AI Founders Unite to Speed Up Roblox’s Reality Vision
▼ Summary
– Roblox acquired Morpheus AI, whose researcher Xun Huang invented Self Forcing, a technique that converts slow offline video models into fast, interactive generation engines for real-time responsiveness.
– Morpheus AI’s pixel-latent world model will merge with the Roblox Engine to create Roblox Reality, layering visual components like high-fidelity textures and physics while the engine manages structured logic and multiplayer state.
– Dynamics Lab founder Joe Chen built a real-time generative world engine allowing users to turn any image into an interactive, text-promptable world that friends can join instantly.
– Lucid AI founder Alberto Hojel invented the Roblox game cartridge harness, augmenting video world models with deterministic game engine logic to ensure reliable, consistent, and predictable multiplayer experiences.
– Roblox Reality aims for 4K resolution at 60 Hz using H200/B200-class GPUs and Morpheus AI’s long-context world models, integrating game engine and video models for consistent photorealistic multiuser worlds.
The road to truly interactive artificial intelligence is blocked by three critical obstacles: latency, consistency, and quality. By acquiring the team and technology behind Morpheus AI, along with the expertise of Dynamics Lab and Lucid AI founders, Roblox has moved significantly closer to realizing its vision for an interactive AI-powered world. For years, video models have been confined to slow, offline, bidirectional generation. After extensive post-training experimentation, researcher Xun Huang achieved a breakthrough with Self Forcing, a novel method that transforms sluggish offline video models into fast, autoregressive, interactive generation engines. Huang founded Morpheus AI to pursue real-time responsiveness, and this core technology now serves as the foundation built by an exceptionally strong research team.
Morpheus AI originally aimed to train an egocentric video world model entirely in pixel latent space, simulating real-world dynamics and physics. That mission finds its ultimate expression in Roblox Reality. At Roblox, we recognized that standalone world models, while powerful, fundamentally lack the long-term memory, consistent logic, and structured user input required for a massive multiplayer platform. This creates a clear opportunity for convergence with the Roblox Engine. Our goal is for Morpheus AI’s models to accelerate the evolution of Roblox Reality’s Roblox Video Model (Super Upsampler), layering on visual enhancements like higher-fidelity textures, realistic lighting, and fluid physics, while the Roblox Engine handles structured symbolic logic and multiplayer state synchronization.
But Morpheus and Huang’s groundbreaking work is only one piece of Roblox’s broader AI innovation.
In late 2025, Dynamics Lab founder Joe Chen joined forces with Roblox. At Dynamics Lab, Chen and his team built a powerful, flexible, and responsive real-time, general-domain generative world engine. Users could upload any image,photos, concept art, classic paintings, or children’s drawings,and step into it as a live, interactive environment. They could prompt the world with text to create surreal scenes or wild events, then share a link for friends to jump in and play instantly.
Also in late 2025, Lucid AI founder Alberto Hojel joined Roblox. Hojel founded Lucid AI on the principle that AI-generated worlds must be reliable and consistent, balancing stunning visuals with deterministic game logic. He recognized that while video world models can generate impressive, physics-rich realities, they fundamentally lack the long-term state and predictable progression required for a true game. Based on this insight, Hojel invented the Roblox game cartridge harness, which augments video world models with the reliable, deterministic logic of a dedicated game engine like Roblox’s. This convergence transforms a mere explorable world into a high-fidelity, interactive, multiplayer gaming platform.
The ambition of the Roblox Reality architecture,achieving 4K resolution at 60 Hz,represents an unprecedented technical leap for video world models. That is why deep expertise in optimization and KV compression is so critical. We are moving the Roblox Video Model (Super Upsampler) to adjacent edge data centers powered by H200/B200-class GPUs. For example, Morpheus AI’s work pioneered the long context abilities of world models, which will be vital for maintaining visual consistency across the immersive experiences Roblox creators will build.
The integration of game engine and video models is the missing piece to unlocking consistent, photorealistic multiuser worlds. This hybrid architecture combines the structured simulation power of the Roblox Engine with the generative vision of Morpheus AI world models. Building a team of innovative founders creates a talent-dense environment that accelerates innovation and streamlines development of high-fidelity, interactive world creation.
We are incredibly excited for these leading AI innovators to join forces with Roblox. We are ready to put technologies like Self Forcing and pixel-latent world models to work, building the foundational layers of photorealistic multiplayer games. Stay tuned for the transformative moments ahead.
(Source: Roblox.com)