Shadows and Lighting Clues in Fake Images

Light leaves geometric clues in photographs. A cast shadow depends on an object, surface and light source; shading and reflections add further information about scene geometry. Digital-forensics research by Farid and colleagues has developed methods for testing whether shadow constraints can fit a proposed light source. An inconsistency may indicate manipulation, but the analysis is more demanding than eyeballing two dark shapes. Human observers can miss subtle violations or see problems where none exist.

Begin with comparable objects and surfaces. Shadows on a level ground plane under the same distant sun may be constrained in a way that shadows on different tilted surfaces are not. Perspective changes how directions appear in the image. Indoor scenes can include windows, lamps, flash and bounced light; a person can cast more than one shadow. Reflections involve viewpoint and material, not merely the direction of a cast shadow. A model-generated picture may contain impossible geometry, but a real composite or unusual lighting setup can produce similar visual oddities.

Check the source file and surrounding frames if available. Identify the objects, contact points, surfaces and candidate lights before drawing lines through shadows. If you cannot establish the geometry, describe the clue as uncertain. A technical analysis may estimate constraints from multiple shadows rather than assume one simple line. Compare with the scene’s time, weather and location when independently known, and seek other evidence such as earlier publication or camera provenance.

Avoid treating one shadow mismatch as proof of AI generation. Older edited photographs and modern generative images can both be inconsistent, while some generated images have plausible lighting. The practical value of shadow analysis is to ask a testable question about physical coherence, then corroborate it. A responsible public conclusion states the assumptions and the alternatives that remain.