alt-llava

A super-simple utility for generating alt-text for images using the LVM "llava" via Ollama.

Building + Running

First, clone this repo to your local system:

git clone https://github.com/seanmtracey/alt-llava

Building

To build, you will need Go >= v1.22.2. cd into your cloned repo, and then run the following:

go mod tidy # Install dependencies
go build . 

This will create a binary alt-llava.

Running

To run with minimal configuration, you will need to set at least 1 environment variable: IMAGE_URL with a URL that links to an image. This image will be downloaded and stored at ./images with a UUID and a file extension derived from the image's MIME type.

https://ollama.com must be running locally in order to accept and process the requests.

IMAGE_URL=<URL_TO_IMAGE> ./alt-llava

alt-llava will then generate the alt-text for the image and log it to the CLI.

Flags

If you wish, you can write the output of alt-llava to a text file by passing the --out flag with the filepath you wish to write your results to.

./alt-llava --out="./output/results.txt"

Environment Variables.

1. OLLAMA_HOST - The hostname where the Ollama server is running.
2. OLLAMA_PORT - The port the Ollama server is listening on.
3. SILENT_OUTPUT - Only output the alt-text to stdout. Default: false
4. OLLAMA_MODEL - The model that Ollama should run to generate the alt-text. Default: llava
5. OLLAMA_PROTOCOL - http or https. Default: http
6. IMAGE_URL - The image that should be downloaded, and have alt-text generated for.
7. PROMPT_TEXT - The prompt for the LVM that you'd like to use to describe the intended output for the model.