paligemma

PaliGemma is a vision-language multimodal model introduced by Google as part of its Gemma umbrella of lightweight open models, first made available to developers on May 14, 2024. It was positioned as the first member of the Gemma family focused specifically on translating visual information into written language, sitting alongside sibling variants such as CodeGemma and RecurrentGemma. The model's purpose is to bridge image understanding and natural language output in a form factor that can run efficiently without demanding heavy compute resources.

In practice, PaliGemma targets three core tasks: image captioning, visual question answering, and image retrieval, making it useful for applications that need to describe or reason about visual content in text form. Its small-language-model footprint is designed for resource-constrained environments like smartphones, IoT devices, and personal computers, opening up use cases such as accessibility tools for visually impaired users, richer in-app content generation, and enhanced image-based search. Google subsequently followed this release with a PaliGemma 2 generation in late 2024, indicating an active development trajectory for the vision-language line.