v2.0 version application introduction & input suggestions
This is an adjusted version based on the previous model, selected after testing multiple language models.
JoyCaption2Split language model. Before the image enters the latent space, I first performed scaling, which is a very critical step because some images with insufficient pixels will lose a lot of details. However, after shrinking and then enlarging
it can complete the pixels. These steps are very critical) and can significantly improve the image quality without changing the original image size.
Input suggestions
A redraw amplitude of around 0.3 is more suitable. For portraits, facial changes are still relatively difficult to control, so you can reduce the redraw amplitude to around 0.2 and lower the CFG weight.