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.<h3 id="v2-l0ttoauqo">v2</h3>Using artists is not only not necessary, but tends to produce worse results, so it is in general not recommended.<h3 id="v1-b3cjvmejn">v1</h3>Using artists is not necessary, but can still be used to some extend in my experienceRecommended settings:Sampler/scheduler: dpm++ 2m / karrasCFG: 5-6Steps: 30Positive Prompt: add "masterpiece, best quality, amazing quality," at the end of your promptNegative Prompt: "lowres, (worst quality, bad quality:1.2), bad anatomy, sketch, jpeg artifacts, signature, watermark,"<h3 id="creating-your-own-mix-lwsg2q7es">Creating your own mix</h3>If you are interested in transferring the character knowledge of LHC onto a different base model, you can check out this article <a target="_blank" rel="ugc" href="https://civarchive.com/articles/11887">https://civarchive.com/articles/11887</a>.

Particle-Accelerated Mix (Illustrious)

About this versionA simple merge of <a target="_blank" rel="ugc" href="https://civitai.com/models/940235">LHC</a> and <a target="_blank" rel="ugc" href="https://civitai.com/models/926443?modelVersionId=1159421">NTRmix-XII</a> using the perpendicular component strategy used in <a target="_blank" rel="ugc" href="https://civitai.com/models/879640">NoobIEater</a>. Additionally, I merged the <a target="_blank" rel="ugc" href="https://civitai.com/models/929652/audelete-or-artist-style-or-illustrious-xl">Au_(d_elete)</a> and <a target="_blank" rel="ugc" href="https://civitai.com/models/834822/nyalia-illustrious">nyalia</a> Loras at 0.4 and 0.5 strength respectively.The list of added characters is the same as the one for <a rel="ugc" href="https://huggingface.co/Jyrrata/LHC_XL/blob/main/alpha_v05/vtubers_valpha05.txt">LHCv0.5</a> . Character knowledge seems slightly weaker than the base model, but the result has been much better than trying to do a Lora extract.Recommended settings:Sampler/scheduler: dpm++ 2m / karrasCFG: 6Steps: 30Positive Prompt: add "masterpiece, best quality, amazing quality," at the end of your promptNegative Prompt: "lowres, (worst quality, bad quality:1.2), bad anatomy, sketch, jpeg artifacts, signature, watermark,"

Files with SHA256: dcd0cc1cecf49c879c545a4dc804d756d3ad69c61de103abae77470936f0ddd9