Spaces:
Running
Very disappointed. "99% Detection Accuracy" claim is misleading, hurtful to those actually in the field, and an insult to open source community.
I hate to be so negative, and I really do commend projects that are tackling the growing issue of deepfakes around the world. Although I don't know you personally and can't comment on the true intent, it's obvious that you are trying to sell a service a profit -- which is totally fine.
What you should be ashamed of, however, is that these false claims of 99% accuracy on image detection is completely false, maybe unless somehow you entered a time machine and ended up back in 2024. The 99% claim is completely misleading, and ultimately hurts the industry as a whole.
Even a manual test of 20 generated images with zero adversarial tactics, zero manipulations, just straight raw generated images from models released in the past 6 months, have shown your "detection model" to have an accuracy rate of a meager 20%. I know its a small sample size, but I think its safe to say that whatever models you are using, are severely outdated and essentially provide zero value in real-life modern day detection. And you know that. Shame on you.
To the community: do NOT be fooled, this space is a complete marketing ploy for revenue generation. There are much better open source detection systems, some even on HF. I won't name the few that are actually still innovating, but its a rough domain to be in, so I hope everyone does their research.