AI image generation has advanced at an incredible pace. Tools like Midjourney and DALL·E can now produce photorealistic images that look, at first glance, like something straight out of a DSLR. Yet even when the resolution is high and the textures are sharp, many of these images still give us a subtle sense that something isn’t quite right.

Let’s break down why these uncanny details creep in.

Inconsistent Lighting and Shadows

Humans are remarkably good at reading light. We instinctively know how shadows should fall given a light source. AI models, however, don’t understand physics. They’re sampling patterns from training data. This often leads to mismatched shadows; a nose casting one direction while a hand casts another. Even small inconsistencies make an image feel fake because our visual system is hypersensitive to light geometry.

Broken Vanishing Points and Perspective

Perspective rules in real-world photography are rigid, parallel lines converge predictably. AI models trained on billions of photos approximate this but don’t apply strict projective geometry. The result: windows that almost line up, buildings with skewed angles, or furniture that looks fine individually but doesn’t sit correctly in the room. To the eye, it feels like an M.C. Escher optical illusion hiding inside a stock photo.

Unrealistic Depth of Field and Focus

In real cameras, lenses obey optical laws; depth of field changes smoothly, and blur follows mathematical curves. AI generators mimic blur but often overdo it or apply it inconsistently. You’ll see objects at the same depth with wildly different sharpness, or background bokeh that looks like someone applied a Photoshop filter. This creates an uncanny valley feeling.

Microstructures and Symmetry Errors

When zoomed in, generated images sometimes show skin pores, hair strands, or fabric weaves that are almost right but repeat unnaturally. Symmetry is another giveaway; faces where one eye is slightly misaligned, or hands with an extra bend. Humans are highly tuned to spot these near-misses.

Why It Matters

For casual users, these glitches are amusing quirks. But for industries like advertising, film, or even scientific visualization, they highlight the gap between statistical pattern-matching and true physical modelling. Until AI models incorporate explicit constraints for geometry, optics, and physics, “realistic” images will remain plausible illusions rather than perfect substitutes for photographs.