liquidai

Liquid AI's open-source edge multimodal models: Convert images directly to JSON, running on just 1.6B parameters.

Liquid AI has launched two compact multimodal models, LFM2.5-VL-1.6B-Extract and 450M-Extract, specifically optimized for extracting structured data from images. These models can convert images to JSON format data directly on-device based on a specified field list, all without needing a GPU.

Why it matters: Edge AI is evolving from 'seeing and describing images' to 'seeing and extracting structured data', significantly reducing deployment costs and privacy risks in scenarios like AI document processing and industrial inspection.

#LiquidAI #AI #多模态 #open-source

Liquid AI drops a local data de-identification tool, ShieldFlow, that runs offline without needing a GPU.

Liquid AI has rolled out ShieldFlow, a local privacy data de-identification tool that can operate offline on your device, automatically spotting and masking sensitive data. It doesn't require a GPU and has a very low memory footprint, allowing it to run in real-time on most standard PCs. It was showcased at Microsoft Build 2026 and Taipei Computex 2026.

Why it matters: One of the biggest hurdles for enterprise AI applications is data privacy. ShieldFlow tackles the trust issue of "AI in the cloud = data breaches" by running purely locally.

#LiquidAI #隐私 #AIsecurity