VERSION 1.5:
The dataset and training parameters have been improved, which has significantly improved quality.. Unfortunately me have to intentionally overtrain her to get her face in the right style, which sometimes leads to pov arms, extra heads, and multiple views, but all of this is easily fixed with negative prompts. With the right approach you can get great results in most cases.
If you want to get results close to mine, I advise you to use the "Based64" model. Weight 1.0 is OK.