The parent company of TikTok, ByteDance, recently unveiled a new AI system called OmniHuman-1. This AI model can turn a photo into a video of a person singing, talking, and moving around naturally. OmniHuman-1 takes AI-generated media to a whole new level, surpassing older models that could only animate faces or upper bodies.
This AI tool’s secret lies in its massive training data sets and smart AI design. ByteDance researchers fed this AI model over 18,700 hours of human video footage using a unique “omni-conditions” approach (a strategy where a system or model learns from and integrates information across several different conditions or data sources simultaneously).
ByteDance researchers shared that animation has improved a lot in recent times, but existing methods still struggle to scale up. Using multiple input types, the team prepared this AI to be even more efficient and capable of creating lifelike human movements.
There are AI music generators impacting how people compose music, but Omni Human-1 can create videos of people from images that deliver speech, play music, and perform many other activities with realistic motion.
The tool puts ByteDance, and therefore TikTok, in the competitive race with tech giants like to create the most realistic-looking footage of AI-generated humans.
