Show HN: Nuanced – a tool to detect AI-generated images

[aymandfire]

We built a tool to detect AI-generated images: https://trial.nuanced.dev/

As AI-generated images continue to blur the line between real and artificial, the risk for fraud and misinformation increases. Not being able to trust what you see online threatens whatever level of "realness" or authenticity online material has, including user profile images, documents, IDs, or media images.

Our models are trained on various architectures, such as Dalle-3, Midjourney, and SDXL, with continuous data integration from the latest AI image generators.

[paul_brook]

Very cool! Good job guys. Is it possible to quantify the accuracy in any way?

[john_noumenonic]

The preliminary model gets ~93% accuracy on an "in-the-wild" test dataset of ~10k images pulled from the widely used models, i.e. midjourney v5/v6, SDXL, DallE-3, etc.