From video and audio manipulation through to the ability to generate believable misleading news articles, the past few years have seen a rapid advancement in manipulation software. Now – based off of research done by Adobe, Stanford University, the Max Planck Institute for Informatics and Princeton University – the ability to manipulate video has been made easier than ever.
Deepfakes – the term given to digitally altered video footage – have long relied on a somewhat complex method of inputting data from a reference video source in order to manipulate the target video. While the results of such manipulations are close to flawless, the steps taken in this process could be simplified.
According to the research published by the team, their new AI can manipulate footage based off of whatever edits you make to the videos transcript. Simply retyping a sentence results in the person on screen saying things they never actually said – complete with realistic mouth movements and believable delivery of each word.
The AI works by analysing a source video for several key pieces of information – tracking the movement of the subjects face as they speak. It then listens for each phoneme the person speaks (for reference, the english language contains over 40 phonemes). As this is happening, the AI also takes notice of each facial expression and movement associated with a particular phoneme being spoken.
Using this analysis of visual and audio components the user can then rewrite parts of the videos transcript, with the AI generating a believable voice and re-animation of the face to match the new transcript.
As terrifying as this system is, certain limitations mean that its use is still constrained. For an accurate analysis and manipulation of visuals and audio, the source video must be at least 40 minutes long. Shorter videos are not viable – for the time being – however long form interviews, speeches and similar talking head appearances are now at risk of manipulation.
To counter the obvious risks associated with such a powerful AI system falling into the wrong hands, the team has acknowledged the danger and highlighted the need for regulations. “We believe that a robust public conversation is necessary to create a set of appropriate regulations and laws that would balance the risks of misuse of these tools against the importance of creative, consensual use cases.” But when has that ever stopped anyone.