ANIMA is fantastic and robust but has some weaknesses, such as lackluster tongue fetish ability IMO. This LoRA aims to mitigate that by adding more detail to the face, mouth, tongue, uvula, and to a small extent, ahegao (though I plan on training a LoRA specifically for ahegao).

Trained to 4000 steps @ 0.000025 LR using ~175 human-curated photos and captioned with danbooru-style tags using Grok. No trigger word necessary, though I never used trigger words for any LoRA past the SD 1.5 days anyway.

Based on my prompting style + negative prompt (see ComfyUI workflow in my example images), 0.5 - 0.75 strength appears to be the sweet spot. YMMV depending on your tastes, as higher strengths rapidly shift toward stronger realism.