Accuracy and limits
AI image detection is an unsolved problem. This tool is designed to give you evidence and context, not a verdict.
What the result can mean
Low
We found little evidence of AI generation. That does not make an image trustworthy: new generators, edits and re-saves can leave nothing to find.
Medium
The evidence is weak, missing or conflicting. This is the most common honest answer for images shared on social media.
High
We found strong evidence, such as a generator tag, valid Content Credentials declaring AI generation, or a detector reporting a high likelihood. Even then, treat it as a strong lead rather than proof.
Known limits
- Screenshots and re-compressed images. Messaging apps and social networks strip metadata, resize and recompress images. This removes most of the evidence and can create artefacts that look suspicious.
- Edited photos. Retouching, filters, AI-assisted editing and upscaling can trigger forensic signals on genuine photos.
- Forensic heuristics are not calibrated detectors. ELA, noise and sharpness measures flag patterns. They do not identify a generator, and thresholds can be wrong for any given image.
- Metadata can be forged or removed. Camera EXIF lowers the signal slightly but can be copied onto any file.
- Content Credentials are rare. Only some tools and cameras write them, and platforms often remove them.
- Lighting, shadows and reflections need a visual AI model. They are always shown as "Not checked" unless a future review service is connected.
- Reverse image search is done by you on third-party sites. We do not see the result until you tell us.
How to use the result well
Combine the signal with the source check: look for earlier copies with a reverse image search, find the original publisher, and consider whether the context makes sense. For important decisions, rely on trusted sources and professional verification.