Overworld Unveils Real-Time Diffusion World Model for Playable, AI-Native Worlds

PROVIDENCE, R.I., Jan. 20, 2026 (GLOBE NEWSWIRE) -- Today, Overworld (formerly Wayfarer Labs) announced the release of a research preview of its local-first diffusion world model. This model is an early, experimental system targeted toward researchers, engineers, and builders. The company is inviting these audiences to explore how diffusion-based world models can operate entirely on-device and at frame rates suitable for play.

With this release, the Overworld team is laying the groundwork for a go-to foundation for world models, but it does not represent the caliber of visual fidelity, stability, or depth that the company plans to bring to the public in the future.

The work is supported by a $4.5 million pre-seed round led by Kindred Ventures, with participation from Amplify.LA, Garage Capital, Northside Ventures, Vital Stage, East Sunshine, and notable angel investors, including Logan Kilpatrick and senior leaders from Snowflake and Roblox.

This release provides an early glimpse into the company’s approach to interactive artificial intelligence (AI). The preview demonstrates that diffusion models can be structured as continuous, real-time systems that incorporate user input into every frame. Local execution, modifiability, and real-time feedback are at the core of Overworld’s model. It uses AI as the foundation in developing a technology that couldn’t exist otherwise: open-ended, human-directed worlds that behave more like lucid dreams than software.

“Technology should give people more room to imagine, not less,” said Louis Castricato, co-founder and CEO of Overworld. “The release of our research preview is meant for people who want to understand and experiment with these systems. Our team is dedicated to putting this model in the hands of researchers, engineers, and builders to help shape what this platform will become.”

With Overworld, real-time, responsive AI doesn’t need proprietary hardware or massive servers. From the outset, the team has focused on the concerns and frustrations people have regarding AI. The model runs on local GPUs to avoid round-trips to data centers and unpredictable latency. It also eliminates the environmental overhead associated with large shared computer clusters, delivers a faster experience, and keeps worlds in the hands of those building them. Anyone will be able to access and run the model, whether they have a Chromebook, gaming PC, or consumer console-class hardware.

“Since I played my first video games, I have wanted to build worlds that respond as quickly as you imagine them,” said Shahbuland Matiana, co-founder and Chief Science Officer at Overworld. “This release focuses on understanding the mechanics of this type of technology and proving diffusion models’ ability to respond to player input in real time on consumer hardware.”

Overworld was founded around the idea that interactive AI should feel personal, immediate, and fun to build with. This research preview is the first step toward a wider creative frontier. With the availability of its model, researchers, experimenters, and engineers are invited to build, test, and imagine with a new kind of engine. For more information and access, please visit the site https://over.world.

About Overworld
Overworld is a research and development studio building real-time, local-first diffusion world models. Using AI, the company is creating a new kind of interactive experience: living, adaptive worlds shaped directly by human imagination. Since its models run on consumer hardware, Overworld is able to offer increased privacy, reduced environmental impact, and the ability for players, artists, and builders to maintain complete creative control. Overworld is committed to openness, intentionality, and human agency throughout all stages of development. For more information, visit https://over.world.

Media Contact
Nicholas Young
nicholas@bulleitgroup.com

Legal Disclaimer:

EIN Presswire provides this news content "as is" without warranty of any kind. We do not accept any responsibility or liability for the accuracy, content, images, videos, licenses, completeness, legality, or reliability of the information contained in this article. If you have any complaints or copyright issues related to this article, kindly contact the author above.