A new project, Cells2Pixels, has been developed to solve a key problem in Neural Cellular Automata (NCA): computational complexity when working at high resolutions. Thanks to a hybrid approach, this technology allows for the real-time generation of detailed PBR textures and complex biological structures.

image

What Happened

The Cells2Pixels architecture has been introduced, utilizing a coarse grid in combination with a lightweight neural decoder called LPPN (Local Pattern Producing Network). This combination allows the primary computations to be shifted to a low resolution, while the decoder renders details at any arbitrary resolution, including the synthesis of PBR textures for 2D and 3D objects.

Context

Traditional Neural Cellular Automata (NCA) face the problem of quadratic growth in computational costs as grid resolution increases, making their use in highly detailed environments practically impossible. The project proposes a way to bypass this limitation through the use of an implicit decoder.

Why It Matters for the Industry

This technology paves the way for creating high-quality procedural textures and dynamic objects with biological plausibility suitable for real-time rendering. This could lead to the integration of such methods into game development pipelines (Unreal Engine, Unity) and the creation of tools for the automatic generation of living, self-organizing environments.

Why It Matters for Users

3D content developers and procedural generation specialists gain a tool for creating detailed, self-organizing structures and textures without critical GPU load. The project is already available as open-source code on GitHub and includes an interactive demo for prototyping.

What Is Not Yet Known / Limitations

There is a difference in assessments of potential: the technical community focuses on architectural efficiency, while business-oriented market participants place primary emphasis on the commercial potential in game development.

Sources

Author

Look at AI, Editorial Staff