Hugging Face's logo

Join the conversation

Join the community of Machine Learners and AI enthusiasts.

Sign Up
Back to feed
Parveshiiii 
posted an update 14 days ago
Post
2,789
Most AI image detectors work fine in the lab and fail on social media.

The reason is simple. Real photos get compressed, resized and re-uploaded again and again, and detectors trained on clean images read that damage as a sign of fakery. A real photo that has passed through a few platforms often gets flagged as AI-generated.

AIRealNet-2 is built to fix that. It is the upgrade to our original Modotte/AIRealNet, trained on images collected from platforms like Instagram and X, so it has seen the kinds of quality loss and compression real images go through online.

The training set has 98k images, split 50/50 between AI-generated and real, so the model has no class bias to lean on. Personal photos were excluded by design, including AI-edited ones.

The architecture is unchanged: a SwinV2 Tiny backbone with a binary classification head, about 0.2B parameters, working at 256x256 resolution. It stays fast and easy to deploy for research or production.

Model: Modotte/AIRealNet-2
WideChaos

The v2 card's "Demo and Usage" snippet still loads Modotte/AIRealNet. Could you change it to Modotte/AIRealNet-2 so copying the example tests this release? https://huggingface.co/Modotte/AIRealNet-2#demo-and-usage

·
Parveshiiii
•
edited 14 days ago

Thanks a lot for pointing it out I have updated it
Kreout

A useful stress test would keep one held-out set of original images, then derive JPEG/re-sized/re-uploaded versions of each. Split by the original image before creating those variants, so copies of the same picture cannot land in both training and test. Reporting false-positive rates for real photos at a fixed detection threshold, separately for each transformation chain, would make the social-media robustness claim much easier to assess than one aggregate accuracy score.

·
Parveshiiii

we have done it but due to copyright laws we will be unable to publish the held out eval set the numbers we publish would look like a gimic
In this post