READ VERSION NOTES IF YOU WANT TO USE V3!
Version 2 works great with Lucid's Mix. Images are made with strength 1 to 0.8 and 512x704 ratio with hires fix Latent or R-ESRGAN 4x+ Anime6B upscale of 1,5 and about 0,6 Denoising strength. Some are also upscaled with img2img.
VAE I use is kl-f8-anime2.
Tested with various amount of other LoRAs, works fine with most Anime and style LoRAs.
Share some of your creations in the comments, I want to see what you can create with it.
Description
This version works well on LawlasMix_prunde_fp16. I personally used for these images a merge of LawlasMix_prunde_fp16 with AbyssOrangeMix2_Hard with 0.2 weighted sum.
Mainly used Euler a sampler with 20 steps. 7 CFG scale. 512x704 resolution with hiresfix upscale of 1.5, 20 hires steps and Latent with 0.5 denoising or R-ESRGAN 4x+ Anime6B upscaler. LoRA weight was mainly on 0.8. Clip skip 2 is recommended.
Some basic prompts are: 1girl, body fur, solo, raccoon ears, two-tone hair, short hair, (raccoon tail:1.0), (body fur, dark body, fluffy fur, brown fur, intricate fur:1.1), multicolored eyes
If you want fur and generally more furry version of Michiru add: (detailed fluffy fur), (body fur, dark body, fluffy fur, brown fur, intricate fur:1.1)
I also like to use these: (high quality), (digital painting), (8k)
Negatives:
I used these as basic negative prompts on most images: (worst quality, low quality:1.4), (bad_prompt_version2:0.7), (bad-artist:0.8), (boring_e621:1.0)
You might need to install some of the embeddings.
If you want to have Michiru with skin add "body fur, chest fur" to negative and remove any fur tags from positive.
Eyes are kinda blurry sometimes. I inpainted the eyes with only masked and denoising at 0.3 to make them look better(didnt do that on the sample images here)
Tail is missing sometimes or in a wrong color. Editing it blue and using inpaint is an easy fix if that happens.