How do I send an image to an Ollama model?

Include an images array of Base64-encoded strings in the request, alongside your prompt . With the ollama Python client you can pass file paths directly and it encodes them for you.

Which Ollama models can process images?

Vision models such as llava , llama3.2-vision , and moondream . A text-only model has no vision architecture, so it cannot interpret an image even if you send one.

Can I use Ollama vision models for UI testing?

Yes. Send a screenshot and ask the model to list elements, describe layout, or spot issues. It works well for local, automated UI and accessibility checks without any cloud service.

Why does the model ignore my image?

Most likely you are using a text-only model. Switch to a vision model like llava . Also confirm the image is valid Base64 and sits in the images field, not the prompt text.

Build / AI / Ollama

Vision Models With Ollama

Send images to local vision models like llava for captioning, UI analysis, and reading text.

TL;DR

Send images as Base64 strings to a vision model.
Use a vision model like llava or llama3.2-vision.
Pass an images array alongside your text prompt.

Jump to section

Vision Basics
Encode Images
cURL Example
Use Cases
Tips
Warnings
In Practice
FAQ

Vision Basics

Pull A Vision Model

Download a model with a vision architecture.

ollama pull llava

images Field

Attach Base64 images to the request.

# body: { model, prompt, images: [b64] }

Ask About It

Pair the image with a specific question.

"prompt": "Describe this screenshot"

Encode Images

Base64 Encode

Read a file and encode it to Base64.

import base64
b64 = base64.b64encode(data).decode()

Python Client

The ollama client takes file paths directly.

ollama.chat(model="llava",
  messages=[{"role": "user", ...}])

images In Message

Add an images list to the message.

"images": ["screenshot.png"]

cURL Example

generate With Image

Send a Base64 image to /api/generate.

curl http://localhost:11434/api/generate -d '{
  "model": "llava",
  "prompt": "What is in this image?",
  "images": ["<base64>"],
  "stream": false
}'

Base64 String

The images array holds encoded strings.

"images": ["iVBORw0KGgo..."]

stream false

Get one response for scripting.

"stream": false

Use Cases

UI Analysis

Describe or audit a UI screenshot automatically.

"prompt": "List the buttons in this UI"

Captioning

Generate alt text for images in bulk.

"prompt": "Write concise alt text"

OCR-Like Reading

Read visible text from an image.

"prompt": "Transcribe the text shown"

Tips

Use the ollama Python client, which accepts image file paths in images and Base64-encodes them for you automatically.
Ask a specific question about the image, such as extracting text or listing UI elements, to get a focused, useful answer.

Warnings

A text-only model ignores the images field; the model must have a vision architecture like llava to see pictures.
Large images inflate token use and slow inference; resize or compress them before Base64-encoding for faster responses.

In Practice

Use the ollama Python client to send a screenshot to a vision model and get a caption back.

The ollama client reads the image path and Base64-encodes it.
llava has a vision architecture, so it can see the image.
The content question focuses the model on what you need.
The reply is a text caption describing the screenshot.

import ollama

resp = ollama.chat(
    model="llava",
    messages=[{
        "role": "user",
        "content": "Describe this screenshot",
        "images": ["screenshot.png"],
    }],
)
print(resp.message.content)