Updated October 2026
Portraits are the single hardest common subject this tool converts, for a reason specific to faces: a recognizable likeness depends on soft tonal gradients across cheekbones, eye sockets, and jawlines, not just a clean outline — and soft gradients are precisely what an edge detector is worst at turning into clean, deliberate lines.
A face's "features" aren't hard-edged shapes the way a logo or bold-lined object is — the transition from forehead to temple, or along the bridge of the nose, is a gradual brightness change rather than a crisp boundary. The Sobel detector needs real contrast to register a line; where that transition is gradual, it either produces a faint, broken line or misses it, and where lighting happens to create harsher shadow lines (under the brow, beside the nose), it can register those as if they were intentional outline — sometimes producing a result that reads as heavier shadow-lines than actual facial structure.
Soft, diffuse, fairly direct lighting (overcast daylight, a softbox, or even a bright window with sheer curtains) minimizes harsh shadow edges that the detector would otherwise pick up as unwanted lines, while still preserving enough real structural contrast to define key features. Harsh single-source side lighting — dramatic in a regular photo — is the worst case for stencil conversion, since it throws strong shadow edges across the face that have nothing to do with the actual likeness and will show up as stray lines in the output.
A portrait photographed at a dramatic angle introduces additional shadow and foreshortening that complicates edge detection further. A more direct, front-on (or clean three-quarter) angle with even lighting gives the cleanest structural contrast for conversion — this is a case where the "best" reference photo for a stencil and the most artistically interesting photo for its own sake aren't necessarily the same image.
Even with ideal lighting and careful settings, a fully automated portrait conversion often needs more manual touch-up afterward than other subjects — redrawing an eye that under-defined, or erasing a stray shadow line that doesn't belong. For a portrait that needs real interpretive judgment about which facial details matter most to the likeness, a hand-drawing approach (see stencil maker vs. Procreate) sometimes produces a more faithful result than pushing this tool's automated detection further.
Many portrait stencils work best as a hybrid: run the photo through this tool for a fast structural first pass (getting the major feature placement roughly right), then refine by hand — adding detail the automated pass missed, removing lines it invented — rather than treating either the automated result or a from-scratch hand-trace as the only option.
Why does my portrait stencil look like random shadow lines instead of a face? Likely harsh or uneven lighting in the source photo creating shadow edges the detector reads as outline — try a more evenly, softly lit reference photo and moderate sensitivity.
Should I convert a color photo to black-and-white first for a portrait? The tool does this internally regardless, but previewing in grayscale first can help you judge whether there's enough real tonal structure in the face before adjusting sliders.
Does this work for pet or animal portraits the same way? Similar principles apply — fur texture behaves a bit like skin texture (fine noise to manage with smoothing), and even, direct lighting still helps define the underlying facial structure cleanly.