Google DeepMind Unveils Genie 4, a New AI Model Combining Human Voices and Videos

Google DeepMind has announced Genie 4, a new artificial intelligence model that can generate photorealistic videos from natural language prompts and automatically add matching ambient sounds and human speech to these visuals.

Unlike its predecessor Genie 3, Genie 4 has the ability to generate sound simultaneously with visual scenes. In addition to ambient sounds like wind or wave noise, the model can create audio simulations where characters can speak to one another and respond to environmental events. Google states that this new model simulates the laws of physics better and responds dynamically to changes in lighting and camera angles.

The most striking feature of the model stands out as its ability to simulate human movements and facial expressions more fluidly. Google states that the characters of Genie 4 move naturally and can be directed in real time. Shared with a limited test group for now, the model is being used by video game developers in prototyping processes.

Google announced that it has taken strict measures regarding the security of Genie 4 and subjected it to rigorous evaluations to prevent the generation of harmful content. Furthermore, it was stated that all content generated by the model will be marked with the SynthID watermark. It was noted that the model, which is still in the development stage, occasionally struggles with visual consistency in long and complex simulations.