An AI-powered DeepFake Detection system capable of detecting manipulated images and videos using Deep Learning techniques. This project performs comparative analysis of multiple CNN-based architectures and implements a Flask web application for real-time prediction.
Deepfake technology uses Artificial Intelligence and Deep Learning to generate highly realistic manipulated images and videos, making it increasingly difficult to distinguish fake content from authentic media.
This project proposes a Spatio-Temporal Deep Learning Framework that combines image-based and video-based deepfake detection using:
- CNN
- MobileNetV2
- Xception
- VGG16
- CNN-LSTM Hybrid Model
The project evaluates model performance on HiDF and FaceForensics++ benchmark datasets and deploys the best-performing model through a Flask web application.
- DeepFake Image Detection
- DeepFake Video Detection
- CNN-based Spatial Feature Extraction
- CNN-LSTM based Spatio-Temporal Learning
- Comparative Performance Analysis
- Flask Web Application
- Confusion Matrix
- ROC Curve
- Classification Report
- Accuracy, Precision, Recall & F1 Score Evaluation
- CNN
- MobileNetV2
- Xception
- VGG16
- CNN + LSTM Hybrid
Due to GitHub's file size limitations, the datasets are not included in this repository.
You can download them from the following sources:
🔗 https://zenodo.org/records/16140829
🔗 https://www.kaggle.com/datasets/xdxd003/ff-c23
After downloading, place the datasets in the appropriate project directories before training.
The following metrics were used to evaluate model performance:
- Accuracy
- Precision
- Recall
- F1-Score
- Confusion Matrix
- ROC Curve
- AUC Score
- Python
- TensorFlow
- Keras
- OpenCV
- Flask
- Jupyter Notebook
- Visual Studio Code
- Git
- GitHub
DeepFake_Detection_DeepLearning/
│
├── static/
├── templates/
├── datasets/
├── App.py
├── extract_frames.py
├── video_processing.py
├── CNN_Training_HiDF_Dataset.ipynb
├── DeepLearning_Model_HiDF_Dataset.ipynb
├── DeepFakeModel_FaceForensics++_Dataset.ipynb
├── EDA.ipynb
├── requirements.txt
└── README.md
Clone the repository
git clone https://github.com/IAmBiswabhusan/DeepFake_Detection_DeepLearning.gitMove into the project
cd DeepFake_Detection_DeepLearningInstall dependencies
pip install -r requirements.txtRun the Flask application
python App.pyOpen your browser
http://127.0.0.1:5000/
- MobileNetV2 achieved the best overall performance for image detection.
- CNN also demonstrated competitive accuracy.
- Xception effectively detected subtle manipulation artifacts.
- VGG16 showed comparatively lower performance.
- CNN achieved the highest accuracy among all evaluated models.
- MobileNetV2 and Xception showed good generalization capability.
- CNN-LSTM successfully combined spatial and temporal feature learning for deepfake video detection.
The web application allows users to:
- Upload Images
- Upload Videos
- Detect whether the uploaded media is Real or Fake
- Display prediction confidence
- Real-time DeepFake Detection
- Explainable AI (Grad-CAM)
- Transformer-based architectures
- Audio DeepFake Detection
- Multimodal DeepFake Detection
- Cloud Deployment
- GPU Optimization
Biswabhusan Mohapatra
MCA
School of Computer Science
Odisha University of Technology and Research (OUTR)
GitHub: https://github.com/IAmBiswabhusan
Copyright (c) 2026 Biswabhusan Mohapatra This project is intended for academic and research purposes.