ModelsLab/[SD1.5] OpenSolera - Anime Mix - OpenSolera-α5 [Serif]

Internal name: OpenSolera-a5-Serif-Complete.safetensors

Both a hotfix to Noto to make it less blurry during prompting and some changes to the model to align it closer to Alpha 3.

This version's codename, Serif, came from the serifs in many popular fonts like Times New Roman. The fact that the Noto font family also has a serif typeface makes this codename better imo.

Changelog:

• More mature character type: During my merges, something happened that makes the body size more anatomically accurate, which makes some characters more mature than they should be...

• Sharper images and clearer lines, obviously

• Weird but semi-consistently good hands: About the same, if not slightly better than Noto, which is above average at generating hands in my experience

• Prompt adherence is slightly worse at medium-high CFG value. Landscape images are kinda fucked

• About the same level of vibrancy as post-fix Alpha 4 and Alpha 3 (with "vibrant colors" in positive)

• Anthro tag still works, and so do the monster girls tags (most of the time)

• Styles seem to be much more required than before

• NEGATIVE EMBEDDINGS ARE STILL BAD ON THIS MODEL. Still don't have a way to fix it, nor do I want to fix it.