MatLoom: Layered Text-to-Material Generation in a Compact Program Space
MatLoom turns text prompts into compact editable programs that render layered PBR material maps.
The paper describes a layer-oriented language whose alpha-masked layers share spatial expressions across coverage, color, and relief. Its pipeline uses pretrained language models, parser-guided repair, preview critique, and seed search without task-specific fine-tuning. On a 141-prompt benchmark, the best setup beat three diffusion baselines across four flat-layout alignment metrics. In a blind comparison, participants chose MatLoom renders 59.2% of the time. ArXiv · AI/CL/LG's note
The paper describes a layer-oriented language whose alpha-masked layers share spatial expressions across coverage, color, and relief. Its pipeline uses pretrained language models, parser-guided repair, preview critique, and seed search without task-specific fine-tuning. On a 141-prompt benchmark, the best setup beat three diffusion baselines across four flat-layout alignment metrics. In a blind comparison, participants chose MatLoom renders 59.2% of the time. ArXiv · AI/CL/LG's note
score 4