2024highReal-Time
GameNGen: Diffusion Models Are Real-Time Game Engines
Valevski et al. (Google Research)
DOOM at 20fps on TPU. PSNR 29.4. Noise augmentation halts autoregressive drift. ICLR 2025.
2024highReal-TimeState
DIAMOND: Diffusion for World Modeling
Alonso et al. (Geneva/Edinburgh)
Diffusion world model for Atari. Trains in 3 days on RTX 4090. Scales to Counter-Strike at 381M params. NeurIPS 2024 Spotlight. Open source.
2024highReal-Time
Oasis: A Universe in a Transformer
Decart/Etched
Real-time Minecraft generation. V2V layer transforms state into aesthetics at 1080p/30fps. $53M from Sequoia.
2024highReal-TimeState
Genie 2: A Large-Scale Foundation World Model
Google DeepMind
11B-parameter foundation world model. Persistent 3D worlds from single images.
2025highReal-TimeIntent
Genie 3: A New Frontier for World Models
Google DeepMind
24fps, 720p, text-to-interactive-3D. Current SOTA for general world simulation. Multi-minute spatial coherence.
2025highReal-Time
StreamDiffusion: Real-Time Interactive Generation
Aki et al.
91fps image-to-image on RTX 4090. Proves 60fps neural rendering is achievable today. ICCV 2025.
2023highReal-Time
Consistency Models
Song et al.
Foundational — single-step generation enabling real-time diffusion inference. ICML 2023.
2025highReal-Time
SANA-Sprint
NVIDIA
100ms for 1024x1024. Best quality-per-millisecond for image generation. ICCV 2025.
2025mediumReal-Time
TurboDiffusion
Dec 2025
100-200x end-to-end speedup, open source on RTX 5090.
2025highReal-Time
Shortcut Models
ICLR 2025 Oral
Dynamic quality/speed tradeoff per frame — perfect for variable-complexity scenes.
2025mediumReal-Time
LTX-Video
Lightricks
First open-source model generating video faster than playback speed.
2025mediumReal-Time
CausVid
CVPR 2025
Distills bidirectional diffusion into causal streaming at 9.4fps.
2025highReal-Time
Yan — Interactive World Generation
Tencent
Claims 1080p/60fps interactive via 3D-VAE + KV-cache shift-window denoising.
2025mediumReal-Time
HY-World 1.5
Tencent
24fps with geometric consistency, open source, full pipeline.
2024highStateReal-Time
Diffusion Forcing: Next-token Prediction Meets Full-Sequence Diffusion
Chen et al.
Bridges autoregressive and diffusion approaches. Variable-horizon coherence without catastrophic drift. NeurIPS 2024.
2026highState
PERSIST: Persistent 3D World Generation
March 2026
Persistent 3D state in latent space, camera as query, geometric consistency by construction. THE most important paper for state coherence.
2025highState
StateSpaceDiffuser
NeurIPS 2025
+14dB PSNR over DIAMOND, 50-step coherence vs 4-step via dual SSM+diffusion architecture.
2025highState
EDELINE: SSM World Models
NeurIPS 2025
Replaces DIAMOND's LSTM with Mamba SSM for immediate coherence gains.
2025highState
Self Forcing
NeurIPS 2025 Spotlight
Solves autoregressive drift by training on self-generated outputs. Directly attacks state drift.
2026highState
MultiGen: Memory for Diffusion Game Engines
March 2026
External memory for diffusion game engines. Decomposes into Memory + Observation + Dynamics modules.
2025mediumState
Long-Context SSM Video World Models
ICCV 2025
Block-wise SSM extends temporal memory beyond attention windows.
2025highState
DreamerV3
Hafner et al. (Nature 2025)
General world model maintaining state across 150+ diverse tasks with single config.
2025highIntent
PAN: Language-Conditioned World Actions
MBZUAI (Nov 2025)
Natural language control of world actions. Highest fidelity among open-source models.
2025highIntent
FOUNDER: Bridging LLMs with World Models
ICML 2025
Bridges LLMs (intent/narrative) with world models (physics/dynamics). THE architecture paper for intent understanding.
2025highIntentBoundary
A2UI Protocol
Google (Dec 2025)
Declarative intent → rendered output protocol. Shows how to structure intent-to-experience pipelines.
2024highIntent
UniSim: Universal Simulation
DeepMind (ICLR 2024)
Simulates both high-level instructions and low-level controls. Multi-modal intent understanding.
2025mediumIntent
GameGen-X: Interactive Game Generation
ICLR 2025
InstructNet for interactive control of generated game content from natural language instructions.
2026highVerificationTrust
World-in-World: Closed-Loop WM Evaluation
ICLR 2026 Oral
First benchmark platform for closed-loop world model evaluation. Key finding: controllability > visual quality.
2025highVerification
MaaG: Model as a Game
Microsoft (ICCV 2025 Workshop)
LogicNet for numerical consistency, spatial memory. Directly attacks score-tracking and physics verification.
2025highVerification
PIWM: Physics-Informed World Model
Sep 2025
60.6% improvement in physics consistency. Soft mask training + warm start inference.
2025highVerificationTrust
Critiques of World Models as Planning
Jul 2025
Systematically catalogs failure modes — shallow coherence, error explosion, generality limits.
2025highBoundary
Vid2World: Interactive Worlds from Video Diffusion
May 2025
Converts pretrained video diffusion into interactive world models. THE bridge between passive and interactive systems.
2025highBoundary
NVIDIA Cosmos: World Foundation Models
NVIDIA (Jan 2025)
Foundation model platform with fine-tuning pipeline. Shows how to build boundary interfaces for domain-specific deployment.
2025highBoundary
Wayve GAIA-2: Driving World Model
Wayve (March 2025)
Domain-specific world model for autonomous driving. Demonstrates rigid-system integration with sensors, maps, controls.
2024highTrust
Is Sora a World Simulator?
May 2024
Challenges visual quality = world understanding. Critical framing for trust calibration.
2025mediumReal-TimeState
Understanding World or Predicting Future?
ACM CSUR 2025
Comprehensive survey of world model approaches. Taxonomy and comparison.
2024mediumReal-Time
Genie 1: Generative Interactive Environments
DeepMind (ICML 2024)
Foundation work. 11B-param model generating 2D platformer environments from images.
2023mediumReal-TimeState
IRIS: Imagination with Auto-Regression over an Inner Speech
ICLR 2023
Early discrete-token world model. Pioneered autoregressive game simulation.
2025mediumReal-Time
V-JEPA 2
Meta (Jun 2025)
Video prediction via joint-embedding. Self-supervised world model approach.
2025mediumReal-Time
MineWorld
Microsoft (Apr 2025)
Minecraft world generation with structured reasoning.
2025mediumReal-Time
GameFactory
ICCV 2025 Highlight
Modular game generation pipeline from descriptions.
More papers being reviewed and added continuously.