World Labs' Justin Johnson Details Persistent AI Worlds

At a fal-hosted event, World Labs co-founder Justin Johnson spoke with a16z's Justine Moore about building generated 3D environments that stay geometrically stable as a virtual camera moves around them.

World Labs co-founder Justin Johnson appeared alongside a16z partner Justine Moore at a stage event hosted by AI infrastructure company fal, where he walked through the research he co-leads on generating navigable, persistent 3D worlds from AI models. According to fal's own recap of the appearance, Johnson's work centers on anchoring generated environments in actual spatial and geometric structure, rather than treating each frame as a fresh guess. That grounding is what allows a room, a hallway, or an outdoor scene to hold its shape and layout even as a viewer's perspective shifts, instead of warping or reinventing itself the way many video generation systems still do when the camera pans or moves through space. The practical hook fal drew from the conversation is straightforward: if a generated space and its contents reliably stay put, that consistency is what makes it usable as more than a visual backdrop. A stable, spatially coherent generated environment starts to look less like a video clip and more like a buildable level or set, the kind of foundation game designers and virtual production teams need rather than a one-off animated shot. Details beyond the event description are limited, and fal did not lay out specifics on the underlying model architecture, release timeline, or how the technology might be packaged for developers. Still, the appearance signals continued momentum for World Labs, a company built around exactly this kind of spatial world generation, and underscores growing industry interest in generative video systems that behave more like consistent 3D spaces than sequences of disconnected frames. The original post from fal, including video from the discussion, can be viewed on X.