The first release of Waypoint proved that real-time generative worlds were possible and marked an early step toward closing the gap between generating worlds and actually experiencing them. But it also exposed two limitations. Visual fidelity needed improvement, and accessibility needed to be expanded.
Waypoint-1.5 builds directly on that foundation. We are not trying to build better videos. We are trying to build worlds people can get lost in.
This release improves the visual quality of Overworld’s generated worlds while dramatically expanding the hardware capable of running them locally. Waypoint-1.5 produces real-time environments at up to 720p and 60 frames per second while running on a wide range of consumer hardware. The release introduces two model tiers: a 720p model for higher-performance systems and a 360p model designed to run smoothly across a variety of gaming PCs, including modern NVIDIA RTX GPUs and, eventually, Apple Silicon Macs.
Key System Improvements
Waypoint-1.5 is built around Overworld’s central thesis that generative worlds need to run in real time on local machines. That direction shapes both the model and the system around it, across accessibility, scale, and efficiency.
The most important change in Waypoint-1.5 is the introduction of dual model tiers: the 720p model targeting higher-end systems, and the 360p model optimized for a broader range of hardware. This allows generative worlds to move beyond a limited set of machines toward something that can be easily explored, modified, and played by almost anyone.
Another big improvement in Waypoint-1.5 is the scale of training data. Waypoint-1.5 was trained on nearly 100 times more data than Waypoint-1. This significantly improves the model’s ability to generate coherent environments and consistent motion.
To support real-time performance, Waypoint-1.5 also incorporates more efficient video modeling techniques. This includes approaches that reduce redundant computation across frames to help maintain responsiveness and coherence while operating within the constraints of consumer hardware.
Why This Matters for AI World Models
Much of the recent progress in generative world models focuses on visual fidelity. These systems are trained on massive datasets and large clusters that produce increasingly realistic scenes and long video sequences. But that is not what people describe when they talk about their experience. They describe the interactions, the environment, and the emergent behavior that led them there.
What makes a world feel real is how it responds, evolves, and holds together as you explore it. The difference between watching a scene and feeling inside it is what we call the immersion gap.
Closing that gap requires low latency, responsiveness, and controllability. Environments need to react instantly to player input by evolving and remaining coherent as users explore them. If generative world models require large GPU clusters to run, they remain impressive demonstrations but are difficult to use as interactive systems. If they can run locally on consumer hardware, they become a foundation for real-time interactive environments. Our latest update is designed around that idea.
Waypoint-1.5 is the next step in our path toward closing that gap. It improves visual quality while expanding the range of machines that can run generative worlds. By combining higher-resolution generation, larger training datasets, and more efficient video model techniques, it moves generative worlds closer to running anywhere.
How to Experience Waypoint-1.5
With Waypoint-1.5, we focused on improving the model and making it easier to use. There are two ways to use Waypoint-1.5: instant access and local control.
You can run Waypoint-1.5 locally through the Overworld Biome runtime. This version is designed to run across a wide range of hardware configurations, enabling real-time generative environments to run on a wide range of consumer hardware. This update also introduces a simple EXE installer, enabling users to go from download to running the model locally in minutes. If you’re interested in learning more about how the Overworld Biome was built, check out our previous blog.
The second option is to run the model instantly using Overworld.stream. This service runs the model in your browser and enables users to jump into generative environments without any setup required.
Whether you want to try the system immediately or run it on your own machine, Waypoint-1.5 is designed to support both.
The Continued Path Toward AI-Native Worlds
Waypoint all started with us asking what it would take for generative worlds to become truly interactive. Early generative models demonstrated that AI could produce convincing images and videos, but creating environments that people can explore, control, and interact with in real time presents a new set of challenges.
Waypoint-1.5 builds on our foundation with improved visual fidelity, expanded training data, and architectural improvements. Each of these strides brings us closer to a future where generative worlds are accessible to anyone to explore, build, and play in.
Download Waypoint-1.5 and run it locally with Biome in minutes to explore generative worlds on your own hardware. If you want to jump in immediately, you can also try it on Overworld.stream.
Join our Discord to share what you build and help shape what these worlds become.