Can I use llm to describe/explain images?

[glaucocustodio]

gpt-4-vision-preview is not listed on https://llm.datasette.io/en/stable/openai-models.html

Does anyone know if we can talk about images (eg asking questions) using any of the plugins? Is it even supported by llm itself?

[connabyss]

Yes, since llm 0.17 (Oct 2024). Images are "attachments", passed with -a (a file path or a URL):

llm -m gpt-4o-mini "describe this image" -a photo.jpg
llm -m gpt-4o-mini "what's different?" -a one.png -a two.png

To see which file types a model accepts: llm models --options -m gpt-4o-mini (look under "Attachment types"; on 0.36 that's JPEG, PNG, GIF, WEBP and PDF).

Docs: https://llm.datasette.io/en/stable/usage.html#attachments

(Checked with help from Claude against llm 0.36.)