This project explores language-based machine learning as a generative design tool, positioning artificial intelligence (AI) not as a means of optimisation but as co-author of architecture. At its centre is a fine-tuned language model that translates weather-based text prompts into tree geometries, each shaped by specific environmental conditions. These forms are not predefined but probabilistically generated, embedding ecological logic into their structure.

The tool, Text2Forest, is deployed across a site to algorithmically plant a forest — each tree unique and tied to its own climate narrative. This forest is not decorative, but productive: a living index of environmental variation and a material resource for architectural intervention. Architecture emerges not from idealised forms but from what the forest provides, shifting authorship toward procedural cultivation.

By merging natural language processing with growth algorithms, the project challenges conventional design control. It embraces indeterminacy and decentralised authorship, reframing the relationship between environment, computation, and material outcome. Forestry becomes a digital, linguistic, and architectural act.