I believe the workflow used to generate the samples was created by a former user - playtime_ai
It utilizes the MoE ksampler.
I get best results using the following settings:
Sampler - er_sde
Scheduler - beta57
Shift - 8
Steps - 8
Lightning - i2v_lightx2v_4step_1022
Str - 1 on both high and low.
(Yes, I use the i2v lightning lora for T2V generations. I don't know why, but it always gives more more dynamic and realistic results).
Trained with the following trigger words:
cock_worship
cock_lick
cock_slap
blowjob
deepthroat
titfuck
handjob
Does somethings better than others. Slapping and titfuck it struggles with. Occasionally does some fun body horror.
Follow up edit: I've added some clarity to the description and will attach a zip file to the T2V Low noise model for all the captions used for Lora training. They all follow the same structure - start with the trigger word(s) featured in the video trained followed by solely focusing on the movement during the video. All videos were 320…
