(It's been a long-term exploration to enhance similarity and so on.)Training images are old inevitably. They are processed by Topaz to get higher definition and rid of mosaic. Yet then LoRA training steps must be restrained in case of blur after hires.fix.~~Both version suggested. Suggested weighted around 0.66 on 020. Notice "About this version" info.💁‍♀️Refer to sample images and notice their unique negative prompts and sampling method of each checkpoint.<pre><code>(Welcome to feedback, review &amp; give me more stars and red hearts!)</code></pre>~~An actress active in 1980~1990s.Be kind to these lovely.<h3>Legal &amp; Risk</h3>It is prohibited to use this model for commercial purposes and any scenarios of illegal acts and purposes.

Nina Li Chi (Li Zhi)

Suggested weighted around 0.66.Refer to sample images and notice their unique negative prompts and sampling method of each checkpoint.e.g. KanpiroMix is good at DPM++ 2M Karras,but not DPM++ 2S a Karras.

Files with SHA256: cb77da7b1fe70cbfb8c4003799322ed05350bad91236e39949b8db37522bbd13