This free AI tool lets you become anyone during video-calls
Deep-Live-Cam, a new software package, is gaining significant traction on social media due to its unique face-swapping capabilities. The AI program can extract a person's face from a single photo, and superimpose it onto a live webcam video feed. This process mimics the pose, lighting conditions, and expressions of the individual in real-time. Despite some imperfections in the results, this technology demonstrates rapid advancement in remote deception techniques. It is available for free download.
Deep-Live-Cam's rise to prominence
The development of Deep-Live-Cam began in late 2023, and has recently gained popularity due to videos demonstrating its capabilities. These include individuals impersonating figures like Elon Musk and US's Republican Vice Presidential candidate JD Vance. The surge in interest propelled this open-source project to the top of GitHub's trending repositories list, where it remains highly ranked.
How does Deep-Live-Cam work?
Deep-Live-Cam first identifies faces in source and target images, then uses a pre-trained AI model called "inswapper" for face swap. Another model, GFPGAN, enhances the quality of swapped faces, by correcting artifacts that occur during the process. The inswapper model can predict how a person might look with different expressions and angles, due to its training on a vast dataset comprising millions of facial images of thousands of people, captured from various angles under different lighting conditions with diverse expressions.
A tool for deception
The rise of Deep-Live-Cam has sparked discussions about its potential misuse. Illustrator Corey Brickley commented on the trend, stating, "Weird how all the major innovations coming out of tech lately are under the Fraud skill tree." He further emphasized the need for code words among family and friends to verify identities, due to such tools' capacity for remote deception.
Evolution and risks of deepfake technology
Deepfake technology, which involves swapping faces in videos using AI, has been around since 2017. Initially, it was slow and expensive, but projects like Deep-Live-Cam have made it more accessible for home use. This democratization of deepfake tech raises concerns about potential misuse. Earlier this year, a fraudulent video call led to a $25 million heist in Hong Kong. Experts warn that with easily available real-time deepfake software, instances of remote video fraud could rise significantly.
