Ollama Vision Models - Image Analysis with LLaVA

ShopMax India manages thousands of product images for its electronics catalogue. Using LLaVA (Large Language and Vision Assistant) through Ollama, you can automatically generate product descriptions and detect image quality issues without sending data to cloud APIs.

LLaVA is a multimodal model that accepts both text prompts and images. In Ollama, you pass images as base64-encoded strings in the messages array. The model returns text descriptions, detects objects, and answers questions about image content. Pull the model first with: ollama pull llava

The below example shows how ShopMax India analyses a product image to generate a catalogue description using Ollama with LLaVA.

It gives the following output,

The image shows a large flat-screen television with a slim bezel design.
Key features visible:
- Ultra-thin profile suitable for wall mounting
- Multiple ports on the rear panel (HDMI, USB)
- Stand with cable management slot
Recommended for: Home theatre setups in Mumbai and Bangalore showrooms.

Image resolution affects accuracy - use images of at least 512x512 pixels for reliable results. For ShopMax India batch processing, use a queue to avoid memory pressure when loading multiple large images. LLaVA works best with clear, well-lit product photos and specific prompts that ask for structured output.