Crystalis Enables LLMs to Generate Complex Multi-View Visualizations

Crystalis is a novel framework designed to enable large language models (LLMs) to generate complex coordinated multi-view visualizations (CMVs), a task previously beyond their capabilities. While LLMs can create individual charts, CMVs require tight field-level coupling across data transformations, visual encodings, and interactions, making them prone to errors. Crystalis addresses this by focusing on structural correctness rather than end-to-end analytical quality, which depends on various external factors. The framework operates by decomposing a CMV into structured queries over a dependency graph, spanning data, visualization, and interaction components across three abstraction levels. It employs two complementary mechanisms: "progressive nucleation" builds each query vertically from requirement to executable object, while "semantic annealing" enforces horizontal consistency across queries at each level through layered logical checks. Evaluated on a 12-task benchmark with five frontier LLMs, Crystalis achieved up to 75% end-to-end success, a substantial improvement over an agentic coding baseline that only reached 8.3%. A user study confirmed the usability of its decomposition and iterative refinement workflow for practitioners.