AI Watermark by Simon
Checks larger images on web pages locally for AI generation and marks hits with an AI badge.
Available on Firefox for Android™Available on Firefox for Android™
3 Users3 Users
Scan the QR code to open this extension in Firefox for Android
Extension Metadata
Screenshots
About this extension
AI Watermark checks images on the web pages you visit for signs of AI generation and marks hits with a small badge. Click a badge for details.
What it checks
Every image whose shorter side is at least 200 px goes through up to three checks:
How reliable is it?
The badge is a hint, not proof. The model flags about 1 % of real photos and catches about half of all AI images as they appear on the web. SynthID detects most unedited Gemini and ChatGPT images, but cropping removes the watermark. Many platforms strip metadata on upload. Detailed numbers: https://github.com/s-gehring/ai-watermark-addon/blob/main/docs/accuracy.md
Privacy
All checks run locally in your browser. No image and no other data leaves your computer. There is no server and no telemetry. To read the original bytes of images from other domains, the add-on fetches each image a second time, usually from the browser cache.
Permissions
"Access your data for all websites" is needed to find and check images on any page.
License
The code is MIT licensed. The bundled model weights are licensed under CC BY-NC 4.0 because they were trained on OpenFake, so the add-on may only be used and redistributed non-commercially.
Source code, issues and documentation: https://github.com/s-gehring/ai-watermark-addon
What it checks
Every image whose shorter side is at least 200 px goes through up to three checks:
- Metadata: AI labels in C2PA manifests and IPTC/XMP, generation parameters from Stable Diffusion, ComfyUI, NovelAI and InvokeAI, and the Chinese AIGC label.
- SynthID watermark: the invisible watermark that Google (Gemini, Imagen) and OpenAI (ChatGPT) embed in generated images.
- Model: an image classifier that estimates how likely an image is AI-generated.
How reliable is it?
The badge is a hint, not proof. The model flags about 1 % of real photos and catches about half of all AI images as they appear on the web. SynthID detects most unedited Gemini and ChatGPT images, but cropping removes the watermark. Many platforms strip metadata on upload. Detailed numbers: https://github.com/s-gehring/ai-watermark-addon/blob/main/docs/accuracy.md
Privacy
All checks run locally in your browser. No image and no other data leaves your computer. There is no server and no telemetry. To read the original bytes of images from other domains, the add-on fetches each image a second time, usually from the browser cache.
Permissions
"Access your data for all websites" is needed to find and check images on any page.
License
The code is MIT licensed. The bundled model weights are licensed under CC BY-NC 4.0 because they were trained on OpenFake, so the add-on may only be used and redistributed non-commercially.
Source code, issues and documentation: https://github.com/s-gehring/ai-watermark-addon
AI has been used in order to create this addon.
Rated 4 by 1 reviewer
Permissions and data
Required permissions:
- Access your data for all websites
Optional permissions:
- Access your data for all websites
Data collection:
- The developer says this extension doesn't require data collection.
More information
- Add-on Links
- Version
- 0.6.0
- Size
- 39.35 MB
- Last updated
- a day ago (Nov 18, 2026)
- Related Categories
- Version History
- Add to collection