Type | |
Stats | 264 |
Reviews | (46) |
Published | Dec 24, 2024 |
Base Model | |
Hash | AutoV2 10F129C4FB |
Default rules as for any other Illustrious model:
"explicit" keyword for 18+ content, "sensitive" for questionable, and "general" for default SFW content.
Use quality tags like "good qulity, best quality, etc." at the end of your prompt.
The model supports this LCM lora partially (give it smaller weights if you see strange artifacts): https://civitai.com/models/216190/lcmandturbomix-lora-only-12mb-8-step-sampling-effect-is-superior-to-using-lcm-or-turbo-alone
So you can generate images in around 12 steps by using this lora, euler_a_cfg++ sampler and sgm_uniform (faster) scheduler. You can use normal scheduler, which provides better quality with ~15 steps.
It's mix of 3 models:
https://civitai.com/models/926443?modelVersionId=1061268 - NTRMix 4.0 (for some part of it's style)
https://civitai.com/models/811067?modelVersionId=1015877 - can't remember exact version. The model was used to get less annoying traits from loras that was used in NTRMix + small quality improvements
https://civitai.com/models/950531?modelVersionId=1064213 - RouWei. Good but unstable model. It improves knowledge of some rare tags. Also it gives more diversity in results.