Deepfakes Just Got a Lot More Convincing

Generating a deepfake is now so easy it’s become a TikTok trend. But the vast majority wouldn’t fool you. Enter: new research from IIIT Hyderabad and the University of Bath.

Getting N*Sync

The paper proposes a neural network that can not only parse whether a frame’s generated lip movement matches that moment of audio, but also contextualize it with surrounding frames. Translation: Deepfakes are becoming more convincing.

The researchers also laid out real-life use cases besides making political figures vibe to a trending song:

• Online lecture videos in any language
• More accurately dubbed movies
• Public press conferences/addresses in any language
• Easier animation for CGI characters in movies and video games

The flip side: There’s a reason why deepfakes have a bad rap—their potential for large-scale misuse. The more realistic deepfakes become, the more power bad actors have to spread misinformation.

• The researchers have open sourced their work, hoping it can also boost detection of potentially harmful deepfakes.

Big picture: Deepfakes are crossing a new threshold. What's less clear is who will ultimately benefit from these advancements.