Mistral unveils Mistral Clay: a fine-tunable 13B multimodal model for creative tooling

AI · 5 min read

Mistral's new model, Clay, is a 13B-parameter multimodal model trained with a mix of composition-focused datasets, layout annotations, and paired text-image design briefs. The company markets Clay as a model that understands page composition and can generate or adapt assets to existing layouts.

Clay ships with fine-tuning kits and instruction datasets tailored for design tasks—things like 'generate a hero image with safe crop margins' or 'suggest 3 typographic hierarchies for this product card.' Mistral emphasized the model's ability to maintain visual consistency across multiple assets for campaigns.

Early adopters in adtech and creative studios noted that Clay's layout-awareness improved automated asset resizing and replaced many manual cropping and retouching steps. Mistral will offer hosted inference and self-hosting options for enterprise customers who need full data control.