The future of gaming isn't built level by level by expert designers—it's imagined, generated, and played by anyone, in real time.
Meet Mirage: the world's first real-time generative engine enabling live UGC gameplay through state-of-the-art AI World Models. Built to support dynamic, interactive, and sustained gameplay, it enables entire worlds to be generated and modified live—through natural language, keyboard, or controller input.
Mirage supports various genres, and today we are releasing two playable demos: Urban chaos (GTA-style) and Coastal drift (Forza Horizon-style) —both are fully generated on the fly. These aren't scripted experiences; they're living, playable simulations shaped in real time by each player.
Mirage sits at the cutting edge of AI. Compared to recent works, such as AI Doom and Genie from Google, AI Minecraft from Decart, and AI Quake II from Microsoft, Mirage offers several distinct advantages:
It enables user-generated content (UGC) through text input at any moment during gameplay-marking a significant leap beyond previous world-model-based game generation approaches.
It produces more photorealistic visuals, moving beyond the pixelated or blocky styles in earlier systems.
It supports much longer interactive experiences, with playable sequences extending beyond ten minutes.
UGC 2.0: The Rise of Generative Play
Traditional games are pre-authored: the city is laid out, missions are scripted, and the experience is ultimately finite. Mirage breaks this boundary by letting players create new experiences dynamically as they play.
Using natural language, keyboard input, or controller actions, players will be able to request an alley to escape through, spawn a vehicle, or expand the city skyline on the fly. The game responds instantly, weaving these user-generated elements seamlessly into the ongoing simulation. The world isn't just interactive—it's co-evolving with the player.
Mirage can generate games in various genres—from racing to RPGs to platformers: ◀ ▶
This is UGC 2.0:
Where anyone can generate their own games from just a text prompt
Where players create, evolve, and reshape the game live as they play
And where every experience is unique, dynamic, and never pre-scripted
Built on Next-Gen AI Mirage is powered by cutting-edge World Model research, featuring a large-scale, transformer-based autoregressive diffusion model capable of generating controllable, high-fidelity video game sequences. This technology framework integrates several key innovations: A Foundation Trained on Games Mirage is built on a robust training foundation designed for understanding and generating rich gaming experiences. This foundation begins with the large-scale collection of diverse game data from across the internet—providing the breadth needed to capture a wide array of game mechanics and styles. To complement this, we built a specialized data recording tool that captures high-quality, human-recorded gameplay interactions. These curated sessions enrich the dataset with nuanced, high-fidelity examples crucial for training a model to grasp intricate player behaviors and contextual game logic. The collected and recorded data is then fed into a vertical training pipeline—a domain-specific training approach focused on games. This allows the model to deeply internalize game world rules, systemic patterns, and interactive dynamics. The result is a model capable of generating coherent, realistic, and flexible game content, moving beyond the constraints of traditional, pre-authored games. Interactive Generation Meets Real-Time Control Mirage redefines real-time interaction by integrating frame-level prompt processing into its core. This enables players to input commands or prompts that are immediately interpreted in the context of ongoing gameplay. Playable anywhere via cloud gaming: Dynamic Input System: Mirage processes player input—primarily via keyboard and text—with ultra-low latency for near-instant response.
Real-Time Output: Visual updates are streamed back to the browser through a full-duplex communication pipeline, where input and output flow in parallel, eliminating lag and ensuring fluid interaction. Under the hood, Mirage is powered by a customized causal Transformer model, augmented with: Specialized visual encoders
Revised positional encoding
A structure optimized for extended interactive sequences Borrowing from both LLMs and diffusion models, Mirage supports coherent, high-quality content generation. We applied an advanced distillation strategy to the diffusion component that ensures both speed and quality. Players can reshape the environment at any time using natural language, triggering instant world updates. A long-context window powered by KV caching maintains visual consistency, even as the world evolves in real time. Key Capabilities Playable at 16 FPS: Engage in real-time interactions delivered in Standard Definition (SD).
Engage in real-time interactions delivered in Standard Definition (SD). Dynamic UGC: Players can alter the world using natural language commands.
Players can alter the world using natural language commands. Extended Gameplay: Mirage generates minutes-long interactive gameplay with consistent visual.
Mirage generates minutes-long interactive gameplay with consistent visual. Cloud Streaming: Instant cross-platform gameplay from anywhere. No download needed.
Instant cross-platform gameplay from anywhere. No download needed. Infinite Replayability: Every playthrough is unique.
Every playthrough is unique. Multimodal Control: Accepts text, keys, and controller input.
Built by a World-Class Team Mirage was created by a deeply technical, creatively-driven team of AI researchers, engineers, and designers. With experience from Google, Nvidia, Amazon, SEGA, Apple, Microsoft, Carnegie Mellon University, and UC San Diego, we are focused on pushing the boundaries of generative world models. The Road Ahead Generative play is more than a feature—it's a new medium. Mirage points toward a future where games aren't downloaded or designed—they're imagined, prompted, and lived. For creative platforms, it unlocks immediate, personalized content loops.
For entertainment ecosystems, it offers infinitely replayable formats.
For communities, it lowers the barrier to game creation to a single idea.
And many more... We're just getting started.
Acknowledgement: Mirage Coastal Drift demo used training data from Feng, Ruili, et al. "The Matrix: Infinite-horizon World Generation with Real-Time Moving Control."
For inquiries, email us at [email protected]