Anal Helper LoRA (Not a slider)
I have found it frustrating that most LoRAs and checkpoints cant understand the difference between anal and vaginal. If this is because people use LLMs to caption their dataset, or Krea2 is just bad at understanding the difference i dont know. No matter why it is like this, propting for anal seems almost impossible from time to time. So i made this small helper LoRA that was trained on only croped anal images without faces. It was trained low and slow for 14000 steps to have as little impact as possible on other LoRAs, and should mainly just help making the penis, fingers and dildos go in the bum.
Only 27mb file size.
It is design to be used in combination with other LoRAs or checkpoints, and Its not so strong that it will do much if used alone.
Trained on:
16 dildo images
25 fingering images
29 penis images
Description
FAQ
Comments (21)
I had this exact problem. You are awesome. and low model size.
You hit the nail on the head with your description. I really appreciate your work on this one. Thanks again for all your wonderful Lora's. I would hit you up with buzz but I have none and no crypto.
Could you make a deep penetration LORA?
in the bum :D
thanks for making this.
It works good on its own with i2i and the controlnet node even at 100% denoise, it is truly functional.
Nice.
Only thing I wish it would not affect the vagina. But I guess it is too much to crop in the dataset then. AI will have problems to learn the concept?
Anyways, I am still thankful for this lora. Thank you for sharing with us!
I dont think the concept would stick to well without. But there is worth a try, since this is the most well recieved LoRA i have made in a while. I just need to gather a dataset that is a bit more cherrypicked. But it will take some time to create a good dataset, and to make all the captions. I caption all images myself, so that alone take a couple of hours on a dataset of this size. And the training takes a day as well, so it takes up a lot of GPU time i dont have available atm.
Thank you for this one. Its great as always.
I have the exact opposite problem, all my loras do is push anal. I wonder if you use a negative number it will force vaginal?
No. It will just degrade quality. Big diff between sliders and normal loras in that sense.
This is very useful👍
fantastic, thank you, works perfectly.
Do you plan to do the same for vaginal ?
If im making one for vag its probably going to be focused on mainly fingering and toys. The dataset and manual captions i do whats holding me back, since that takes hours..
Dude, you are fantastic. I found one of your LoRAs yesterday, tried it out and really appreciated it, so I downloaded everything you have for Krea 2 and it's all damn good! Most of your LoRAs outdo other popular LoRAs – and those LoRAs clock in at several hundred megabytes, or sometimes even gigabytes.
This is what a LoRA is supposed to be.
This is the first time ever for me leaving a comment on anything, because it's so freaking good. Thank you for your amazing work. I really appreciate it!
//Rob
Finally last thing i needed!
changes the face.
Ok. Thanks for the very constructive message.
The faces in my samples dont seem to change much, at least compared to any other LoRA.
But when you only write 3 words there is not much i can help you with. If you want the faces to stay pixel identical, may i sugest you use photoshop instead.
@Loraholic
Ok. Thanks for the very sarcastic message.
The face doesn't change with realism engine or snofs as they mask the face data out during the training.