Lollipop is an extremely good 2.5D base model. She can easily handle a variety of styles, including but not limited to secondary, illustration, anime characters, etc. Lollipop is based my Flyingsnow model. It is a branch model based on my Flyingsnow model training + mix, Lollipop perfectly inherits the advantages of the Flyingsnow model, and on top of that, mix the Ghostmix middle layer and the 8th-11th output layer with 0.5 weight. This gives Lollipop the perfect inheritance of the strengths of Flyingsnow and Ghostmix, with the use of plugins to fine-tune the colors of the model.

Anyway, I've always thought that a good large model should be versatile and not lean too much towards a certain style that is too fixed, as this can be left to LoRA. Therefore, when doing the final testing of this model, I focused more on its colors, composition, lines, backgrounds, characters, and the model's response to unstyled cue words such as Anime, Photographic, Digital Art, Fantasy Art, Neon Punk, Low Poly, Line Art, 3D Model Pixel Art, etc.

Generation Parameters：

Sampling Steps ：25-30

Sampler method：Restart (regular images), DDIM (images with high quality textures. It will be slower than other samplers, but it doesn't matter if you run it online), PLMS (generates images with oil painting style)

CFG scale：6-9

Hires. fix Parameters：

Step：15-20

Denoising：0.1-0.3

Upscaler：R-ESRGAN 4x+ Anime6B (Anime , Illustration), Nearest (try watercolor