It would be a great option to allow users to train their own models using the currently available ones as a base. For instance, a Film Restoration Artist may have access to both 4K and corresponding SD tape sources of the same movies. They could use these versions to train their own models and share them with the community.
