llava

Visual question answering integrates a vision encoder with an LLM to answer questions about objects and scenes in images. You provide an image and ask questions, and the model analyzes the visual content to respond.

Multi-turn image chat lets you hold conversations about visual content across multiple turns. You provide an image and continue asking follow-up questions, and the model maintains context to discuss the image naturally.

Multimodal document analysis processes documents containing images or figures by combining a vision encoder with an LLM. The model understands both text and visual elements, enabling you to analyze and discuss document content in natural language.

Image-based conversations require a multimodal backbone pairing a vision encoder with an LLM such as Vicuna or LLaMA. You need downloadable weights for both components to support visual instruction following and multi-turn prompts.

Visual instruction following performs tasks described in visual prompts to generate captions and scene descriptions. The model follows user instructions to describe images, answer questions about visual details, and analyze scene content.

Multimodal reasoning works across desktop and cloud environments. The vision-language integration supports multi-turn prompts and image inputs in both contexts, enabling flexible deployment for image chat and visual question answering.