A reflection is a geometric constraint
A mirror image is not just a second version of a person placed beside them. Each visible point has to line up with a point of light that bounced off the mirror and reached the observer. The mirror’s angle determines where that point appears, which way it faces, and what parts of the scene it can show. Change the viewer’s position, and the reflection changes too.
That creates a web of relationships: a hand raised on one side must correspond to the right hand in the reflected view; a wall, doorway, or object behind the person must appear in a position consistent with the room. A physically based graphics program can calculate these relationships from a 3D scene—a representation of objects and their locations in space.
The generator draws the whole scene by association
Most image generators do not build that kind of scene and then trace light through it. They create an image by using patterns learned from many examples to estimate which pixels belong together. They may know what a person, a mirror, and a reflection typically look like, but that familiarity does not guarantee that every part of the picture obeys the same geometry.
The person and reflection can therefore be drawn as two locally plausible figures without being matched point by point. The model may preserve the broad idea of “someone looking into a mirror” while getting the reflected pose, clothing, or surroundings wrong. Those mistakes stand out because people are especially practiced at noticing faces and bodily movement—and because a mirror gives us a precise expectation of how the two views should relate.
Why the mismatch is hard to prompt away
Adding more descriptive words can help specify what a reflection should contain, but it does not necessarily give the generator an explicit map of the room or a rule for matching each reflected feature to its source. The underlying challenge is not simply drawing a mirror-shaped object; it is keeping the person, mirror, viewpoint, and reflected scene consistent at once.
That is why a picture can look polished at a glance yet fail the moment you compare the person with their reflection: the generator has captured the visual idea, but not the chain of geometry that makes a real mirror work.
