This model is more 2.5d than realism compared to the photorealistic models that even 1.5 is capable of nowdays. I have tried to make it a bit better in this regard with this release. with the benefit of refiner models, or passing images into XL to make more realistic with img2img as a second step, it can clearly do much better. None the less, it still looks better than 1.0, and its a bit better at bedwetting and not making pee streams out of diapers. The main appeal here is it can do wet clothes, puddles, bedwetting, as well as diapers, and make them look leaky or used, not perfectly clean like some other models seem limited to. After this update I plan to do an SDXL checkpoint, or maybe SDXL loras.

I have found this model is better with clip skip 1, and is a bit better with sparser negative prompts than my older sample images have, or than I used with my yuzu model. I also have noticed that wet spots on pants are still not as good as they could be, and again I am left with a nod to the newer version of the best lora for this, as I still think it is workable with this model.

Use a VAE obviously but I am not naming a specific one because I cannot tell which one is what I actually use out of the list I think any of the standard ones will do fine. I tend to use negative embeddings but which ones are a matter of personal taste and prompt style. this model seems to be relatively responsive to loras, but anime loras, while they work, will always push the style a bit more towards 2D.