More than three decades ago, when films showed images of one person’s face changing into another, we learnt that it could be done with a technology called morphing. Programmers wrote code for it to happen and morphed images became all the rage for a brief while. Years rolled by and technology absorbed concepts of Artificial Intelligence (AI) more prominently. Images and videos can now be altered convincingly to show a different person or picture altogether. It came to be called deepfake and roused the interest of online users when fake videos and images of known faces began to pop up, often in unexpected circumstances. In the past few days, deepfakes of actors Rashmika Mandanna and Katrina Kaif have led to a lot of discussion on the ill effects of the technology and how it can be misused.
Deepfake, put simply, is a method of digitally altering an image, video, audio or other forms of media. But, so is morphing. The difference comes in the algorithms used for the transformation and the purpose of such an act. Deepfakes use deep learning, in which a computer is trained with a heavy amount of data to do human-like tasks. That is where the ‘deep’ in the name comes from. The ‘fake’ part of it, refers to the artificially generated output, much like a human-generated one.
“Deepfakes are usually created by training a neural network with a large number of images or videos of a specific person. The system learns how to replicate the person's facial movements, expressions, and voice, which can then be superimposed onto another person's performance to create convincingly realistic fabrications,” says Hrishikesh Bhaskaran, VP of Engineering at Kochi-based IT company Entri.
Morphing, on the other hand, does not rely on AI.
It is often used in movies, music videos, and other forms of entertainment to create a transformation effect. Morphing uses a straightforward technique, blending two images to create another image with simple software. For creating deepfake, the technology used is more sophisticated and produces content with a high degree of realism.
“Deepfakes can be extremely convincing and difficult to detect, sometimes requiring expert analysis or specialised detection software. Morphs are usually easier to spot and often intended to be noticeable as part of a creative effect,” Hrishikesh says.
As a result, deepfakes are often used with manipulative intent, while morphing is mostly used for entertainment purposes.
Hrishikesh says that there are good uses as well as bad uses of deepfakes. It is used for educational purposes, to reconstruct historical events using simulations. It is also used in the entertainment industry to ‘de-age’ actors or to ‘bring back’ deceased actors to life for cameos, he says. The ‘bad uses’ include spreading misinformation and creating fake narratives as it did in the Rashmika Mandanna case, impersonating public figures to manipulate stock prices or political events and cyber-bullying.
Read: Rashmika Mandanna deepfake: Can India’s laws keep up with AI morphing?