Advanced deepfake detection powered by PyTorch and EfficientNet-B7. Extracts and analyzes up to 32 facial frames per video to detect manipulations.
Drop any video file. Our EfficientNet-B7 classifier will automatically extract faces and dissect up to 32 frames.
Drag & drop or click to browse
Max file size: 2GB
Up to 32 frames are extracted evenly across the video. MTCNN accurately detects and crops all identified faces from these frames.
Our EfficientNet-B7 neural network analyzes the extracted facial crops to detect visual anomalies and deepfake manipulation artifacts.
Confidence scores are gathered across all 32 frames. A confident strategy filters outliers to determine a reliable overall authenticity score.
If the final aggregated probability crosses the threshold, the video is flagged as a Deepfake, otherwise it is marked as Authentic Content.
Accurately identifies and isolates faces from video frames, handling varying angles and bounding boxes efficiently.
Utilizes the pretrained EfficientNet-B7 Noisy Student model, fine-tuned specifically for detecting subtle facial manipulations.
Averaging 32 frames provides a comprehensive and trustworthy analysis of the overall video input validity.