moawxx - Female Pleasure Moans + Body Writhing (MiniMax H3)
Trigger word: moawxx
This LoRA is trained to add realistic female pleasure moans and corresponding body movements (writhing, arching, trembling, hip movements) for MiniMax H3.
Recommended settings:
Strength: 0.6 – 0.85 (1.0 often degrades image quality)
Best used with prompts that describe both sound and motion
Example prompts:
text
moawxx, soft sensual female moan, body gently writhing and arching
moawxx, intense high-pitched female orgasm moan, hips bucking, legs trembling
moawxx, deep throaty pleasure moan, woman squirming and twistingNotes:
Trained at 256 resolution, so pure image quality may slightly decrease when the LoRA is applied strongly.
Focus of the training was on audio + motion synchronization, not on visual details.
Works best when the prompt mentions both the sound and the body movement.
Versions:
I recommend testing different epochs. Earlier ones usually preserve image quality better, while later ones have stronger sound and movement effect.
Feel free to post your generations.
Description
FAQ
Comments (16)
Guys, stop realeasing loras that looks that good when im at work! I cant leave early everyday!
Sorry :D
Greattt work!! Please make a version for wet sucking sounds, licking sounds and gagging sounds. We need LoRAs like that it would help A LOT! Thanks!
Thanks a lot!
Glad you like it.
Good idea, I’ll consider making one in the future when I have a solid dataset for it.
epic cool lora . hats off
Many thanks!
Does this work with the audio reference? In other words with the tone of the voice of the moan change if a reference is included?
I'm curious about this as well.
The LoRA mainly adds the trained moan style and body movement when you use the trigger moawxx.
Regarding Audio Reference — it can still influence the base voice tone a little.
Best to test it yourself with different strengths and reference audios, results can vary.
Really nice results!! What trainer did you use for? Also, did you mix images and videos together? Would love to know—thanks!
Thanks!
Trainer: ostris/ai-toolkit (the one with MiniMax H3 support).
I used only videos, no images mixed in. Dataset was 70 short clips (6–10 sec each) focused on the sound + body movement.
сколько видео вы использовали для тренировки и какой learning rate,rank?
Dataset: 70 видео (6–10 sec)
Learning Rate: 1e-4 (0.0001)
Rank: 16
Resolution: 256 Steps: 3000 Audio Loss Multiplier: 2
Как то так)
@xpero спасибо,приятно что знаете русский,если не трудно можете дать совет новичку,пока что мои лоры либо выходят переобученной кашей,либо обучаются не тому что надо(
that is an absolute gamechanger lora for H3 - thank you so much!
its i2v or t2v?