Deep Live Cam It is an application that has gone viral in recent days thanks to the ease with which it allows create videos deepfake of any person from just one photograph. The software is capable of applying any face you want onto another individual’s face in a video stream, adapting it naturally to the poses, lighting and expressions made by the person in front of the webcamThe results are not perfect, but it is another example of how quickly technology is evolving and how fooling others is becoming easier. increasingly easier.
The software has been in development since late last year, but it has been the examples that have gone viral on social media, such as deepfakes of Elon Musk or the Republican candidate for vice president JD Vancethose who have recently worn it to number 1 in trends on GitHub, repository where it is available for free download and use this open source application.
Some experiments – it works almost flawlessly and it’s totally real-time. Took me 5 minutes to install. pic.twitter.com/Ow0QRF7WOj
— joao (@jay_wooow) August 9, 2024
The dangers posed by this type of technology are not new. The term began to gain popularity in 2017 as a combination of deep learning and fakes. At that time, the technology to do them was expensive and slow and did not work in real time, but in just a few years it has progressed enormously. To the point of knowing cases such as that of an employee of a multinational company who transferred almost 24 million euros to the scammers who made him believe he was talking to his CFO on a deepfake group video calllast February. Deep-Live-Cam puts these kinds of capabilities within everyone’s reach.
Like many open source projects on GitHub, Deep-Live-Cam unites several existing programs under a new interface. It first detects faces in both the source and target images, such as a live video frame. Then, it uses a pre-trained AI model called inswapper to perform face swapping and another model called GFPGAN to improve the quality of swapped faces by correcting details and artifacts that occur during the face swapping process.
The inswapper model, developed for a project called InsightFacecan guess what a person in a photo would look like using different expressions and from different angles. To achieve this, it was trained with a vast data set containing Millions of facial images of thousands of individuals captured from various angles, under different lighting conditions and with various expressions.
Deep-Live-Cam is capable of separate the specific characteristics of identitywhich remain constant in different images of the same person, of the specific characteristics of the pose that change with angle and expressionThis separation allows the model to generate new facial images that combine the identity of one face with the pose, expression, and lighting of another.
This isn’t the only face-swapping software that can be found on GitHub. Facefusionfor example, uses the same AI model for face swapping with a different interface. In any case, these are tools that are not aimed at the average user and whose installation and use requires certain technical knowledge, such as working with Python language libraries on the computer.