Meta releases LlamaVision 2: on-device multimodal for creative prototyping

Tech · 4 min read

Meta announced LlamaVision 2, an updated multimodal model engineered for on-device creative tasks like image-to-layout conversion, style transfer, and palette extraction. The new model reduces memory footprint while maintaining strong visual reasoning capabilities that design teams can use in prototyping flows.

A key differentiator for LlamaVision 2 is its offline-first approach: designers can run inference locally on supported mobile and laptop hardware, avoiding cloud uploads for sensitive material. Meta also provided developer libraries for exporting model outputs as design assets (SVGs and layered PSDs) and hooks for common design tools.

Third-party plugin developers are already adapting LlamaVision 2 into photo-editing apps and rapid prototyping tools, highlighting the appeal of low-latency, privacy-preserving workflows. Meta plans to open-source parts of the runtime for research and community contributions later this year.