Why AI Background Removal Fails on Dim Footage and Close Colors

An editor's analysis argues that AI background removers are often sold as a post-production fix, but two common failure modes behave differently. Underexposed footage stores brightness in a compressed dark range, so detail is permanently lost and can only partly be recovered by lifting levels, which introduces banding. Tests on a sample clip showed that denoising before brightening preserves more temporal stability than lifting alone. Color keyers and segmentation tools work from the measured distance between colors, so when a subject's color is close to the background, the tolerance setting cannot separate them without partially erasing the subject. The article recommends checking the share of unrecoverable dark pixels before attempting a fix.