How to use audio and vision modalities in llama.cpp?

How to use audio and vision modalities in llama.cpp with Gemma4 12B it? I’m on release b9494, but when I run llama-cli it shows “modalities: text” only, and crashes if I try to add an image.