Anima Base/P2/P3 prompt:
JoyfulAP, 1boy, furry male, dog boyAnimaPreview1 prompt:
JoyfulAP, 1boy
Recommended samplers:
- Euler Ancestral Scheduler: Simple CFG: 3.0-4.0 Steps: 30-40 Lora strength: 1.0
(er_sde and Euler also work well)
Prompt guide (green - positive, red - negative):
- fat - helps in unusual angles/shots
- closed mouth, parted lips, open mouth - helps produce correct looking teeth/mouth flaps
- claw/claws - helps with producing his canonical hands
- four toe claws - should help with canonical 4 toes, can be used at high weight .e.g (four toe claws:2)
- collared shirt, tank top, shorts, pants, jeans, sleeves rolled up, suspenders, necktie, shirt, shoes, black jacket, belt, jockstrap, vest - canonical clothes tags
Anima Base recommended checkpoint/Loras:
WAI-Anima 1.0
RDBT Lora/Turbo Lora v0.2
Anima Preview recommended checkpoint:
AnimaYume v0.4/WAI-Anima 1.0 [Preview3]
AnimaYume v0.25/v0.3 [Preview2]
AnimaYume v0.1 [Preview1]
Last image in the gallery can be imported into ComfyUI with a basic workflow:
No custom nodes required.
Description
More training on Anima. This time I went with something harder: a fat furry male character.
I wanted to train Joy by kumagayasin on Illustrious for a while but the model always failed to capture all the small details.
Anima once again proves to be more capable than SDXL, however, its not perfect. While it knows furry anatomy it sometimes struggles with smaller details like claws/teeth/fangs etc. I also had to deliberately remove the caption "furry male" and "dog boy" because it was causing issues with ehem private parts. Basically Anima sometimes thinks that fingers with claws are actually penis tips and transfers that knowledge when prompting for nude NSFW images.
I managed to somehow control this by removing all references to furries in my captions and make sure to tag claws.
The Lora is not perfect but I am happy with it considering the difficult of the character + limited dataset.