A team of researchers from EPFL and Google Research unveiled a high-resolution Neural Cellular Automata (NCA) system that enables the generation of complex, self-organizing patterns at unprecedented resolutions. The work, presented at SIGGRAPH 2026 and detailed on cells2pixels.github.io, addresses previous limitations in NCA outputs, which were confined to low-resolution textures due to computational constraints.
The new approach overcomes the quadratic growth in training time and memory requirements associated with increasing grid sizes by optimizing local update rules applied iteratively by identical cells. This bio-inspired system exhibits properties such as regeneration, robustness, and spontaneous dynamics, enabling the synthesis of detailed textures and morphogenesis patterns. The project includes interactive demos showcasing brush-based manipulation and 3D texture generation, with code available on GitHub.
NCAs have been studied primarily for their ability to model biological growth and texture synthesis, but prior implementations struggled with scaling to high-resolution outputs. This advancement expands the practical applications of NCAs in graphics and computational biology, potentially impacting areas like procedural content generation and material design. The collaboration between EPFL and Google Research highlights the growing interest in combining neural networks with cellular automata principles.
The research paper is accessible on arXiv, and the project repository on GitHub provides tools for further exploration. The SIGGRAPH 2026 presentation marks a milestone in neural cellular automata research, enabling more detailed and scalable pattern generation than previously possible.