AI Watermark version history - 4 versions
AI Watermark by Simon
AI Watermark version history - 4 versions
Be careful with old versions! These versions are displayed for testing and reference purposes.You should always use the latest version of an add-on.
Latest version
Version 0.6.0
Released Nov 18, 2026 - 39.35 MBWorks with firefox 140.0 and later, android 142.0 and later- Finds more AI images among thumbnails: when a page shows only a small preview, the classifier scores the larger version of the image.
- Reads the AI labels of pixiv, DeviantArt and Wikimedia Commons and shows them as "AI · pixiv", "AI · DeviantArt" or "AI · Commons". This can be switched off in the settings.
- Checks the uploaded file behind previews on Civitai and X and the 1024 px version on Flickr.
- Finds Google's SynthID in more Gemini images that were recompressed or scaled down.
- Runs the classifier on WebGPU on Apple Silicon Macs with Firefox 153 or newer, where a model run takes about a third of the time.
- Shows results sooner after a page change and while scrolling: images in view on the tab in front are checked first, and a page that was left no longer holds up the others.
- New option, off by default: a weaker "AI?" badge for images that score just below the threshold.
- Fixes images that stayed without a result, a grey "?" on placeholders that a page swaps later and slow scanning on busy pages.
Source code released under MIT (code) and CC BY-NC 4.0 (model weights)
Older versions
Version 0.5.0
Released Oct 24, 2026 - 33.05 MBWorks with firefox 140.0 and later, android 142.0 and later- Finds SynthID in cropped Gemini and ChatGPT images, in Gemini images after WebP and JPEG recompression and in ChatGPT images of every output size.
- Checks the original image behind the WebP renditions on Google's blogs.
- Recognizes the visible marks of Bing Image Creator, Meta AI and DALL·E 2.
- Reads the EXIF tags of Civitai and Reve and recognizes more ComfyUI workflows.
- Checks images inside web components.
- Fewer false alarms on WebP photos: the classifier scores lossy WebP images as JPEG.
Source code released under MIT (code) and CC BY-NC 4.0 (model weights)
Version 0.4.0
Released Oct 22, 2026 - 35.8 MBWorks with firefox 140.0 and later, android 142.0 and later- Recognizes the visible marks of Gemini, Kling, Doubao, Jimeng, Qwen and Grok.
- Reads the EXIF tags of Grok and Ideogram and finds the faint tile pattern of Doubao and Jimeng.
- Finds SynthID more reliably after JPEG and WebP compression and in all ChatGPT output sizes.
- Checks the largest version of an image the page offers for metadata and watermarks.
- Scores borderline images again at higher resolution.
- New classifier, trained with images from the newest generators and self-generated images, shipped with 8-bit weights (12 MB smaller download). The default threshold is 0.95; a threshold you set yourself is reset to it.
Source code released under MIT (code) and CC BY-NC 4.0 (model weights)
Version 0.3.0
Released Oct 13, 2026 - 47.39 MBWorks with firefox 140.0 and later, android 142.0 and later- Show the changelog on addons.mozilla.org and fix the AMO lint warning
- Keep badges on their images while the page scrolls
- Detect the actual SDXL watermark
- Detect the Amazon Nova Canvas watermark and Stability AI metadata
- Detect the Stable Diffusion and SDXL invisible watermark
Source code released under MIT (code) and CC BY-NC 4.0 (model weights)