Connecting the dots: A new wave of AI-assisted creation promises to redefine how virtual worlds are built, but it also ignites questions about authorship, consent, and control. The latest example comes from Roblox, which has released a beta tool that allows players to convert text prompts into functional 3D objects within live games. At the same time, Japan's creative community is issuing one of the most forceful rebukes yet of generative models, calling them a structural threat to artistic livelihoods. Taken together, these developments capture both the momentum behind – and the unease surrounding – the next stage of automated creativity.
The new Roblox feature is built on the same underlying technology as Roblox Cube, the company's 3D asset generator introduced last year. While the first iteration produced only static geometry, the updated engine links prompt-based generation to the platform's runtime physics, collision, and scripting layers, allowing newly created assets to be immediately interactive.
For now, Roblox's AI system interprets prompts using two predefined schemas: one for single-mesh static objects, and another for physics-bound, four-wheeled vehicles that respond to user input.
Each generated asset inherits parameters from its schema – including bounding boxes, material properties, and mobility constraints – before being compiled into the runtime environment. In internal testing, users of Wish Master, a sandbox designed to trial the feature, created more than 160,000 AI-generated items over six months. Developers reported a 64 percent increase in player engagement and session length.
Roblox engineers describe the current implementation as a foundation for a fully open "vocabulary schema system," which would allow prompts to generate any type of object or behavior that the platform's physics and scripting APIs can support.
Behind the scenes, Roblox is reportedly integrating elements of its ongoing "real-time dreaming" research – a project that fuses large language model inference with procedural scene synthesis. Demonstrations by CEO Dave Baszucki show prototype environments that respond to natural-language prompts by dynamically rebuilding terrain and lighting effects.
The announcement comes just days after Google debuted its Genie project, which generates playable, physics-aware "worlds" directly from video or text input. Investor enthusiasm surrounding Genie triggered short-term declines in Roblox's share price, suggesting concern that broader, model-driven world building could disrupt traditional interactive development pipelines.
The push into text-to-game technology underscores how quickly generative models are moving from concept to deployment. As creation becomes increasingly algorithmically assisted, questions about who benefits – and who loses – are no longer theoretical.
In Japan, where illustration and manga underpin much of the country's creative economy, the same technical milestones are being met with alarm. The Freelance League of Japan (FLJ) surveyed nearly 25,000 creative professionals – most of them visual artists – to gauge sentiment toward generative AI. The results reveal overwhelming apprehension: 88.6 percent of respondents consider AI a "serious threat" to their ability to earn a living, and 93.3 percent fear displacement in current or future contracts.
Roughly 12 percent of respondents say they have already experienced income loss attributable to AI image generators replacing commissioned artwork, and around 10 percent report taking secondary jobs outside creative work.
Beyond the financial impact, respondents emphasized accountability. More than 92 percent want generative systems to disclose the exact copyrighted materials used as training data – a requirement that would expose the opaque dataset sourcing common in large-scale model development.
A majority, 61.6 percent, favor a "permission-first" framework in which creators must explicitly approve the use of their data for training – an inversion of the opt-out model prevalent today. Meanwhile, 26.6 percent support outright prohibition. Even proposals for royalty-sharing or licensing systems drew limited enthusiasm, with roughly one-third of respondents rejecting all presented options as inadequate.