To tell if art is AI-generated, start with the artist rather than the image: look for a posting history, work-in-progress shots and a style that develops over time. Then look closely at the piece for rendering errors AI still makes (dissolving details, meaningless ornaments, garbled text and signatures, inconsistent line work) and run it through an AI art detector. Weigh all three. A detector score alone isn’t enough to accuse an artist, because false positives on human art happen and the damage to a real artist’s reputation is hard to undo.
Why art is harder to judge than photos #
With photos, a detector can compare an image against what cameras produce: sensor noise, lens effects, phone processing. Digital paintings, illustrations and anime-style art have none of that. They’re made on a computer either way, so there’s less of a baseline, and a detector has to lean on the traces generators leave in rendering and texture.
People aren’t perfect either. A 2024 study from the University of Chicago tested five automated detectors and three human groups (180 crowdworkers, more than 4,000 professional artists and 13 expert artists experienced at spotting AI) on telling human art from AI art (Ha et al.). Hive’s detector and the expert artists both did very well, but they made different mistakes: Hive was weaker against adversarial perturbations, and the experts produced more false positives. The authors found a human and a detector working together gave the best results.
Step 1: Look at the artist #
This is the strongest evidence you have.
- Posting history. Real artists usually have years of work that shows their skill changing, including older, weaker pieces. An account that appeared recently with dozens of polished pieces in varied styles is a red flag.
- Process. Sketches, line art, color studies, timelapses and layer screenshots are hard to fake at volume.
- Consistency. Human artists repeat their own habits: the way they draw eyes, hands, hair and fabric. AI accounts often drift in style from piece to piece.
- Output rate. Several finished, detailed illustrations a day, every day, is unusual for a person.
- Engagement. Artists talk about their tools, brushes and struggles. Ask a friendly question about how a piece was made.
Step 2: Look at the piece #
Zoom in on the areas that take the most deliberate decisions.
| Where to look | Common AI tell |
|---|---|
| Small details | Jewelry, buttons, buckles and embroidery that dissolve into squiggles |
| Ornament and pattern | Decorative elements that look intricate but follow no logic or repeat |
| Text and signatures | Lettering and signatures that resemble writing but say nothing |
| Hands and objects | Fingers merging with weapons, cups or instruments; strings that don’t connect |
| Line work | Line weight that changes randomly, lines that start and stop without purpose |
| Background | Architecture that doesn’t hold together, props that fade into texture |
| Lighting | A glossy, even glow with no clear light source |
| Composition | Subject dead center, generic pose, background that could be anywhere |
Human art has flaws too, but they’re usually consistent flaws: a proportion the artist always gets slightly off, a hand hidden behind a sleeve because hands are hard. AI errors are random: a perfect face next to a belt buckle that makes no sense. Our guide to AI image artifacts goes deeper on why these errors happen.
Step 3: Check the file and run a detector #
If you have the original file, check its metadata. Some generators and interfaces write their name or even the prompt into the file, and some tools add Content Credentials. How to check image metadata for AI shows how.
Then run an AI art detector. Expose AI checks any image on your phone: share the artwork to it from your browser or a social app, crop away the interface if it’s a screenshot, and it returns Likely real, Uncertain or Likely AI-generated with a confidence meter. Its metadata pass recognizes tags from tools used for art, including Midjourney, Stable Diffusion, NovelAI, Leonardo.Ai, Ideogram and Adobe Firefly, and a credible AI claim in the file decides the verdict on its own. When there’s no such tag, its on-device model judges the pixels. The image isn’t uploaded, and if you’re checking a whole portfolio, Pro’s batch check takes up to 12 images at once.
Some online detectors add features that help with art, such as Illuminarty’s paid localized detection, which highlights regions that look generated. Our best AI image detector roundup compares the options.
Step 4: Weigh it before you say anything #
| Evidence | Weight |
|---|---|
| Metadata names an AI tool | Strong |
| Long history, process posts, consistent style | Strong evidence of a human artist |
| Several random rendering errors | Moderate to strong |
| Detector says Likely AI-generated, high confidence | Moderate to strong |
| Detector says Uncertain | Weak either way |
| “It just looks too polished” | Weak |
If you’re buying a commission or judging a contest, ask for evidence directly: work-in-progress stages, a layered file, or a timelapse. That’s a normal request, and most artists are happy to show process.
If you’re a viewer who suspects an account on social media, think twice before calling it out publicly. Human artists have been accused of using AI because their clean style resembled generated work, and a detector’s false positive is a poor reason to pile onto someone. Why real images get flagged as AI explains how that happens.
What about art that uses AI for part of the work? #
Many artists now use AI somewhere in their process: generating a reference, extending a background, upscaling the final image. Whether that counts as “AI art” depends on the context. A contest may ban any AI use, a client may only care about the final result, and a marketplace may require disclosure. Detectors can pick up partial AI use, but they can’t tell you whether it breaks a particular rule.
Frequently asked questions #
Can AI art detectors be wrong? #
Yes. They can flag clean, digital human art as AI and miss AI art that has been painted over or heavily edited. Research on human versus AI art found even the best detector had specific weaknesses, so use a detector alongside the artist’s history and process.
How can artists prove their art isn’t AI? #
Keep and share process evidence: sketches, layered files, timelapse recordings, and dated work-in-progress posts. Some creative apps can also attach Content Credentials that record how a file was made. A consistent posting history over time is also convincing.
Why is anime art often flagged as AI? #
Anime and similar styles use clean lines, flat color and smooth shading, which overlap with the look of popular AI anime models. That makes both human judgment and detectors less certain, so process evidence matters more for these styles.
Is it AI art if the artist painted over an AI image? #
That’s a judgment call that depends on the rules of the context. Painting over a generated base is AI-assisted work, and many contests and marketplaces require it to be disclosed. A detector may or may not flag it, depending on how much of the original generated texture survives.