I was aiming for more of 2.5d/semi-realistic look with this one. It's a continuation of Shidoshi Fun Mix but I blended back to the NAI model (just a bit). The results were pretty good.

No VAE burned in. And if anyone wants the full float for their own mixing purposes, I will upload it (10.7Gb).

Euler samplers don't seem to work so great with this one. I use DPM++ 2M Karras (1st pass) and DPM++ 2M SDE Karras for the 2nd pass/ hires fix.

This model likes rendering everything and everyone nude.

I still recommend : NEGS: (worst quality:2), (low quality:1.8), (normal quality:1.6),

Please check the grids I made with varying quality chants. I included "Masterpiece" in the positive prompt (y-axis) as well. You be the judge.

Most of my samples don't use embeddings. If the image has too many fingers, a simple "too many fingers" in the negative prompt usually fixes it (and "fewer fingers" for not enough fingers, "too many hands", "too many feet", etc...).

I will use the BadDream + UnrealisticDream (Negative Embeddings) though.

I like positive and negative feedback! And post some samples.