An experimental hypernetwork trained on 100+ different anime characters. Do we actually need to create specific networks for each and every character or could we work on improving our base models instead?

Since the list is rather large, see each version for which characters were added. Later versions include all characters of the previous versions as well.

This kind of model certainly requires a lot more training than just doing a single character and some characters might not be as well finished as others. However adding more characters does not negatively affect previously trained characters, indicating that the network have no issues representing them and that it is simply an issue of training it enough.