The update of this version is to create a good-looking anime-style face. This version leans a bit towards 2.5D, focusing on training faces in the rainbow style. The color representation remains excellent. Now, you can create unique images without needing Lora.

The sample size I used for testing is 768*1024*2, and the sampling method is not very particular, just use what you are accustomed to. For upscaling, I generally use: R-ESRGAN 4x+ Anime6B, and the testing was also done using this. Sampling steps: 25-30.