How to process multiple images with a single vision model request

What this does

Passes multiple image files to a vision model in a single request via the Ollama API, enabling multi-image comparison or scene analysis without separate calls. After this guide batch image processing will work in a single command.

Steps

1. Encode multiple images to base64. Converts each image independently.

   IMG1=$(base64 -w0 /path/to/image1.jpg)
   IMG2=$(base64 -w0 /path/to/image2.jpg)
   

   Using -w0 prevents line wrapping, required for valid JSON.

2. Send a multi-image request via the API. Passes both base64 strings in the images array.

   curl -X POST http://localhost:11434/api/generate -d "{\"model\":\"llava\",\"prompt\":\"Compare these two images. What do they have in common?\",\"images\":[\"$IMG1\",\"$IMG2\"]}"
   

   The model receives both images simultaneously and can reference them in a single response.

3. Ask a comparison question. Tests whether the model distinguishes between the images.

   curl -s -X POST http://localhost:11434/api/generate -d "{\"model\":\"llava\",\"prompt\":\"Which image contains more people?\",\"images\":[\"$IMG1\",\"$IMG2\"]}" | jq '.response'
   

   Expected output: A response that correctly identifies each image and answers the question.

• Record the local run evidence. Save the exact command, runtime or package version, model name if applicable, and observed output so the result can be reproduced later.

Verification

curl -s -X POST http://localhost:11434/api/generate -d "{\"model\":\"llava\",\"prompt\":\"List objects visible in both images.\",\"images\":[\"$IMG1\",\"$IMG2\"]}" | jq '.response'
# Expected: Text referencing content from both provided images

Common failures

• images field not an array - Payload must use an array even for a single image; for multiple images, add more elements.
• mismatched image order - Model processes images in array order; swap elements if behavior seems reversed.
• payload too large - Combined base64 for large images causes timeouts; compress images before encoding.
• missing -w0 in base64 - Newlines in base64 break JSON parsing; always include the flag.
• connection refused - Ollama daemon not running; confirm with ps aux | grep ollama.

Related guides

• How to format image prompts correctly for LLaVA models
• How to run LLaVA vision model to analyze images