DreamWorld brings text-to-3D asset generation to a Steam MMO

Building a custom fortress in sandbox games usually means hours of placing blocks by hand or firing up complex modeling software to import assets from scratch.

DreamWorld wants to turn generative AI into a built-in hotkey for world building.

California-based startup DreamWorld announced that its sandbox title, DreamWorld: The Infinite Sandbox MMO, will host its first Steam playtest starting October 10. Access is limited and will roll out in batches through Steam's sign-up system, with the upcoming session scheduled to run for 12 hours.

Text prompts meet sandbox building

The headline feature of this playtest is an in-game text-to-3D asset generator built in partnership with 3D generative platform Meshy.

Instead of switching between external software suites, players can simply type a text prompt directly inside the game client to generate custom 3D models and drop them into the world.

DreamWorld is pairing this AI tool with a suite of advanced construction features, including copy-paste commands, volumetric selection, and a fine-tuning HUD. The company claims these systems allow players to build larger structures significantly faster than in traditional sandbox games.

Beyond building, the game features an infinite, procedurally generated world with diverse biomes where players can explore, craft, and team up with friends to fight massive world bosses like Vyne.

A single-server blend of Minecraft and WoW

DreamWorld operates out of Redwood City with a lean crew of 15 people, but it has already raised $3.2 million from notable backers including YCombinator, a16z partner Andrew Chen, and Twitch co-founder Kevin Lin.

CEO Garrison Bellack brings an unconventional engineering pedigree to the table, having previously worked at Apple, Google, and Meta. He helped remove the physical home button on the iPhone X and spent three years building autonomous driving AI before founding DreamWorld during the pandemic.

Bellack explained that his goal was to merge the boundless creative sandbox of Minecraft with the massive social scale of World of Warcraft. According to him, what sets DreamWorld apart is that every player and piece of user-generated content resides on a single massive shared server.

Why it matters

Baking real-time text-to-3D generation straight into a live multiplayer game is an ambitious move. For casual players, it virtually eliminates the learning curve of 3D modeling, turning creative ideas into tangible in-game objects in seconds. For dedicated builders, it could dramatically shorten the time needed to prototype custom props and set pieces.

Across the broader gaming industry, user-generated content platforms have spent years trying to lower creative friction. While platforms like Roblox still rely heavily on desktop creator suites or simple block-building, DreamWorld is betting that generative AI can live inside the gameplay loop itself. If it works, rival virtual worlds will feel heavy pressure to embed native AI generation directly into their clients.

That said, we think there are glaring technical hurdles ahead. AI-generated 3D meshes are notoriously messy, often plagued by unoptimized polygons and inconsistent textures. When thousands of players on a single shared server start spawning custom models on demand, rendering budgets and server bandwidth will face an extreme test. This 12-hour playtest is an intriguing experiment, but it will quickly reveal whether prompt-based MMOs are ready for prime time.