SEX MACHINE
This is my first LoRa trained for the WAN 2.2 5B. I chose this model because it's fast enough and delivers good quality for my needs.
Prompt example:
video loop. stationary camera. rsc. fast sex, rhythmic motion. rsc. girl's thighs are trembling. she is using a sex machine with a dildo attached to a metal rod. The dildo thrusts back and forth into her vagina, sex machine penetrating her. deep penetration. hard thursting. pussy is squirting. secondary tissue motion. Realistic soft-body physics, ripples move across her skin with every thrust. breasts jiggling and bouncing.Limitations: This first version was trained on only one position, so it may exhibit inconsistencies in others.
I'm using this workflow WAN 2.2 5b i2v Workflow.
Description
FAQ
Comments (16)
5B is seriously underrated.
People would rather spend hours waiting on 14B than explore new options, often without even having the right hardware to run it.
thanks for this
Yes, I myself realized that I get better quality results with the 5B model because my hardware doesn't allow me to push the 14B model too far.
Are there GGUF versions of 5b?
@TheGlowingGuardian
GGUF: QuantStack/Wan2.2-TI2V-5B-GGUF at main
FP8: meimeilook/Wan2.2-TI2V-5B_FP8 at main
@W2R You rock, thank you kind sir
I get very deformed nipples with 5B (t2i), were they censored? Anyone has the same experience? Are there good LoRAs as a workaround?
@zz5 I'm not sure but I think pretty much everyone uses breasts LoRA as the base model can't generate good ones.
@zz5 Use this lora its great: https://civitai.com/models/1811313/wan-dr34ml4y-all-in-one-nsfw With that u can do all sort of nsfw stuff and u will no more have the nipple problems. And use fastwan lora for 4 steps T2V and turbo lora for I2V 4 steps.
@AndyZocker thank you! I will try it. The full link for those who can't find the right version: https://civitai.com/models/1811313?modelVersionId=2099309
EDIT:
I'm using 2 loras now: dr34ml4y-aio-nsfw weight=0.3 and pov-titfuck-lora weight=0.55
any plan for 14b?
If someone could figure out how to make a character identity retention lora for 5b i2v then we would seriously be in business.
Just describe your character's features in the prompt. In most cases it keeps things consistent. You only really run into problems if a lora is trained "incorrectly" and likenesses are baked in (either due to bad training, or no other option).
"character identity retention lora" is your character lora. its that simple.
@TurboCoomer no it's not, 5b sucks ass at character retention. Especially in I2V if it's not a headshot. It's the main reason there is barely any downloadable content for it. (None of it has anything to do with this lora, which works great)
@TheNecr0mancer what do you mean "its not", you are seriously asking for a lora that will magically preserve ANY character while refusing to use character lora that are intended to serve such purpose?
Wow. Would you consider porting this to the 14B variant?