New AI Tool Detects When AI Art Copies Real Artists

If you're an artist, the rise of AI image generators like Stable Diffusion has been a double-edged sword. These tools can create stunning images in seconds, but they often do so by learning from — and closely imitating — the styles of living artists. Now, a team of researchers has developed a way to measure just how much an AI image copies a human artwork, and it's surprisingly accurate.

The system works like a high-tech fingerprint comparison. First, the researchers used an AI image generator to create copies of original paintings. Then they fed pairs of original and AI-generated images into a "Siamese network" — a type of AI that compares two things and decides how similar they are. The network wasn't just checking for pixel-by-pixel matches; it looked at deeper visual and semantic features, like color palettes, brushstroke patterns, and even the mood of the piece. In tests, the tool correctly identified copied images 99.4% of the time.

The stakes are real. Previous research has shown that AI models can replicate an artist's style up to 81% of the time and achieve 90% visual similarity to the original. That's the kind of closeness that could easily lead to unpaid, unrecognized imitation. This new detection tool could give artists and lawyers a way to prove when an AI image has crossed the line from "inspired by" to "copied from."

The honest limitation? This is a research paper, not a consumer app. The tool works best when it has a specific original artwork to compare against, and it can't tell you whether an AI image illegally copied an artist unless you already suspect a particular source. It's a big step toward fairness in AI art, but there's still a long way to go before artists can use it in everyday disputes.