CIFAKE: Image Classification and Explainable Identification of AI-Generated Synthetic Images

DOI

https://www.doi.org/10.59256/ijrtmr.20250506030

Published In

Volume 5, Issue 6 (2025)

Published Online

November-December 2025

Pages

227-230

Article Type

Original Article

Check for updates

A quick rise in fake image tools like GANs and LDMs makes it hard to tell real photos from fake ones. Our work tackles this by sorting real and AI-made pictures using the CIFAKE set. Instead of copying old designs, we built a unique CNN that learns better patterns. To see how it decides, we added Grad-CAM so users can view focus areas. The tool runs live online through Flask, showing results instantly. On top of that, it uses YOLOv8 to spot objects and check scene logic. Our findings show strong accuracy while revealing how the model pays more attention to tiny pixel flaws or small background issues instead of the central figure when spotting fake pictures. This way of working makes AI-driven deepfake identification clearer and easier to rely on.