This model comes from 7 real materials for more than 1 minute and was trained on Nvidia a6000 for 24 hours
He works very well when text-generating videos, because I use the base of Wensheng Video to train
You can use a {woman lying on the bathroom wall, shake ass, full body shot} or use Chinese directly
I haven't tested the effects of images to videos yet. Maybe you can experiment with them. The weight can be opened to 1.2-1.5 under the premise of matching other loras.
Of course, if you only use a single model, you can open it to a smaller value, about 0.5-0.9
Because it is trained under an uncompressed model, it is not recommended to use unofficial nodes, as the effect will be very bad, especially on gguf, which is extremely limited
My suggestion is to use a larger model because it contains a lot of actions, rather than a compressed model, it will cut out a lot of actions.
Description
This model comes from 7 real materials for more than 1 minute and was trained on Nvidia a6000 for 24 hours
He works very well when text-generating videos, because I use the base of Wensheng Video to train
You can use a {woman lying on the bathroom wall, shake ass, full body shot} or use Chinese directly
I haven't tested the effects of images to videos yet. Maybe you can experiment with them. The weight can be opened to 1.2-1.5 under the premise of matching other loras.
Of course, if you only use a single model, you can open it to a smaller value, about 0.5-0.9
Because it is trained under an uncompressed model, it is not recommended to use unofficial nodes, as the effect will be very bad, especially on gguf, which is extremely limited
My suggestion is to use a larger model because it contains a lot of actions, rather than a compressed model, it will cut out a lot of actions.
FAQ
Comments (16)
Maybe everyone can share your works to open my eyes!!!
Block Recommendation 19, Sampler euler, step 23, num33, shift3
do image2video
please consider making an I2V version
Text to video by nature works on image to video. t2v loras are the best quality but i2v can't be backwards compatable.
不需要clip训练，因为这是最大兼容度
does it work with image to video
sure
Thanks a lot for this model! I've been experimenting with it in i2v mode, and I realized that it's currently the best model for everything that involves ass bouncing and even twerking.
I'm still trying to find the perfect prompts for the desired movements, but it's clear to me that your model is currently the very best one for it, so I congratulate you for your great work!
If someday you plan to work on a new version and if you need suggestions or ideas I'd be glad to help :)
Yo, thanks, 'cause his collection's all top-notch picks!
@spaceai2077 Hehe to be honest, I didn't know the woman you used to build this model but I bet her content must be great because your model is great!
I've tried to combine your model with various other models in order to get a twerking effect, and sometimes I get satisfying results. But it's very dependent on the prompt sometimes.
I've also realized that there are different "variants" with ass shaking/twerking in real life, and I believe it would be a cool idea to categorize them so that the model knows what we're trying to achieve.
But for now, your model is the one that is the most complete in this area, and by far!
Super hit and miss for me but when it works. Perfection... Thumbs up.
your vid is so perfact..
@spaceai2077 Oh awesome! Thanks for the pins. :)
When will this be updated?
Will this work on Wan2.1 1.3B?
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.