Oasis AI and a single NVIDIA H100 GPU has created a playable 'AI Minecraft' at 720p 20 FPS

Oasis AI, from Decart, is described as "the world's first real-time AI world model" for gaming. It offers a real-time playable AI-generated version of Minecraft running on a single NVIDIA H100 GPU to demonstrate its capabilities. It takes user input, from movement to using a tool, to generate the next frame.

It's an impressive little tech demo because what you see isn't rendered in the traditional sense that we associate with gaming; however, in its current state, playing Minecraft at a heavily compressed 720p at 20 FPS isn't exactly immersive or engaging. It's more proof of concept - generating entire frames, physics, lighting, and mechanics after being trained on watching hours and hours of Minecraft gameplay.

After playing the demo, which you can do via the Oasis AI site, you will see one pretty big flaw with how it works as a game - there is no real object permanence.

When you turn around or backtrack, the Oasis AI version of Minecraft often rearranges objects or "renders" a different version of the environment you just saw and interacted with - which can be disorienting. So, it's more of an approximation of Minecraft than the actual game. Of course, Minecraft has been playable on tablets, phones, consoles, and PCs for several years, so it's not meant to be a replacement but a look at the potential of AI-generated open-world environments. And for experiences that aren't limited to being games.

"Oasis marks our initial foray into more complex interactive worlds and offers a glimpse of what we call and hope to coin a 'Generative Interactive Experience.'" Decart explains. "This new technology could change many types of experiences by creating interactive videos that users can control in real time. Imagine a world where AI models make entertainment more personalized."

What differentiates Oasis AI from similar AI-generated game examples that we've seen is that it can generate each frame with low latency and is optimized to run on a single GPU or multiple GPUs.

"This breakthrough stems from innovations across our entire technology stack," Decart adds. "We've developed a model architecture that enables low action-to-frame latency, coupled with proprietary system optimizations targeting specific GPU and server hardware architectures."

• Read more: Half-Life 3 to push boundaries of game physics, destruction, NPC AI behavior in fresh leaks
• Read more: AI creates a playable version of the original Doom, generating each frame in real-time
• Read more: AMD's RDNA team says its working toward AI-powered 4K Path Tracing for Radeon GPUs
• Read more: OpenAI releases AI capable of thinking video games into existence