Text-guided flow matching enables sample-efficient crystal structure generation
TFMat, a text-conditioned flow-matching framework, uses structured materials language as a semantic prior for a CrystalFlow generator. Across Perov-5, Carbon-24 and MP-20 crystal structure prediction benchmarks, TFMat improves one-candidate match rates over CrystalFlow and reaches a 92.04% MP-20 match rate with 20 candidates. In de novo generation, it improves element-count and density distribution alignment while retaining coarse property consistency in composition-selected outputs.
Crystal generators can propose periodic structures, but their control interfaces remain poorly matched to the mixed descriptors used in materials design. Text provides a compact way to combine composition, symmetry, prototype and property cues, yet it has not been clear whether such information can steer flow-based crystal generation. TFMat introduces a text-conditioned flow-matching framework that uses structured materials language as a semantic prior for a CrystalFlow generator. Across Perov-5, Carbon-24 and MP-20 crystal structure prediction benchmarks, TFMat improves one-candidate match rates over CrystalFlow and reaches a 92.04% MP-20 match rate with 20 candidates; in de novo generation, it improves element-count and density distribution alignment while retaining coarse property consistency in composition-selected outputs. These results position structured text as an inspectable control layer for translating human-readable materials intent into candidate crystals for downstream simulation and validation.
TFMat integrates text conditioning into a flow-matching generative model for crystal structures. The structured materials language acts as a semantic prior, guiding the CrystalFlow generator. The reported 92.04% MP-20 match rate with 20 candidates indicates improved sample efficiency over the baseline CrystalFlow. The de novo generation results suggest better alignment of element-count and density distributions, implying the text prior helps capture compositional and structural constraints.
This work demonstrates a practical interface for materials design: using natural language to specify desired crystal properties. It could lower the barrier for domain experts to interact with generative models, potentially accelerating materials discovery workflows. The inspectable text control layer may facilitate integration with downstream simulation and validation pipelines.
The ability to generate candidate crystal structures from text descriptions could streamline early-stage materials screening, reducing time and cost in R&D for industries such as energy storage, catalysis, and semiconductors. The improved sample efficiency may lower computational requirements for candidate generation.
Next observable signals include adoption of TFMat in materials science research, extensions to other crystal families or property targets, and comparisons with alternative conditioning methods. Further validation on experimental synthesis or property prediction tasks would strengthen the case for real-world impact.