For more LoRAs and updates, Join my Discord:
https://discord.gg/ZVWVhT43GW
You can deploy MiniMax on Runpod with my template:
👉 https://get.runpod.io/minimax-template
👉 Deployment tutorial
This is a very experimental LoRA trained on a relatively small dataset (47 videos) and actually works.
It's really not the best, but does the trick.
V1 is a big improvement with better fingering and squirting texture, still not the best.
Probably needs a larger dataset with more examples.
RECOMMENDED SETTINGS
Sampler - Euler
Scheduler - Simple
Steps - 12
Shift - 6
Turbo LoRA - Strength 0.5
Use the HMPussy LoRA at strength 0.4-0.8 with this LoRA for best results.
Description
Same dataset as V1.0 different training settings, better fingering, still needs the vagina LoRA
FAQ
Comments (14)
Just what I needed! Thank you🙏❤️
I wonder, if the same dataset can be used to make LTX 2.3 Lora?
Yes but I won't train it since I don't like LTX
@HearmemanAI It looks like I'll have to learn & switch to MiniMax then ;)
Brilliant concept, appreciate that. Could you also train squirt-only (no fingering) lora as well? The liquid is not so realistic.
Piss. You're talking about piss.
@brnfd24434343d Squirting is fluid expelled from the vagina during orgasm, and it can include urine components—but it isn’t always just ordinary bathroom pee. Some people’s squirting also involves secretions from the Skene’s (paraurethral) glands, so the fluid can be more than only urine.
It can be just pee though lol.
@makiaeveli it's pee
Dis shit dope
Can it be used on the ref model?
normally flv models can get something on the ref model, but rarely the other way around. You could try it at 0.8 str and increase or decrease
fl2va model has superior quality and less issues when it comes to using loras. It also functions the exact same as the ref model when plugged into the ref node.
@SweetAyanna It does not function exactly the same, hence why it is a separate model. R2VA has been trained on additional direction and reference themes.
@jm112368767 I should have rephrased. On paper the ref model might be different. The reality is if you plug fl2va into the ref node, the output on average will be better across the board. Plus loras that were trained with the ref model work without problem on fl2va, but not vice versa. fl2va has no issues with references and it maintains identity as well as the ref model does, so yes. While it seems different. Based on what people (and I) have generated the fl2va just feels like a superior model overall.