The goal was to make LHC more user friendly, by introducing a base style, while keeping the character knowledge of LHC in tact. As such, no artist tags are needed for a consistent style. The components and method used for merging can be found in the version details.
v2
Using artists is not only not necessary, but tends to produce worse results, so it is in general not recommended.
v1
Using artists is not necessary, but can still be used to some extend in my experience
Recommended settings:
Sampler/scheduler: dpm++ 2m / karras
CFG: 5-6
Steps: 30
Positive Prompt: add "masterpiece, best quality, amazing quality," at the end of your prompt
Negative Prompt: "lowres, (worst quality, bad quality:1.2), bad anatomy, sketch, jpeg artifacts, signature, watermark,"
Creating your own mix
If you are interested in transferring the character knowledge of LHC onto a different base model, you can check out this article https://civarchive.com/articles/11887.
Description
This model is a v-prediction model. Make sure to use a UI that supports v-prediction.
A simple merge of LHC and noob_v_pencil-XL using the perpendicular component strategy used in NoobIEater.
The list of added characters is the same as the one for LHCv0.4, which is https://huggingface.co/Jyrrata/LHC_XL/blob/main/characters/alpha04.txt . Character knowledge seems slightly weaker than the base model, but the result has been much better than just using the Lora-extract.
Recommended settings:
Sampler/scheduler: Euler/sgm uniform
CFG: 3-5
Steps: 25-30