/vt/ - Virtual Youtubers

>>37968803
This is all trial and error right now to learn how many vectors an embedding needs. There is no way to know how many you will need beforehand. It all depends on the complexity of the object/subject. Properly explaining how vectors/tokens works means explaining how stable diffusion as a whole works and someone else can do that better than me.

You want just "enough" vectors, less could be considered better. With more vectors you are more likely to pick up random noise in while training and create artifacts.
Picrel is an exaggerated example but thats the noise of too many vectors.