Accuracy guide
AI Image Detector Accuracy: What the Score Can and Cannot Tell You
Accuracy is variable, not fixed. Compression, screenshots, edits, missing metadata, and model coverage all affect detector confidence.
This result is a probability estimate, not proof. Use it as one signal, not a final decision.
Accuracy is variable, not fixed
Detector scores depend on image quality, transformation history, and the models used to create or edit the image. A high score is not legal proof and a low score does not prove camera origin.
False positives and false negatives
- False positives can happen when a camera image is heavily compressed or stylized.
- False negatives can happen when AI images are post-processed, screenshotted, or blended with camera content.
- Confidence should shape the next step, not replace it.
Metadata and C2PA as supporting signals
EXIF metadata, C2PA credentials, and source context can strengthen or weaken an interpretation, but missing data is not proof of AI generation.
Responsible-use checklist
- Check the original source.
- Run reverse image search.
- Review metadata and provenance.
- Ask a human reviewer for important decisions.