Real vs AI photos: what actually differs

Two origins, one similar look

A camera works with a lens, a sensor and a shutter: it captures the light that happened to be in front of it, at a specific moment, with a specific exposure. An image generator works with patterns: it produces a result that matches what millions of photographs of similar scenes look like.

That is why the two are not opposites in appearance. Both aim at a plausible photograph. The difference becomes visible where “usually” is not enough: at strict rules like text, hands and reflections, in unusual situations, and in the small disorder of a real moment.

What real photos usually carry with them

Genuine photographs contain the traces of how they were made. None of these traces is guaranteed, but together they form what a real picture tends to look like:

• Motion blur, grain and noise, especially in low light
• Accidental framing: something cut off at the edge, a passer-by in the background
• One consistent light direction that all shadows and highlights follow
• Depth of field that follows optics: sharp where focused, soft in front of and behind that plane
• Wear and disorder: rust, dust, wrinkles, tangled cables
• Sometimes technical metadata or content credentials that record camera and edits

What AI images tend to do

Generated images reproduce the average of the photographs they were trained on. That average is tidy:

• Everything fully visible and neatly arranged, nothing accidentally cut off
• Even, flattering light without a clear single source
• Skin, fabric and food that look smoother than the material allows
• Text that looks like writing but forms no readable word in any language
• Reflections that show the wrong thing, or nothing at all
• Repetition: identical windows, identical faces in a crowd
• No metadata and no file history before the upload

Where the comparison breaks down

Real and AI are not two sealed boxes. A genuine photo edited with generative fill has both origins. A product photo is cleaned up until it looks synthetic. A phone snapshot is computed from several exposures, denoised and sharpened, and can end up with edges and textures that resemble generated images.

The direction that surprises people most: at high difficulty, AI images are built to look like unpolished snapshots, with grain, slight underexposure and motion blur. The stylistic evidence people rely on is now reproduced deliberately.

How to compare two images of the same scene

When the same event appears in two pictures, the comparison is often more informative than either image on its own:

• Compare the geometry: are windows, doors, shadows and the position of the people identical?
• Compare the reflections: do both versions show the same thing in glass, water and sunglasses?
• Compare the sharpness: is one version sharper in a way the optics cannot explain?
• Compare the framing: is one version cropped exactly where an object would have been difficult?
• Compare the files: do both carry the same metadata or provenance information?

What to do with the result

Decide on a confidence level instead of true or false: certainly AI, probably AI, unsure, probably real. “Unsure” is a legitimate result and the honest one whenever your only evidence is a style impression.

If you want to practise the difference, the quiz shows you both kinds of images and gives you the answer after every round — that is where you find out whether your intuition about real and AI photos holds up.