Skip to content

Nuanced's Deepfake Detection Tech Outperforms on New SD3 Release

Exciting news! Last week, we put our deepfake detection technology to the test against the newly released SD3, and the results are in!

We ran 581 examples from SD3 through our system, achieving a precision of 97% in detecting AI-generated images. 🎯

On a meta-level, this shows that the progression of these models isn't as rapid as people might think. Nuanced detection tech is keeping up, if not ahead. With everyone already upset about SD3's quality, it's clear that the release is underwhelming and easy to detect. However, despite SD3 lagging in overall image quality, its ability to seamlessly insert generated text into images makes it a likely threat when it comes to document fabrication via AI. This means that SD3 makes it easier to generate fake identification documents, such as driver's licenses and passports, highlighting the ongoing need for robust detection tools to ensure the authenticity of digital documents and prevent fraud.

We are incredibly proud to see our technology generalizing so well, even with the latest models like SD3. Our team remains dedicated to pushing the boundaries of what's possible in deepfake and AI-generated image detection. As the landscape of AI-generated content continues to evolve, Nuanced is committed to staying at the cutting edge, ensuring that our solutions meet and exceed industry standards.