Why AI-Generated Composites Look Fake: The Lighting Coherence Problem

Composited images often appear unconvincing not because of poor facial detail, but due to mismatched lighting between subjects. Human vision is highly sensitive to inconsistencies in color temperature, key light direction, shadow contrast, and edge atmosphere — even when viewers cannot consciously identify the cause. A developer writing on DEV Community argues that lighting analysis must happen before compositing begins, not as a post-processing fix, since critical information is lost once images are combined. Simple code-based estimators can measure color warmth, light direction, and contrast ratio from source photos to determine whether two subjects are even compatible for compositing. When the angular difference between key light directions exceeds 90 degrees, the recommended approach is to change the scene or regenerate one subject's shading entirely, rather than attempting to blend incompatible lighting.