DeepFake Detect

DFDetect is a free web tool that checks photos for deepfaked faces. A user uploads a PNG or JPEG image of up to 2 MB, and the service detects each face in the picture and gives it an authenticity score, the model's estimate of how likely that face is real rather than synthetic, color coded red below 20 percent, orange between 20 and 80 percent and green above 80 percent. It suits journalists, moderators, researchers and anyone who needs a quick check on whether a face in an image has been generated or manipulated. The model uses a customized EfficientNet architecture, and the site reports 96.36 percent accuracy, 94.95 percent precision and 97.94 percent recall on its validation data, with results it says outperform XceptionNet and MesoNet on the FaceForensics++ benchmark. For developers, an MIT licensed GitHub repository provides the full pipeline to train a detector from scratch in Python with Keras and TensorFlow, including scripts for extracting frames from video, detecting faces with MTCNN or Azure Computer Vision, balancing datasets and training the network on public datasets such as FaceForensics++, Celeb DF, Google Deep Fake Detection and the Facebook Deepfake Detection Challenge. The web tool is free to use with no account or payment required. A pricing page exists but its content could not be read, so no paid plans are confirmed. Two limitations are worth knowing: the site states results are for informational and research use with no accuracy guarantee, and by uploading, users agree that images may be stored and used to improve the service, including model training.