Open source AI image detection.
Lens is a free, open source browser extension that detects AI-generated images as you browse — using URL patterns, C2PA metadata, pixel statistics, and frequency watermarks. Everything runs in your browser. No uploads, no account, no tracking.
Detection architecture
Four layers, stopping at confidence
Lens runs up to four detection checks on every image, stopping as soon as it reaches sufficient confidence. Each layer is independently auditable in the source code.
URL & filename recognition
Pattern matching against known AI service URL structures and filename conventions — OpenAI, Midjourney, Adobe Firefly, Stability AI, xAI, Bing Image Creator, and more.
C2PA metadata
Reads Content Credentials provenance embedded by ChatGPT, Adobe Firefly, and Bing Image Creator. When intact, detection is near-certain.
Pixel statistics
Analyses noise patterns, gradient smoothness, and inter-block noise correlation — drawing on academic AI image forensics research. Works when metadata has been stripped by social media platforms.
Frequency watermarks
FFT-based spread-spectrum analysis to detect invisible watermarks, including signals consistent with Google SynthID.
Known limitations
AI image detection is genuinely hard. Social media platforms strip EXIF and C2PA data on upload, which removes the strongest signals. Some models — particularly FLUX.1 and untagged SDXL — produce pixel statistics that overlap significantly with real photographs. Lens is honest about uncertainty: it won’t badge every image, only ones where evidence crosses a threshold, and it shows its work when it does.
Upcoming: a small on-device ML model (~3MB binary classifier) for images that slip past the heuristic checks. Contributions welcome.
Contribute
MIT licensed · Open to contributions
Bug reports, false positive reports, and detection improvements all welcome. Every report helps calibrate the thresholds.
Also available as a browser extension
Lens — AI Image Detector
Install from the Chrome Web Store — free, no account required.

