build a single combined token for Textual Inversions

[ssube]

While the current code does support textual inversions, ones with more than one layer/vector will be split into multiple tokens. There are some benefits to that, since you can weight/omit individual tokens, but some of the embeddings available from Civitai have more than 10 layers - a few even have 70-75 or more. Providing that many tokens in the winter-0, winter-1, winter-2, winter-3, ... form takes a lot of time and space in the prompt.

Add a single token that will either expand into all of the others or invoke the layers correctly without expansion.