{"slug": "infinitediffusion-bridging-learned-fidelity-and-procedural-utility-for-open", "title": "InfiniteDiffusion: Bridging Learned Fidelity and Procedural Utility for Open-World Terrain Generation", "summary": "Researchers introduced InfiniteDiffusion, a training-free algorithm enabling diffusion models to generate infinite, seed-consistent, and random-access content, breaking the trilemma between infinite extent, stateless generation, and learned realism. They demonstrated Terrain Diffusion, a framework for procedural terrain generation that runs 9 times faster than orbital velocity on consumer GPUs, producing realistic, unbounded virtual worlds. This positions diffusion models as a practical foundation for next-generation infinite virtual environments.", "body_md": "For decades, procedural worlds have been built on procedural noise functions such as Perlin noise, which are fast and infinite, yet fundamentally limited in realism and large-scale coherence. Conversely, diffusion models offer unprecedented fidelity but remain generally confined to bounded canvases. We introduce InfiniteDiffusion, a training-free algorithm that reformulates diffusion sampling for lazy and unbounded generation, bridging the fidelity of diffusion models with the properties that made procedural noise indispensable: seamless infinite extent, seed-consistency, and constant-time random access. To demonstrate the utility of this approach, we present Terrain Diffusion, a framework for learned procedural terrain generation with a procedural noise-like interface. Our framework outpaces orbital velocity by 9 times on a consumer GPU, enabling realistic terrain generation at interactive rates. We integrate a hierarchical stack of diffusion models to couple planetary context with local detail, a compact Laplacian encoding to stabilize outputs across Earth-scale dynamic ranges, and an open-source infinite-tensor framework for constant-memory manipulation of unbounded tensors. Together, these components position diffusion models as a practical foundation for the next generation of infinite virtual worlds.\n\nUntil now, content generation has faced a fundamental trilemma: **infinite extent, stateless generation, and learned realism** - pick any two.\nDiffusion models achieve realism but are bounded.\nClassical procedural noise is infinite and stateless, but cannot learn.\nAuto-regressive outpainting allows learned unbounded generation, but requires a shared global state that precludes determinism and random access.\n\n**InfiniteDiffusion breaks this trilemma**, transforming any diffusion model into an\ninfinite, logically stateless array, indexed only by seed and coordinates, supporting O(1) random access,\nfull determinism, and embarrassing parallelism. It internally uses only a bounded LRU cache as a performance optimization. No persistent or external state.\n\nInfiniteDiffusion achieves this by generalizing MultiDiffusion for infinite or larger-than-memory domains, acting as a drop-in replacement that reformulates the diffusion process as a lazy computation that generates only the region you request, when you request it.\n\nEach image below compares **MultiDiffusion** (top), which operates over a pre-defined, eagerly-generated, and bounded canvas, with **InfiniteDiffusion** (bottom), which imposes no such bounds.\nInfiniteDiffusion introduces little to no quality degradation compared to MultiDiffusion, while providing the benefits of infinite, stateless, and lazy generation.\n\nThe only competing paradigm for unbounded or lazy generation is auto-regression, which carries fundamental limitations that InfiniteDiffusion avoids entirely:\n\n| Auto-Regression | InfiniteDiffusion | |\n|---|---|---|\n| Random Access | O(n) | O(1) |\n| Determinism | No; Order-dependent | Yes; Order-invariant |\n| Errors | Compound | No compounding |\n| Parallelization | Sequential | Embarrassingly parallel |\n| State | Global | Functionally Stateless |\n| Training-free | No | Yes |\n\n**Terrain Diffusion is the first learned procedural terrain generator.** I introduced a technique\nthat enables diffusion models to generate outputs spanning massive dynamic ranges, from -10000m in the Mariana\ntrench to nearly 9000m at Mt Everest, all in one world. But vertical scale alone is not enough. By utilizing a cascade of diffusion models,\nTerrain Diffusion **generates features spanning hundreds of real-world kilometers**, with continents spanning millions of square kilometers.\nEach 1024x1024 relief map of terrain below spans 100km in width. And it **runs locally on consumer hardware**.\n\n**Built on InfiniteDiffusion, Terrain Diffusion inherits all of its properties:** It's functionally stateless,\ntrivially integrates into any game engine, and has almost no practical limitations.\nTo demonstrate this, it was shipped as an open-source\n[ Minecraft mod](https://github.com/xandergos/terrain-diffusion-mc)\nwith no external dependencies. Worlds can be\nshared by seed, players can teleport millions of miles instantly, and it runs in multiplayer.\nIt is also demonstrated in Unity, where the player is able to comfortably fly around the world at 3 times orbital velocity on consumer hardware.\n\nTechnical Demo (Unity + Minecraft)\n\n(Cinematic) Mod Showcase by [AsianHalfSquat](https://www.youtube.com/@AsianHalfSquat)\n\n```\n      @inproceedings{goslin2026infinitediffusion,\n        author    = {Goslin, Alexander},\n        title     = {InfiniteDiffusion: Bridging Learned Fidelity and Procedural Utility for Open-World Terrain Generation},\n        booktitle = {Special Interest Group on Computer Graphics and Interactive Techniques Conference Conference Papers},\n        year      = {2026},\n        pages     = {10 pages},\n        publisher = {ACM},\n        address   = {New York, NY, USA},\n        doi       = {10.1145/3799902.3811080},\n        url       = {https://doi.org/10.1145/3799902.3811080},\n        series    = {SIGGRAPH Conference Papers '26}\n      }\n```\n\n", "url": "https://wpnews.pro/news/infinitediffusion-bridging-learned-fidelity-and-procedural-utility-for-open", "canonical_source": "https://xandergos.github.io/terrain-diffusion/", "published_at": "2026-07-12 19:56:07+00:00", "updated_at": "2026-07-12 20:06:36.013164+00:00", "lang": "en", "topics": ["artificial-intelligence", "machine-learning", "generative-ai", "computer-vision", "ai-research"], "entities": ["InfiniteDiffusion", "Terrain Diffusion", "MultiDiffusion", "Perlin noise"], "alternates": {"html": "https://wpnews.pro/news/infinitediffusion-bridging-learned-fidelity-and-procedural-utility-for-open", "markdown": "https://wpnews.pro/news/infinitediffusion-bridging-learned-fidelity-and-procedural-utility-for-open.md", "text": "https://wpnews.pro/news/infinitediffusion-bridging-learned-fidelity-and-procedural-utility-for-open.txt", "jsonld": "https://wpnews.pro/news/infinitediffusion-bridging-learned-fidelity-and-procedural-utility-for-open.jsonld"}}