use Ollama to host a vision language model, by default using the llava vlm. it can grab a frame from opencv and describe it based on the system prompt.

[kwatters]

See if we can give a more meaningful understanding of what the inmoov sees.