It would be inefficient if git only had access to the binary model files instead of each image in it. This way, if one image gets updated, an entire new model package doesn't need to be stored in git's objects.
Model files are zip archives with a "model.toml" file at the root describing the model and all its textures.