Version 0.3 is here!

New large dataset with cherry-picked aesthetic fox- cat- wolf- girls and others cute characters, beautiful sceneries, landscapes and interiors, natural and hybrid captioning with CogVLM.

Experimental fine tuning merged with some popular mixes. Focused on generating cute characters with animal ears and tails in anime style (not furry, no relations with ~fluff models).

For best result write artist tag (by ..., if used) first and character tag in second. Very strong natural text capabilities, use it in combination with booru tags.

V0.3 comes with general improvement of character knowledge and better ???????. Correct tails (~90% with swimsuits/underwear/..., ~80% with general clothes, ~40% with interactions like "lifted by tail"), correct ears (right angles in side/back views, no more human ears spamming). As an added bonus - fine performance in NSFW, see examples.

768 base resolution, features v-pred with all its benefits (do not forget to download YAML config and put it in the same folder with model!).

Subversions:

• Main - pick this first, base version.

• Juicy - produce high contrast and attractive pictures with epic sceneries, but sligtly less coherence

• Alt - alternative merge, sometimes better then main in bodies, but simpler backgrounds. Best for upscale.

• Soft - generates softer images

If you are getting strange colour noise - download model config and place it in folder with safetensors file.