This lora was trained to enhance your character's thumb going into someone's anus. This lora also teaches the MiniMax-H3 model the action of putting objects into a anus.
Trigger word: thum1n8utt
trained prompts: the man inserts his right thumb in her anus causing the woman to moan with pleasure.
the man inserts his right thumb in her anus.
the man inserts his left thumb in her anus.
the man uses his right thumb to rub around her anus in a circular motion then inserts his right thumb in her anus.
the man uses his right thumb to rub around her anus in a circular motion then inserts his right thumb in her anus causing the woman to moan with pleasure.
Be creative with your prompting, interesting prompts will lead to interesting results. You can see how the lora behaves in the videos I posted.
This lora was trained for I2V. No audio was trained. If your character is positioned where the vagina is above the anus, the thumb will most likely be inserted into the vagina instead. It kinda works in R2V and T2V but it does not look too good. R2V might be decent if you use a image of your character's anus already well exposed.
If your enjoy my work go ahead and support me at
Ko-fi: https://ko-fi.com/captjay25895
You can find my loras not allowed on this website at my Hugging Face - https://huggingface.co/CaptJay178
Description
FAQ
Comments (17)
Thank you so much for creating this!
I've started working on a finger in anus h3 lora, but results haven't been great. I think it's because it requires capturing both motion + detail and on my local setup, I can't run )still in the process of experimenting) high resolution clips to capture both.
What does your dataset look like? What hardware and software did you train on?
I locally trained with a RTX 5090 and trained at 512 resolution in ai toolkit. The dataset was thumbs being inserted to anuses.
@captjay178 Thanks! How many clips and what duration were the clips?
@sugarfreewafers like 10 or 11, all 5 seconds
@captjay178 Thanks!
@captjay178 Thank you so much for your information since I have a 5090 I wanted to train loras I was wondering about how long did it take to train with 11 videos and I saw the AI toolkit video on YouTube and in that one he also used images in his dataset. Many thanks for your lora and the info
@Hzoid it took around 8 hours training at 512 resolution. There are many factors on how long training will take. I tried training at higher resolutions using video clips encoded in h.265 and linear PCM codecs and after 8 hours the steps were only around 100/3000...Your electricity bill will increase pretty fast lol.
@captjay178 Also curious, do you let it run for the full 8h, or is there a pause button, or only train on the first 3 videos and continue training afterwards (pause,....resume continues 4...5....6.)
@right144 ai toolkit does have a pause button, but I just let the whole thing train while I go to sleep.
The link to click to copy the trigger word is being flagged as a ClickFix attack.
In the model details? That would be a Civitai issue then, authors can only set regular text - not arbitrary code.
What do you mean exactly?
just copy the trigger from the main description, maybe Civitai is up to something...
Great LoRa!! Thank you!
MiniMax was already curiously good at doing this concept... this does seem to guide it a bit even with less careful prompting.
Definitely makes a difference, at gaping a-holes, AI was constantly gaping her peach while with this lora it gapes her a-hole nice and large.
If the strength is larger than 0.5, it creates severe artifacts elsewhere. But it does not work if it is less than 0.7. Can you create the version 2 with limited focus?