This is an NSFW LoRA intended to produce better futanari anatomy, with a bias towards realistic/3d (but capable in other styles), now for Pony Diffusion V6!

Usage:

• The keyword futacock is absolutely necessary. I prefer to use it as an adjective alongside whatever descriptors you prefer, as well as replacement for tags like "woman, 1girl" etc.

  • Example: score_9, score_8_up, score_7_up, source_comic, rating_explicit, futacock, brown hair, huge breasts, standing outside

• You will notice i always include the score tags, the rating tag (explicit or questionable). Futacock should ideally be the next tag after those, before describing the character or scene

• Since tags like futanari, huge penis, etc. are somewhat supported by Pony Diffusion nativley, using these tags alongside this lora is sort of redundant but not harmful at all.

Some quirks and issues you might encounter:

• It's somewhat competent with established characters within Pony Diffusion.

• Variable results for different styles (comic, anime, etc), just try generating again, or turning the strength down.

• Your subject will occasionally have strangely smooth/plastic looking skin.

  • To fix this, tinker with your tags surrounding your lighting and environment. It's usually a prompting issue.

    • You can put things like "beautiful", "photography", "realistic" at the start of your prompt and when describing your subject, include "attractive" as a descriptor

    • Start with "glossy skin" in your negative prompt, but...

      • If you get really desperate, try putting "barbie doll, plastic, fake" in your negative prompt. This might make your subjects a touch less attractive though.

• The tag "huge breasts" is great in my opinion, but could be very heavy-handed for others. Use as you see fit.

Limits & Future Improvements:

• This LoRA was trained around "realism" and 3D renders, so I make no guarantee how well it interacts with other style tags (like 2D or anime) or LoRAs. I hope to improve this in the future though.

• Pose variety can be limited.

  • Most of the training data involves standing, sitting, or squatting so that's what you'll get in your generations.

  • Try experimenting with controlnet if you really want more variety.

• Your subject will almost always be nude by default.

  • The majority of the training data didn't involve clothes, so the LoRA might struggle to do full outfits. Forcing clothing will probably affect the anatomy of the penis as well.

  • For safekeeping, feel free to include the tag "nude" even though it's not super necessary

  • It can do things like jewelry or thigh-highs to some success, though.