>>47949444You probably mean VAE? Which based on that image I think you have working correctly. When it's missing you get faded / washed out colours.
Sorta simplifying here but the model is basically what's in charge of knowing what various concepts look like. Models have their own look/style to them, but probably more relevant to you right now is that they can also understand different concepts.
An anon herehas made/finetuned multiple models with vtuber images. Most models used in these threads are model mixes made with that as a base. If you use a model without that base, you can definitely get some vtubers with careful prompting but many will have a lot of messed up details or will not work entirely.
Picrel is some different models, the last 3 don't "know" who ayame is and you can see they're clearly not as good at making her.
If you want more specific advice, link a catbox or one of your gens so everyone can see the prompts/settings/model/etc. And on the same vein, if you want to learn by copying then use catbox links of images you think look nice and see what prompts/settings/model they are using