Why AI Image Detectors Flag Real Art, and What to Do

Plenty of artists have now had the same unpleasant experience: a painting they made stroke by stroke is run through an online AI image detector, and the tool announces that it is probably generated. Sometimes it is a stranger doing the checking. Sometimes it is a contest, a shop, or a client. This guide explains why a detector can get hand-made work wrong, what the best available research says about how often, and how both artists and the people judging them should treat a score.

What an image detector is actually measuring

Most automated detectors are classifiers. They are trained on large collections of images labeled real or generated, and they learn statistical patterns that separate the two groups: tiny regularities in noise, texture, color transitions, and edges that people cannot see. When you upload a picture, the detector compares it to those learned patterns and returns a probability. It never watched you draw. It has no access to your sketches, your layers, or your timelapse. It only knows how closely the finished pixels resemble the images it was trained on.

Why stylized digital art is the hardest case

The features that make a lot of digital illustration look polished are the same features detectors associate with generation. That is an uncomfortable overlap for anyone who paints in a clean, finished style.

• Smooth gradients and soft blending. Airbrushed skin and glowing light, of the kind built with the stack in layer blend modes for shading, produce very even transitions that resemble model output.
• Clean cel shading and flat fills. Large areas of perfectly uniform color leave little of the irregular texture a classifier expects from a camera or a physical medium.
• Heavy compression. Social platforms resize and recompress uploads, which wipes out the fine detail many detectors rely on and pushes real and generated images toward the same blurred statistics.
• Small crops and screenshots. A thumbnail or a screenshot of a screenshot gives the tool almost nothing to work with, and uncertain inputs produce unreliable outputs.

What the research found

The most useful study for artists so far comes from researchers at the University of Chicago, published in 2024 under the title Organic or Diffused: Can We Distinguish Human Art from AI-generated Images? The team collected real human art across seven styles, generated matching images with five generative models, and tested eight kinds of detector: five automated tools and three groups of people, including 180 crowdworkers, more than 4,000 professional artists, and 13 expert artists experienced at spotting generated work.

The best commercial detector and the expert artists both performed very well, but they failed in different ways. The automated tool was weaker when images had been deliberately perturbed, while the experts produced more false positives, calling real art generated. The authors concluded that a combined team of human and automated review gave the best mix of accuracy and robustness. In other words, neither a tool alone nor an expert alone should be the last word.

Standards bodies take a similar view. The US National Institute of Standards and Technology, in its 2024 overview of technical approaches to synthetic content, treats detection as one tool among several, alongside provenance records and watermarking, rather than a standalone answer.

If a detector flags your work

If you run a contest, a shop, or a moderation queue

• Never treat a detector score as a verdict on its own. Use it, at most, as a reason to look closer.
• Publish process evidence rules in advance, so every entrant knows what may be asked for and nobody is singled out.
• Have a trained eye review flagged work, using the construction, edge, and light checks in telling generated images from hand-made work.
• Give the artist a real chance to respond before any public decision, and accept timelapse, layered files, and sketches as evidence.
• Record how each case was decided, so the process can be checked and improved.

The honest limits

No detector and no human judge is perfect. Errors run in both directions: real paintings get flagged, and edited generations slip through. That is not a reason to give up on judging images, but it is a strong reason to change the question. Instead of asking what a tool says about a picture, ask what evidence exists of the process behind it. A working artist almost always has that evidence. A generated image almost never does. Built into your habits, that difference protects you far better than any score.

Sources and further reading

Primary references for the facts and definitions in this piece, all from independent publishers:

• Ha et al., Organic or Diffused (2024): the study of eight detectors, including professional and expert artists.
• NIST AI 100-4, Reducing Risks Posed by Synthetic Content (2024): why detection is one approach among several.
• Content Credentials: Verify: checking the provenance record on your own file.