Decided to split my base model into realistic and anime styles because trying to both will worsen the quality of both styles.This model leans on anime to cgi styles.<h3 id="heading-56">Recommended Parameters</h3>Sampler: DPM++2M Karras, DPM++ SDE Karras, Euler a, DDIMSampling Steps: 20-30CFG: 7-10Clip: 2Negative Embeddings: <a target="_blank" rel="ugc" href="https://civarchive.com/models/23178">bad_pictures</a>

Versamix-Anime

Much better prompt following over the first version. The model can now "wink with one eye closed" and produces better image quality with just (worst quality, low quality:1.4) negative prompt.

Files with SHA256: a590f14dc617cd6b3e8d6f3a8cb189d28cec902aae7949f03fd643cdbc222e32