None of this requires a multimodal LLM.

You can do this with traditional CNNs. For place, use Geospy or another geospatial model.