LMN (Large Nature Model)

The Large Nature Model (LMN) is an open-source generative AI system developed by Refik Anadol Studio as the computational foundation of the DATALAND project. Unlike conventional large models trained predominantly on human-generated text and imagery, LMN is trained exclusively on data representing the natural world — spanning flora, fauna, fungi, and ecosystems, with particular emphasis on rainforest biodiversity. The dataset was assembled through partnerships with institutions including the Smithsonian, National Geographic, and the Cornell Lab of Ornithology, with infrastructure support from Google Cloud and NVIDIA, and was curated with a strong emphasis on ethical sourcing and environmental stewardship.

Architecturally, LMN follows the broad paradigm of large-scale generative models but is distinguished by its domain-specific training corpus. By grounding the model in multimodal natural world data — including images, audio recordings of wildlife, and ecological metadata — the system learns representations of biological and environmental patterns rather than human cultural artifacts. This makes it capable of generating and interpreting nature-centric content in ways that general-purpose models, trained on internet-scale human data, are not optimized to do.

The significance of LMN extends beyond its technical construction into its conceptual framing. It represents an emerging category of purpose-built foundation models designed around a specific domain or value system rather than general capability. By making the model open-source, Anadol Studio invites researchers, conservationists, and artists to build on the system, potentially enabling applications in biodiversity monitoring, environmental education, and ecological visualization that go well beyond the original artistic context.

LMN also raises important questions about what AI models are trained to reflect and amplify. Most large models implicitly encode human priorities, biases, and cultural perspectives by virtue of their training data. LMN's deliberate reorientation toward the non-human natural world is both an artistic statement and a methodological experiment — probing whether shifting the data distribution of a foundation model can meaningfully shift the kinds of understanding and creativity it enables. Whether as an art project or a scientific tool, it signals growing interest in domain-aligned, values-driven model development.