doesn’t lose detail.

Positive offset darkens the threshold: pixels need to be even darker than the local mean to qualify as ink. This reduces noise in areas where the mean is close to the pixel value (e.g., smooth gradients). Typical values 5-15.

How many pixels became the ink color (vs paper). Very low (<0.5%) means threshold is too low and barely anything becomes ink. Very high (>99.5%) means threshold too high and almost everything is ink. The toast warns in both cases.

JPEG uses DCT compression which introduces “ringing” artifacts at sharp 1-bit edges. PNG is lossless and preserves the binary pixels exactly. For stencils, always PNG. WebP-lossless is also fine.

No. Canvas API runs in the browser. Your image never leaves your device.