Cupping hands!
I think this one turned out pretty well, the triggers are: cupping hand, cupping hands, waiting for cum, cum on hands, clean hands, drinking
Tags
Most face and expression tags also work, here's some more detail on what each tag does
cupping hand: one hand
cupping hands: both hands
waiting for cum: this one is multi purpose, in general, using it with any other tags improves hand quality
cum on hands: also multi purpose, and why I might retrain, works great! but, it's difficult to control cum in mouth and on face independently
clean hands: clean hands, doesn't really work?
drinking: this one! okay. I had like 7 images for this, but it turned out better than I expected, very difficult to control, but when it does work, it's amazing!
Extras!
If you want to make the gif in the sample, I included the prompts in the training data section for that version. You can use that with the prompts to gif script on my animation helper scripts page.
Description
First version; I might retrain for cum on hands to be less powerful, and for drinking to have less impact on style
FAQ
Comments (9)
There is another amazing wonder beside this model: HOW DID YOU GENERATE SUCH CONSISTENT CONTENT TO MAKE THAT GIF ???
I think it's about img to img and adjusting the denoising strength.
@Waterflow Nope. No img2img on this one. It's prompt editing. In the image download section, I have all the prompts for download.
The big thing is training the lora to understand the concept well enough that it can make consistent frames.
@emberbyte Im assuming you consistently used the same seed for all images too?
@ogathereal img2img or inpaint : never use the same seed. Would be nonsensical.
@hansolocambo this video I used the same seed for all the images, didn't use img2img or inpaint, I mainly relied on prompt editing to iterate the frames.
@hansolocambo That said... you could get more consistent quality for animations by adding in inpainting and/or img2img; the main purpose of the gif was to demonstrate how consistent the results from the lora are for different tags
@emberbyte thats what i thought. if it was img2img ofc seed would need to be different but i dont think the result would be as optimal as prompt editing
You can use reference in control net.