slover

DEVLOG 6 - I added OpenAI detection to my AI watermark detector library ! So uhm I was tired with Videoseal, so I just decided to do the openai implementation instead. It was pretty simple, and I tried to do it without using the OpenAI sdk because i didn’t wanted to add yet another dependency. I also added a config to pass to the detector, and a way to ignore a detector.

DEVLOG 5 - It’s been a looong time, it’s so hard… I tried to implement meta videoseal, and after countless of tries, I managed to export the detector to an ONNX model, that I could use in the library. The only problem is that I don’t know a lot about machine learning, so it doesn’t work at all. The JS script made to use the model just gives random results, often giving more score to non-ai images than to watermarked images. If anyone knows how to help with this, I would love a biiit of help. The Github repo is opened to issues and all the code is Open Source :). (Also I did the Jame Gam it was really fun :3)

Devlog 4 - Very tiny update :). I don’t have my main computer, so I can’t do heavy things, which is what I need to do right now. But I used Blob objects instead of paths so that the library is compatible with bot Node and web browsers. I also tried some things with the openai detector, but the public api needs an API key and the private one has a turnstile challenge to solve. If anyone has tips to solve that, I would be really happy to hear them ! I also experimented with pytorch export to ONNX, but I don’t have the right pc right now (not committed yet). So yeah very smoll update, thanks for reading till the end.

DEVLOG 3 - I made a logo for my AI watermark detection tool ! Also, I fixed a bug that made C2PA flag human made c2pa content as ai content.

Devlog 2 - I made the base implémentation for image and video detection !When sending an image or a video, the system will pass the file against each available detector (for now, only SynthID), and return a ImageAnalysisResult object (or a VideoAnalysisResult) with every result, and some methods to easily manage the data.I also created a base README (on mobile, without hackatime), and a ROADMAP with what I plan to implement.Here is the list of detectors I will implement, but feel free to suggest some other watermarks to implement ! C2PA (done), OpenAI verificator (in progress), Stable signature, Meta Content Seal (Audiodeal, Videoseal, Textdeal, Imageseal), Elevenlabs synthid (not sure), Claude Watermark (when public). That’s it for now !

[FIRST DEVLOG] Building my first JS project, Slover !
Slover is a tool to unify every major AI watermarking detection tool (c2pa, openai verificator, synthid-text..) in an easy to use JS library.
For now, there is a start of c2pa detector implementation, but I plan on adding a lot more of tools !
Also, I plan on making a browser extension, and a website to use it, but users will be free to integrate it wherever they want.