There have been huge advancements in recent years in the area of AI “deepfakes”, or fake photos or videos of humans created using neural networks. Fake videos of a person usually require a large number of photos of that individual, but Samsung has figured out how to create realistic talking heads from as little as a single portrait photo.
In a newly published paper titled, “Few-Shot Adversarial Learning of Realistic Neural Talking Head Models,” a team of researchers at the Samsung AI Center in Moscow, Russia, share their new system that has this “few-shot capability.”
The Samsung AI first undergoes extensive “meta-learning” by studying a huge trove of videos. Once it’s familiar with human faces, it’s able to create talking heads of previously unseen people using one or a few shots of that person. For each photo, the AI is able to detect various “landmarks” on the face — things like the eyes, nose, mouth, and various lengths and shapes.
Here are some examples of famous historical figures in “living portraits” created using this technique.
The technology works on not only photographs but portrait paintings as well.
While the new Samsung AI can be trained using just one photo, adding a few more (e.g. 8 or 32 total) helps to create even more convincing results.
Here’s a video that introduces the new technology and shows more examples:
This technology is still relatively young, so it remains to be seen what kind of real-world applications we’ll see it make impacts in.