Pussyjob a.k.a. grinding. This LoRA is trained for LTX-2.3
v1.1 tips
I2V
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
FAQ
Comments (21)
Amazing job. Even T2V looks like I2V, very realistic looking and good audio. One of the best loras so far.
any idea what t2i model and what lora to use to get starting image? Klein, ZIT, Qwen?
Klein 9b is really amazing and fast, better than ZIT by far. And u can both generate and edit images. U can talk naturally to the model and he does just what u want. And there are many loras already for this, i'am using a general nsfw lora for most of the nsfw stuff. But for this lora u don't really need a image just copy the prompt and modify it, it works great for ltx without a start image. I use this workflow so I can edit and generate images simultaneously. It's far better than the official one because it has bugs. https://www.reddit.com/r/StableDiffusion/comments/1qeam1k/i_made_a_simplified_workflow_for_flux_klein_9b/
Excited to try this. Any advice on the Model CGF, Steps and Lora weights for I2V only?
Thanks. The workflows are embedded in the sample videos, so please take a look at them.
@2929dance Yes, I ended up using this wf and got great results! Thank you.
This one is a good thank you!
i2v changes character quite a bit for what ive tried so far
Thanks for the feedback. It seems lower weights like 0.7 sometimes help. Also I found you must describe every visual aspect in the image for better consistency. I'm preparing an improved dataset to prevent this, but in the meantime please try them.
@2929dance did both, much better, thanks :)
damn
I can't get it to EVER show the woman's face for i2v.
I try all sorts of aspect ratios. the word WIDE CAMERA SHOT. nothing works.
Sorry to hear that, and thanks for the feedback. I am aware of the issue, and there are some ways to prevent it. I added a brief note in the description. Hope it helps.
could be wrong, haven't tried this yet, just downloaded it, but, things that arn't explained in prompt, tends to not show up.
so if prompt is something like "she rubs her pussy on the man's penis" it won't show face, you need to describe the face.
you get the same issues in reverse if you overly explain something. like if you says "red shoes" just encase the camera shows feet, it will guarantee it shows the feet.
i've had some luck with weighting things.
@MrReclusive666 i literally used the same prompt as the cheerleader one and it mentions her face
@2929dance is there any lora for an editor that can make the start image? Or a prompt that doesnt just insert the penis?
@kronos1959777 I've been using these: https://civitai.com/models/727117/grindingillpony https://civitai.com/models/507472?modelVersionId=2274687
@kronos1959777 I published v1.1 which should mitigate this issue. I'd appreciate your feedback!
Great work, Loras! I would like to know how your training set was processed by using what method for labeling, and how many training samples there were. It would be even better if I could refer to your training configuration file! thank you