mmproj (multimodal projector)
The separate file that lets a vision-language model actually see images.
A vision-language model in GGUF is two files: the language model and a projector that maps image features into the language model's embedding space. Load only the first and the model runs perfectly well while silently ignoring every image you give it — which is the usual explanation for a vision model that appears not to work.
The projector is small relative to the model, but it must be loaded explicitly by most runtimes.