Many global vegetation and biome models represent plant diversity through plant functional types (PFTs) and physiological processes. To map biomes, established models rely on fixed PFT definitions and established biome attribution schemes, which limit their adaptability to investigate new questions. Increased availability of high-resolution climate and species distribution datasets, as well as advanced computational power, now enable regional parameterization and conceptual extension of biome models, but this potential remains underutilized. Here, we present Biome.jl, a framework that unifies climate-based biome classification and mechanistic simulations of biological processes within a single, modular engine. The Biome.jl framework introduces a generalized dominance-based competition scheme in which the attribution of model pixels to a biome is determined by environmental suitability, potential productivity of PFTs, rules to translate PFTs to biomes, and biome dominance hierarchies. We evaluate Biome.jl against independent datasets, demonstrating that improved flexibility does not come at the cost of predictive power. We then illustrate the framework's customizability through two case studies: a regional parameterization of PFTs using multi-parametric grid search inference based on empirical data, and an extension of the base model with a novel stem succulent PFT and biome definition, achieved without modifying model components. By facilitating the comparison of biome concepts, offering flexible functional parameterization, and supporting multi-resolution implementation, Biome.jl redefines equilibrium biome modeling as a modular, hypothesis- and data-driven endeavor. It provides a foundation for testing climate-vegetation relationships and exploring uncertainty in biome projections under past, current, and future environmental conditions.