1. Bhaskar Trivedi (Leader)
2. Asmit Chaudhary
3. Ganesh Rana

Dr. Udai Shanker (Professor and Head of Department)

Department of Computer Science and Engineering

Madan Mohan Malaviya University of Technology

https://mmmut.ac.in/

The rise of synthetic media and deepfakes is forcing us towards an important and unsettling realization: common belief that video and audio are reliable records of reality is no longer tenable. Hence the detection of AI-generated fake images has become a critical research problem. Our project uses Deep Learning techniques to predict whether an image is real or generated through AI models such as DALL-E, Stable Diffusion, GANs etc. The goal of deepfake detection is to identify such manipulations and distinguish them from real images. In this project, we used a convolutional neural network-based model to detect a particular kind of fakes i.e. facial images with the help of previous studies done in this space. We used Python programming language libraries viz. Tensorflow, keras, matplotlib and achieved an accuracy of 87.38% with custom CNN model and 92% with pretrained models. Our findings can be applied to real-world applications where fake image detection is critical. This system will help the end user with responsible A.I. use and imply ethical standards towards the use of Artificial Intelligence.

1. Open the notebook in Google Colaboratory.
2. Make sure to run all cell in the notebook.
3. The notebook will provide a frontend interface through a link via the Gradio system.
4. Go to the link, an entire application will be launched with the option to upload an image.
5. After the image is uploaded completely, submit it for evaluation or clear and upload another image.
6. After submission, the model outputs its prediction with % accuracy of either class, real or fake.
7. Along with the classification score, it also provides an heat map pertaining to the image showing which areas are used in order to determine the fake nature of the image.