AI Artwork Checker · How it works

Several signals. One considered result.

A useful assessment combines different kinds of evidence, explains uncertainty and leaves room for human judgment.

01 · Prepare the artwork

Start with the best available file

Choose the original digital file where possible. A screenshot, social-media download, photograph of a canvas or repeatedly compressed copy may have different pixel patterns and little useful metadata. Those changes do not make a piece AI-made, but they can reduce evidence available to a detector.

The product is intended to accept common artwork formats such as JPG, PNG and WebP. Upload limits depend on the live checker configuration; confirm the current limits in the upload flow before submitting a file.

02 · Examine visual patterns

Look for statistical traces in the pixels

Generative systems can leave patterns in texture, edges, fine detail and spatial frequencies. Automated analysis can look for features associated with known generation methods. This is statistical evidence, not a visible signature that appears in every generated piece.

Art styles, resizing and compression can change the measured patterns. A brushy digital painting can naturally differ from a photograph; that alone is not evidence of AI generation. Signals need to be interpreted with medium and file history in mind.

03 · Check embedded information

Read metadata and content credentials

Some creative tools include EXIF metadata or C2PA content credentials. When present and intact, these may describe a file's software history or signed provenance. Their absence does not establish that a person made the work: metadata is often removed during export, editing or reposting.

A credential can support a provenance claim, but should be verified according to its issuer and scope. It does not automatically settle who created the art or how much AI assistance was used.

04 · Compare model fingerprints

Match against known generators

Where reference data exists, a detector can compare a file with patterns associated with known model families. A match is a lead, not a guarantee of a specific tool or version. Generators change frequently and many share techniques.

05 · Read the report

Weigh the signals together

A strong report should show a verdict, confidence and the evidence that contributed to it. Conflicting signals should lower confidence or be called out rather than hidden behind an overconfident label. Use the report alongside the artist's process, provenance and context.

Learn more about accuracy and limitations, or see the artwork comparison demo.