Pussyjob a.k.a. grinding. This LoRA is trained for LTX-2.3
v1.1 tips
I2V
Example first images: https://civarchive.com/posts/28787989
Weight: 0.4 - 0.8 (reduce if the reference image is affected)
Try to explain everything in the image in the prompt
T2V
Weight: 1.0
If it affects the reference too much in I2V, the following prompts may help:
Positive prompt: (strangely, "loop video" seems effective for some cases)
(details of the every visual aspect of the scene), loop videoNegative prompt:
close-up, her face is out of frame, dynamic angle, camera motion, handheld camera, transition, cut, undressv1.0 tips
I2V
Weight = 0.7
Try to explain everything in the image in the prompt
T2V
Weight = 1.0
Use other nudity or penis loras to enhance the details
Description
Trained on Sulphur 2 base: https://huggingface.co/SulphurAI/Sulphur-2-base
FAQ
Comments (13)
I briefly tried the old version and managed to make it work for things it wasn't meant for. (Grinding but in the reverse cowgirl position) But the Lora kept wanting to add a penis attached to the labia almost every time. Does the Sulphur version copy the movement itself better or does it still add something between the labia and whatever the person is kneeling on top?
its sulphur. Sulphur suck. Ero10 is x10 better.
For that purpose I'd recommend Pillow Humping LoRA by Playtime_AI. I'm not pretty sure if he has made it for LTX-2.3 though.
Sulphur has a strong tendency to turn scenes into vaginal sex, so I'd imagine it works better if a penis is not apparent in the reference image
@2929dance He only just got started with Sulphur so that one ain't for that I think. I can give it a try though. As for the penis, I very much do want it to be there. It's mainly that V1 wanted to create another penis on top of it or fleshy bits that extend toward the vagina whenever the character lifts their crotch away from the penis the image started with. It's hard to describe without having a video to point to.
we really nwws aome breasts shaking lora,and ass sgake twerking lora
Will it still work for regular LTX, non-sulphur?
No, and vice versa: the LTX-2.3 LoRA also does not produce good results when used with Sulphur. That's why I trained it again
Your LTX 2.3 lora is so strong, I'm curious about how you feel about the Sulphur lora. Do you like the result? Is it different to you than the 2.3 Lora, even running that one on Sulphur?
I think the results are as good as the LTX2.3 versions. I think I haven't found good settings for Sulphur yet, though. LoRAs for 2.3 and Sulphur seem not very compatible with each other, so I needed to train this again.
@2929dance Interesting. I wouldn't have expected that Sulphur would require any different settings at all. Glad to see you dialing it in though.
IMO sulphur definitely requires a different setup, at least if you are running distilled. It always gave me garbage until I tried using TenStrip workflows distilled LoRAs. Same for Eros v1. Not Eros beta however, which works fine in regular WFs. 384 fries it into oblivion, theirs are tuned to the newer model. Their custom nodes use a pretty convoluted technique for anchoring, split-distillation and guidance and the custom WFs are an unholy mess, they didn't clean them up at all, but once you get a feel for the architecture it's fine and produces a nice crisp output. Samplers required some tweaking though. That's been my experience. Results may vary. Consult a slopmaster before making any changes to your workflows. Do not stop using your workflows suddenly, as this may cause a life-threatening drop in slop output. This advice has not been reviewed by the CivitAI AI civet.
@Ponder_Stibbons Is there a workflow that handles the custom nodes correctly to make for good outputs using Eros v1 and/or Sulphur? I've only used my old workflow that I built out over time, and would like to learn from a good setup.
@Ponder_Stibbons Thanks for sharing your experience. It's a total shame that they lack a clean minimal workflow that we can actually tweak.