Put your hands up in the air! Or down in front, I don't care!
🪩 Hip Sway Dance 🪩
High Noise only!
This LORA is not intended to fill in clothes or bodies, just the groove, which the High Noise model can handle on its own.
Wildcard Prompt
hip sway dance,
A beautiful {Russian|French|Swedish|Swiss|African|Latina|Austrian|Dutch|English|American|Siberian|Chinese|Ukrainian|Thai|Czech|Japanese|Indian|Caucasian|Spanish|Portuguese} woman dancing with {arms up above her head, wrists together|bent arms, vertical forearms, fists in front of shoulders|arms down and forward, hands together in front|},
She is wearing {a sexy dress|yoga clothes|a summer dress|jeans and a crop top}.
{smiling, happy|duck face|serious, concentrated|excited, seductive expression},
appropriate lighting for a {bedroom|dive bar|Western bar|night club}Disclaimer
um-tiss-um-tiss-um-tiss-um-tiss...
Description
FAQ
Comments (14)
yessss!!! do gyrating hips dance lora next!
got any good source material? I randomly happened upon the material for this...
@az420 i usually look up stuff like that on tik tok. that's the style i would like the lora to be in after all, a gyrating hips lora kind of based on the tik tok style!
@TypicalLoraLover any chance you'd be down to collect 5-10 clips of the right kind of movement? would save me some time 😊
@az420 i have some already from a dataset! you might have better luck with it than me!
@az420 https://drive.google.com/file/d/1SWP4d3H8vcKJ7OujFzmLOBcqSZWCOe4B/view?usp=sharing here you go! you might have to trim a couple clips and caption them. hopefully you can do something with this!
@TypicalLoraLover ok, not bad, try to get me one or two where they aren't wearing baggy bottoms, seeing the contours of the legs would help HighNoise
So... I'm confused. Why do you do only the HIGH model?
Only needed for this movement based LoRA. Unless there is some fine detail required in the effect, there is no need for LOW.
ohhhhHHHHHHH!!!! ok- I'm new to training WAN 2.2 lora's- So the HIGH model is mainly for MOVEMENT OR ACTION, and the LOW model is for more detail like 'faces' or characters.
@soundgate High Noise represents what you'd still understand if the frames were blurry. You get main motion patterns, and pose, almost like controlnet. The High Noise is the early timesteps, where you're making sense out of noise to start forming the basis of the render.
Low Noise then fills in the rest, the textures, the colors, the more subtle motions and interactions.
When I train High Noise, I skip high resolution buckets completely, and instead focus on frame count. Low noise needs a good bucket mix to capture textures as well as finer movements and interactions.
@soundgate funny that many low noise loras already exists, so you may train only high lora for almost everything and combine it with already existing low noise loras if you wan't. ^_^
p.s. yes, this lora works for i2v with prompt "hip sway dance, A woman dancing."
Thanks for all the comments everyone- It's helping me understand better, good looking out.