shared their training data and I used it to train WAN - it seemed to have worked out well
No specific trigger word, but you'll want to use the following phrases: shaking breasts. A woman is shaking her breasts from side to side.
check out their version here:
https://civarchive.com/models/2537439/shaking-breasts-side-to-side?modelVersionId=2851713
Description
FAQ
Comments (11)
Can this be used for i2v?
I've only tried it with i2v, not sure how well it works for t2v. I listed the model as t2v because that's how it was trained, but almost universally t2v lora for wan are usable for i2v.
@CallMeMaybe Not as well with Wan2.2 compared to Wan2.1, in my experience but you are not wrong. Curious... why would you use another users dataset to train and then post the exact same model, with the exact same starting images? What is the point?
@playtime_ai_ He didn't train WAN - his is for Hunyuan Video (original, not 1.5) said he didn't have a powerful enough computer. He said he'd share the dataset if someone wanted to train WAN, so I asked for it to give it a try and posted the result here.
You should all download this from the original user who gather the datset, captioned it and trained the original model.
https://civitai.com/models/2537439/shaking-breasts-side-to-side?modelVersionId=2851713
Yes, except that's for hunyuan so not the same - although I think their result is honestly better...
@playtime_ai_ Thanks for the extra visibility! Sadly I don't think too many people use original Hunyuan Video/FramePack anymore, but I think the LoRA is decent.
I was happy to share my dataset for someone to make a Wan version, and I don't mind my images being used to make the showcase videos, though I guess it might make more sense to use different videos as the showcase for this one, in order to better distinguish the Hunyuan and Wan versions from each other. Ultimately that's up to @CallMeMaybe though.
@Insult_to_Ninjary - oh I used the same image because 1: I used them to test the results and 2: because I wanted to make it obvious that it was your dataset!
Honestly, I haven't even tried it yet on anything but those example images, I've been curating a dataset for something else.
If you'd prefer I can differentiate it with some different samples!
@CallMeMaybe I don't mind either way, but it might make it easier to distinguish. Or at least maybe use a new video as the main first showcase video, and then keep these two after that one? Up to you in the end though, just thought it might help people tell them apart more quickly/easily!
New showcase vids are muy caliente, nice job!
@Insult_to_Ninjary Thanks! I appreciate your efforts with the training data. There's always so much to learn...