Trusted Reviews is supported by its audience. If you purchase through links on our site, we may earn a commission. Learn more.

Samsung’s deepfake AI can bring your Facebook photos to life

Samsung has developed a new artificial intelligence system capable of generating video footage from a single photo.

The software was created by researchers in a Samsung artificial intelligence lab in Russia and is groundbreaking in its ability to generate convincing deepfake footage without being fed a large catalogue of images.

The technology works by training its algorithm on landmark facial features  the shape of the eyes, mouth, nose, etc  grabbed from a visual database of over 7000 photos of celebs taken from interview clips on YouTube. The software can then map this information onto a still photo to get it moving and talking.

Related: Hollywood is finally fighting back against ‘Deepfakes’ porn

Researchers uploaded a video to YouTube showing their deepfakes  or ‘realistic talking head models’  in action.

The video features a section entitled ‘Living portraits’ within which Marilyn Monroe, Salvador Dalí and the Mona Lisa can be caught moving, talking and smiling. There is certainly an educational benefit to Samsung’s research and this section suggests a potential to use deepfakes to bring history to life. In the research paper, Samsung argues that the ability to create these talking heads has “practical applications for telepresence, including videoconferencing and multi-player games, as well as special effects industry“.

The paper also highlights the difficulties associated with creating the software including the complexity of the human head and the “acuteness of the human visual system” in spotting minor mistakes in the appearance of the human head  a factor which contributes toward the uncanny valley effect.

Unfortunately the negatives seem to outweigh the positives when it comes to the creation of deepfakes. There is fear over these video manipulation techniques being used to spread misinformation and ‘fake news’ leading to election tampering and fraud.

To the surprise of absolutely no-one, the origins of the deepfake can be found in porn. PornHub and Reddit have actually banned the use of machine learning techniques to replace stock video footage with the faces of famous actresses and singers in porn. Taylor Swift, Katy Perry, Daisy Ridley and Sophie Turner have all fallen victim to deepfake creators and PornHub has placed the generated videos in the same category as revenge porn, both being non-consensual.

Related: PornHub and reddit ban ‘deepfakes’ fake celebrity videos

So far, Samsung is the only company to crack the world of deepfakes with a realistic single image model but, at the rate that technology is progressing, it might not be long before just about anyone can hijack your Facebook photos and bring them to life.

Why trust our journalism?

Founded in 2004, Trusted Reviews exists to give our readers thorough, unbiased and independent advice on what to buy.

Today, we have 9 million users a month around the world, and assess more than 1,000 products a year.

author icon

Editorial independence

Editorial independence means being able to give an unbiased verdict about a product or company, with the avoidance of conflicts of interest. To ensure this is possible, every member of the editorial staff follows a clear code of conduct.

author icon

Professional conduct

We also expect our journalists to follow clear ethical standards in their work. Our staff members must strive for honesty and accuracy in everything they do. We follow the IPSO Editors’ code of practice to underpin these standards.