ImgVX AI

Fix bad edges after background removal

AI background removal gets the outline of most subjects right in one click. When it goes wrong, it is almost always at the edges: a pale halo, a jagged line, a missing strand of hair, or a gap where the subject and the background were the same colour. Each has a cause, and most have a quick fix.

Start with a photo that separates well

The model decides, pixel by pixel, what is subject and what is background. It does best when they differ clearly.

Halo: a pale or coloured fringe

Cause: edge pixels are a blend of the subject and the old background. Removing the background leaves that blend behind, which shows as a light rim on a dark new background (or a dark rim on a light one).

Fixes:

Hair and fur

Cause: fine strands are thinner than a pixel and partly transparent. The model has to decide how see-through each edge pixel is, and against a busy or similar-coloured background it can drop strands or keep clumps of background between them.

Fixes: shoot against a plain background, use the Best edges model, and choose a new background close in colour to the old one so partial pixels still look right. For a head-and-shoulders photo, Portrait Blur is often better than full removal: it keeps the real background, just out of focus, and never has to guess at hair.

Missing parts, or background left behind

Cause: the model treated part of the subject as background (a white shirt on a white wall, a glass, a thin handle) or kept a piece of the background that looked like the subject.

Fixes:

Jagged or blocky outlines

Cause: a low-resolution or heavily compressed source, or enlarging the cut-out after removal.

Fixes: remove the background from the largest original you have; if the image is small, enlarge it first with the Image Upscaler, then remove the background. Resize the result down, not up.

Saving without losing the edges