OpenMosaic launches MosaicX: a 1T multimodal design assistant for product teams

AI · 6 min read

OpenMosaic positioned MosaicX as a purpose-built assistant for designers, combining image, layout and text modalities in a single model. The company says the 1T parameter architecture was trained on curated UI kits, accessibility patterns, and real-world product screenshots to reduce hallucinations in layout and copy generation.

MosaicX offers built-in brand conditioning: design systems can be uploaded as token sets and the model will generate variants that obey spacing, color, and typographic rules. OpenMosaic is also highlighting latency improvements for interactive use, claiming sub-second vector sketch refinement in a cloud-hosted tier and <200ms inference for quantized on-prem endpoints.

Enterprise features include audit logs, fine-tuning on private design repos, and a legal-safe data pipeline that blocks third-party scraping. OpenMosaic is launching a graded release with free sandbox access, paid cloud workspaces, and an on-prem appliance for larger teams, aiming to position MosaicX as a central design co-pilot.