Generative AI: MegaPortraits

MegaPortraits: One-shot Megapixel Neural Head Avatars a novel machine learning architecture that generates avatar heads in megapixel resolution. It takes only one portrait image (one-shot), called source, to then transform a person’s video, called driver, accordingly. The transformation happens in great detail and is best in class right now.

In the right column under Ours, there is a screenshot of MegaPortraits’ results - find video link below: 

💡 What could this mean?

  • Business video calls: Have an all-time ready business persona of yourself for video calls. No eye bags, no dark circles, and dressed appropriately for various situations (Business vs. Business casual vs. Urban)

  • Learning experience: Revival of historical or ancient personas eg. in museums. Actors could provide the driver video stream that gets transformed into eg. Cleopatra, Otto von Bismarck, and Churchill.

  • Immersive gaming: Both ways are thinkable. One way, personalise your game character by mapping your face on top of it like for example in GTA. Another way, games that allow video calls amongst players like Fortnite , where your video stream (the driver) is transformed into your character’s appearance (the source).

  • Metaverse: This tech could be a center piece for your, perhaps modified, representation/ avatar life in virtual worlds eg. Sandbox

MegaPortraits Scientific paper.

MegaPortraits Video.

Generative AIFrancesca Tabor