
Today Moonlake AI emerged from stealth with a $28 million seed from AIX Ventures, Threshold, and NVIDIA Ventures, plus more than a dozen leading AI operators and researchers. The pitch: an AI platform that allows anyone to vibe code interactive worlds in minutes.
What Moonlake is building
Moonlake’s system takes a short description “a medieval world with a bustling shopping district,” or “a warehouse with three robot arms testing grasp policies”, and compiles it into an interactive world. Edits are conversational and near-instant. Change physics, add rules, drop in agents, and keep playing without restarting a build.
Under the hood, the platform stitches together:
- Multi-modal reasoning to plan spatial layout and object relationships
- Tool use across modern engines to place assets and wire scenes
- Program synthesis for game logic and agent behavior
- A simulation layer for reinforcement-learning and embodied-AI training
- A real-time diffusion step to reskin and restyle the world
The goal is to remove the early friction of prototyping and world-building while keeping expert-level control when needed.
Why it matters
Interactive worlds are slow and expensive to make. Prototyping often takes months. Small changes ripple across code, art, physics, and tooling. That slows iteration and kills ideas. Moonlake wants to compress that loop to minutes, so teams can test more concepts, and individual creators can ship without a large studio pipeline.
“Game design is about to become 10X better with Moonlake’s AI platform,” said Shaun Johnson, General Partner at AIX Ventures. “This will enable consumers and professional game designers to create high grade games using natural language. Moonlake is at the frontier of AI game design.”
“Moonlake is marrying vibe coding and world generation to create the ultimate reinforcement learning environments,” said Mo Islam, Partner at Threshold Ventures. “This unlocks entirely novel forms of agents in both the digital and physical worlds.”
Founders and origin
Moonlake was founded by Fan-Yun Sun and Sharon Lee while working out of Stanford AI Lab. Both trained in some of the world’s top AI labs:
-
Sun earned a PhD in CS at Stanford and worked at NVIDIA on large-scale 3D world generation for AI training and foundation models. In their launch press release, he said: “Simulations and games are an extension of our ideas. We want anyone to sketch a feeling or mechanic and be able to simulate it within minutes. That unlocks new workflows and gives AI agents the interactive environment it needs to learn.”
-
Lee pursued a CS PhD focused on generative AI and computer graphics, blending diffusion models with 3D engines and building models that reason about space and interactivity. She said: “Too many great worlds never made it past the sketchpad. It is not for lack of imagination, but for lack of the right tools.”
Early use cases
Moonlake says early partners are exploring:
- User-generated interactive worlds and micro-fandoms
- Rapid prototyping for studios and IP holders
- Robotics and embodied-AI simulations for training and eval
- 3D pre-viz for film and animation
- Classroom experiences where teachers build short interactive lessons
Who’s backing it
Along with the lead VC investors, Moonlake’s round includes Jeff Dean (Chief Scientist at Google Research), Steve Chen (Founder of YouTube), Naval Ravikant (Founder of AngelList), Ian Goodfellow (Inventor of GANs), Guillermo Rauch (Founder of Vercel), alongside executives from Hugging Face, Stability AI, Deepmind and OpenAI.
“I’m excited by the mass potential here,” said Steve Chen, YouTube’s cofounder. “YouTube opened a platform for the world to watch and share video; I can see Moonlake doing something similar for interactive content.”
Moonlake is in private alpha and plans to open its beta soon, interested developers and organizations can join at https://www.moonlakeai.com/.