How to run LLaVA vision model to analyze images
Ollama installed and running, LLaVA model pulled (e.g., llava:latest)
What this does
Pulls the LLaVA vision model via Ollama and runs it against an image file to extract descriptions or answer visual questions. After this guide image analysis will be performed entirely offline using the local vision model.
Steps
Pull the LLaVA model. Downloads the vision model from the Ollama library.
ollama pull llavaExpected output: Progress bars followed by
success.Send an image to LLaVA with a question. Uses the CLI with the image path as the final argument.
ollama run llava "Describe what you see in this image." /path/to/image.jpgExpected output: A textual description of the image content.
Ask a specific follow-up question. Targets particular details in the same session.
ollama run llava "What color is the dominant object?" /path/to/image.jpgExpected output: A concise answer addressing the specific 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
ollama list | grep llava && ollama run llava "Is there a car in this image?" /path/to/image.jpg
# Expected: llava listed in model inventory, followed by a yes/no answer about the image
Common failures
- file not found: Image path is incorrect; use an absolute path and confirm with
ls. - model not found: LLaVA model is not pulled; run
ollama pull llavafirst. - blurry analysis: Image resolution may be too low; try a higher-quality image.
- slow response: LLaVA is compute-intensive; ensure sufficient RAM and ideally a GPU.
- unsupported format: Convert to PNG or JPEG using ImageMagick:
convert input.tiff output.jpg.
Operator checkpoint
Before treating this as solved, write down the local runtime, model or package version, hardware/backend if relevant, and the verification output. This keeps the guide useful as a Will-It-Run style decision instead of a one-off command transcript.