Join the NSFW AI community on Discord where over 6,000 NSFW AI developers, creators and enthusiasts gather to share tips and tricks, talk about the latest models, and collaborate on research to push the cutting edge of NSFW AI!
https://discord.gg/mjnStFuCYh
Trained on 500 videos and 300 images to give the most realistic cum available for HunyuanVideo.
You can use this by itself or mix with others to enhance their effects.
It has a tendency to distort generations at 0.8+ strength.
I recommend starting it out with a low strength of 0.1-0.2 and gradually testing with increasing values until results begin to degrade.
Description
FAQ
Comments (10)
Does the job, but oddly I can get better loads to show without the lora and just some excessive prompting.
Nice, definitely share any good prompt tips you have! Could help with captioning future datasets.
using hunyuan wrapper or native comfyui? I've noticed some things can be done easier in the wrapper without lora.
@Redbird Native. It's definitely a challenge for good goop thickness. hunyuan likes to make it very liquidy. This lora helps to thicken it some.
for me it seems to remove any cum from the scene , without it I get some, tried .7 to 1.1 , nothing
@mofo69 Did you try lower settings? It's trained pretty heavily and I noticed best results 0.2-0.6 strength
@BetterPorn Just tried .1 to .6, not a drop. :(
maybe its a conflict with other loras Im using , ill try on its own
edit, yep works ok on its own , thanks but it doesnt seem to play well with any other loras, as soon as another lora is present , no cum, lol
@BetterPorn was this trained on double block only?
@KinkMaster No, I've explored block freezing quite a bit for training but HunyuanVideo learns temporal relativity across all blocks, so freezing can have pretty destructive effects in some ways. Instead, I've been training with either a uniform timestep sampling or using shift with higher values to get motion locked in.
There is sadly still no cum lora for Wan.