For a long time, AI produced images that were easy to expose. That is no longer true — especially in painting. A generator can now create a canvas "in the style of" Monet, Van Gogh or Rembrandt that fools the eye in a fraction of a second. The question is no longer "can AI paint?" but "can you still recognise the hand of a real master?"
Good news: yes, with a little method. Here is why art is a special case, what still gives away a generated painting, and how to train your eye.
Why some styles are easier to imitate
Not all movements are equal against AI. A "loose" style is easier to reproduce than an "exact" one.
- Impressionism is the easiest to fake. Blurry strokes, vague outlines, juxtaposed colours: the style forgives approximation — exactly what AI does naturally. A fake "Monet" can be very convincing.
- The Renaissance and Baroque resist better. The moment you need correct anatomy, rigorous perspective, and exact hands and faces, AI slips up more often. These styles rely on a logic the machine imitates poorly.
In other words: AI captures the mood of a painting, but still stumbles on its mechanics — the structure of the body, the coherence of light, the logic of objects.
6 tells that give away an AI painting
On a screen you can't feel the paint: you judge with your eyes. Here are the most useful signals.
1. Hands, fingers, feet
As with photos, the extremities remain the weak point. An extra finger, a "melting" hand, a badly articulated foot: in a supposedly old painting, that's a strong signal.
2. Geometry and perspective
Look for lines that don't meet, a crooked tiled floor, impossible architecture in the background. The masters commanded perspective; AI only approximates it.
3. Inconsistent light and shadow
In a real painting, every shadow obeys a single light source. AI sometimes places contradictory shadows, or light coming from nowhere.
4. Text and the "signature"
An inscription on a book, a date, a signature: AI often produces a jumble of letters. A genuine public-domain painting has a coherent signature (when there is one).
5. Sloppy secondary details
The main subject is polished, but check the edges: a blurry figure at the back, an object that makes no sense, a strangely repeating pattern. AI "fills" secondary areas with less care.
6. "Random-free" perfection
Paradoxically, an image that is too smooth, too balanced, without a single irregularity, can ring false. The human hand leaves accidents; AI trends toward a slightly too-perfect regularity.
The trap works both ways
Beware overconfidence: you can be just as wrong shouting "AI!" at a real painting. Loose styles like Impressionism resemble what generators produce so closely that a real master's canvas is sometimes taken for an artificial image. The label we're told ("AI-generated" or "authentic") strongly shapes our judgment — which is exactly why testing your eye blind is worth it.
What about the real matter?
An important, honest point: on a screen, a reproduction of a painting and an AI image both arrive "flat". In front of the real work, however, an expert has an asset the machine can't simulate: the matter — the relief of the impasto, the weave of the canvas, the craquelure of the varnish. That's partly why the art market still values the physical object beyond the image alone. In a game, that cue disappears: all you have is visual logic. Good news — it's often enough.
How to prove an image's origin (the real tools)
Beyond the eye, the industry is building standards to trace image origin:
- C2PA / Content Credentials — signed metadata that records an image's history (created by this tool, edited by that one). An open standard backed by major industry players.
- SynthID (Google DeepMind) — an invisible watermark embedded in images generated by certain Google models, designed to survive edits and flag an AI origin.
These tools help, but aren't universal: many images circulate with no metadata or watermark. A trained eye therefore remains a valuable skill.
Train your eye: the Art mode on Fake or Real
Theory isn't enough — detection is learned by practice, with instant feedback. That's the idea behind our new Art mode: each round shows a work, and you decide: authentic or AI painting?
Four movements, from the most "imitable" to the most demanding:
- 🎨 Impressionism — Monet, Renoir, Degas… the blur trap.
- 🖼️ Renaissance — Leonardo, Raphael, Botticelli… the rigour of drawing.
- 🌻 Post-Impressionism — Van Gogh, Cézanne, Gauguin.
- 🕯️ Baroque — Rembrandt, Vermeer, Caravaggio… the chiaroscuro.
Real works come from the public domain (great museums); the fakes are generated "in the style of" each movement. Your job: tell them apart — and earn the Expert Eye and Curator badges.
Ready to play art detective? Real masterpiece or AI painting: could you call it?
Try Art mode →In short
AI superbly imitates the atmosphere of a painting, but still trips on its logic: anatomy, perspective, coherence of light. Loose styles like Impressionism are the easiest to fake; the Renaissance and Baroque resist better. Trust structural clues, beware "random-free" perfection — and train your eye by playing.