A LoRA that teaches/improves a model to show licking and kissing of a woman's vulva
Trained on 21 close-ups and 22 medium shots of real women performing cunnilingus with the receiver on her back. Mostly camera friendly side angles but also a few POV angles too. Manually written captions as auto-captioning had no idea how to describe this action. This has been a labour of love.
Versions:
v1.0_H3 - Minimax H3 version. Currently limited to depicting Receiving Girl on Back and Giving Girl on all fours/stomach
(v1.1_H3 Coming soon) - Minimax H3 version with Giving Girl Eating Out Receiving Girl from behind.
Minimax H3 Version
Trained on my RTX5090 using ai-toolkit. Steps 0-2500 trained on 21 close-ups to capture the detail of the tongue and the distension of the labia. Steps 2500-5500 trained on both close-ups and medium shots to allow the model to understand the body positioning better and prevent zooming or cutting into close-ups during inference.
Usage Notes:
For best results I highly recommend you use TenStrip's 10ErosMax for H3. This checkpoint already delivers good results for this activity, but is very sensitive to camera angle, and sometimes distorts female anatomy too much. This LoRA fixes those issues. My video showcase demonstrates the before/after effect of this LoRA on 10Eros.
You can vary the strength from 0.5 to 1.0 to adjust the speed of licking. Experiment with what works best.
Works with both FL2VA and Ref2VA models (10Eros is a hybrid checkpoint). As always with H3, tight and detailed prompt structure is key when dealing with two-person shots. See my showcase videos for examples.
Use my Lesbian Posing LoRA to help get your characters into position on your seed image!
A big THANK YOU to all of you who donated their buzz to the Early Access unlock in less than 18 hours! If you like my work and wish to support me, then please consider signing up for a paid sub to my riotmodels page where you get to see the full scenes of the Seven Sisters doing their thing.
Description
v1.2 is the first production release for Minimax H3
FAQ
Comments (5)
為何我覺得no lora比較厲害,是不是標記打錯影片🤦♀️
In some cases it is marginal. In other cases, especially at the POV and other angles, the LoRA really fixes anomalies.
Sadly this lora gives visual errors at 15 steps with 0.5 resolution for 15-20 second clip on both FL2V, REF2VA
Thanks for spotting this. I will investigate.
@Generation3dX No prob, hopefully an updated version is released. I have not tested your other loras YET. So idk if they all suffer from this issue. I will keep you posted.
