Materials

Generative and multimodal AI for materials prediction and design: Progress, challenges, and perspectives

Xianyuan Liu, Charles Anjah, Benjamin E. Jolly, Jonathon F. S. Markanday, Haiping Lu

Featured August 4, 2026

AI-generated analysis — This is SciGrove's AI interpretation of the paper, not peer-reviewed content. Always refer to the original paper.

Simply

Designing new materials with AI is hard because real-world performance depends on many factors, not just basic ingredients. This paper suggests we need to look at all kinds of data, from how materials are made to how they behave, to truly find novel and useful ones.

In depth

This perspective paper introduces a materials property hierarchy and a novelty taxonomy to rigorously define what constitutes 'novelty' in AI-driven materials discovery. It argues that current AI models, often focused on composition and idealized structure, struggle to achieve 'physical' and 'deployment' novelty because they lack comprehensive, integrated multimodal data spanning processing, microstructure, and experimental characterization. The authors identify four key opportunities to bridge this gap, emphasizing the need for community-wide standards and infrastructure.

Key Takeaways

1
The paper establishes a materials property hierarchy (intrinsic to extrinsic) and a novelty taxonomy (structural, physical, deployment) to clarify the challenges in AI-driven materials discovery.
2
It highlights that current AI models primarily achieve structural novelty, while 'physical' and 'deployment' novelty are limited by scarce, imbalanced, and heterogeneous multimodal data.
3
The authors propose four critical opportunities: multimodal data construction, process-aware modeling, feasibility-first generation, and deployment-aware benchmarking, to enable the design of experimentally realizable materials.

Conceptual Flow

HIGH LEVEL

Methodology: Bridging the Novelty Gap

The paper explains that finding truly new materials needs more than just new recipes; it needs to consider how they're made and how they perform in the real world.

Basic Material Info

Add More Details

Real-World Performance

Results: The Path to Real-World Materials

They found that current AI is good at making new recipes, but struggles to make materials that actually work well in practice, highlighting a big gap to fill.

Easy Novelty (Recipes)

Harder to Achieve

Useful Novelty (Working Parts)

This breakdown was generated by SciGrove. Get the same analysis — intuition, storyboard, peer review, a runnable prototype and a glossary — on any paper you upload or paste a DOI for.