My Art Was Flagged as AI
A commenter, a contest, a marketplace or a detector said your hand-made piece was AI-generated. This page explains why that happens, how to check what your own file says, and how to prove the work is yours.
First, the honest part. AI image detectors, including ours, return a probability, not a verdict. They are trained to separate camera photographs and generator output. Stylised digital art sits between those two worlds, and some of it lands on the wrong side. A high score on your painting is a false positive, not an accusation, and it is more common for art than for photos.
Why detectors flag human art
Detectors learn statistical signatures of generator output: very smooth gradients, an absence of sensor noise, glossy skin, soft glow, symmetrical composition, and certain output sizes. Digital painting produces several of those on purpose. The pieces most often flagged share some of these traits:
- Airbrushed or heavily blended rendering, especially portraits with soft skin and rim light.
- Glow, bloom and blur filters applied across the whole canvas.
- Canvas sizes generators use: 1024×1024, 1024×1792, 1792×1024 and the SDXL aspect buckets. A square 1024 export reads as a generator size.
- Heavy upscaling or denoising on export, which removes the fine texture that reads as human.
- Exports with no metadata: a screenshot of your canvas or a re-save through a platform strips the software name and history that would have counted for you.
None of these mean you did anything wrong. They mean the file looks, statistically, closer to a render than a photograph. We publish where our model is weak at for fact-checkers.
Check what your own file says
Upload your export here. You get the detector's probability and a read of the file's metadata, so you know exactly what a commenter or a moderator would see.
Drag & drop, paste, or click to upload
PNG · JPG · WEBP · HEIC — your image is analyzed privately and deleted
This is an estimate, not proof — the detector can be wrong, producing both false positives and false negatives. Don't rely on it as your only source; verify anything important. wedetect.ai is not liable for decisions made from these results. See our Terms.
Then go deeper with the same file:
- EXIF viewer: does the export still name your software (Procreate, Photoshop, Clip Studio, Krita) and carry a creation date?
- Edited photo detector: shows the edit history and application markers your file records, which is evidence of a human workflow.
- Content Credentials checker: if your app signs exports (Photoshop, Firefly-free workflows, some cameras), this shows the signed record that says who made the file and how.
How to prove your art is human-made
- Keep the working file. The layered PSD, Procreate, Clip Studio or Krita file, with its layers, is the single strongest proof. Generators do not produce layered sources.
- Keep the sketches. Thumbnails, line art, colour studies, with their original timestamps. A folder of dated stages tells the story on its own.
- Record the process. A timelapse from your app, or a screen recording of a working session. Procreate and Clip Studio record timelapses automatically; turn it on.
- Export with metadata intact. Do not screenshot your canvas. Export from the app so the software name and history travel with the file, and keep that original even if the platform strips it.
- Attach Content Credentials where you can. Adobe apps can sign exports with a record of the tool and the edits. It is the closest thing to a signature that a moderator can verify.
- Post process alongside the piece. A WIP shot or a 10-second timelapse next to the finished image pre-empts most accusations and costs nothing.
Responding to an accusation
Keep it short and evidence-first. Reply with one process shot or the timelapse, name the software, and offer the layered file to a moderator privately. Do not argue about the detector's number; explain that it is a probability that is known to misfire on stylised art, and point to the evidence a generator cannot produce.
If a contest or marketplace used an automated score to reject you, ask which tool and which score, and ask for a human review with your working file. Most reasonable venues accept a layered source as proof.
If you were the one doing the flagging
Before calling out an artist, run the checks in how to tell if a photo is fake and ask for process shots. A single detector score, ours included, is not enough to accuse someone. The AI art detector is built to reduce false positives on stylised human work, and it still gets some wrong.
Frequently asked questions
Why did an AI detector flag my hand-made art as AI?
Detectors measure statistical patterns, not intent. Smooth digital painting, heavy blur and glow, airbrushed skin, symmetrical compositions, noise-free renders and images exported at generator-native sizes (1024×1024, 1024×1792) all look more like generator output than like a camera photo. A stylised piece can score high without any AI involvement. It is a probability, and false positives happen.
How do I prove my art is not AI?
Keep and show the evidence a generator cannot produce: the layered working file (PSD, Procreate, Clip Studio, Krita), early sketches and work-in-progress shots with timestamps, a screen recording or timelapse, and the exported file with its metadata intact. Attach Content Credentials if your software supports them. A detector score is one input; a process trail settles it.
Does the metadata on my export help?
Yes, if you keep it. Most art software writes its name into the file, and Adobe apps keep an edit history. Social platforms strip all of that, so share the original file when it matters. Our EXIF viewer and Content Credentials checker show exactly what your export carries.
Should I run my own art through a detector before posting?
It helps to know the score in advance. If a piece scores high, you know to post process shots alongside it and to keep the working file ready. It does not mean you should change how you paint.
Can I ask you to correct a false positive?
Yes. Every result on this site has a thumbs-down button. We read those, and pieces that are wrongly flagged are exactly the cases that improve the next version of the model.
Related: AI art detector · EXIF viewer · Content Credentials checker · how accurate the detector is, and where it is weak · AI detection for digital art.