Put something like this in your prompt:
breasts bounce, shake, jiggle and sway.Should work with I2VA. Idk about REF2VA, but feel free to try it out and share results! The first 2 video examples are T2VA with the lora strength set to 1, workflow and basic prompt included from video. I trained at 3000 steps and 10 epochs. This is epoch 5. Thanks to anonimoose for sharing the dataset.
MiniMax H3 Prompt guide for you/LLM.
https://huggingface.co/MiniMaxAI/MiniMax-H3/blob/main/docs/VIDEO_PROMPT_WRITING_GUIDE_base_en.md
https://huggingface.co/MiniMaxAI/MiniMax-H3/blob/main/docs/VIDEO_PROMPT_WRITING_GUIDE_ref_en.md
NEW* H3 SLOP TWERK: https://civarchive.com/models/2885070/slop-twerk-minimax-h3-lora
If body is too big, try pairing with a breast/body type lora and slightly lower both strengths.
If clothes are vanishing, you may have to prompt clothing.
Description
Put something like this in your prompt: shakes her hips side to side causing her breasts to bounce, shake, jiggle and sway.
Should work with I2V but probably lower strength. I trained at 3000 steps and 10 epochs. This is epoch 5. Thanks to anonimoose for sharing the dataset.
FAQ
Comments (28)
Woohoo!!!!
Who's the hero?! It's you!!
r2v works. But it seems to force the slow motion even if I prompt about speed and frequency
And the tits are always out, the size is always huge.
@AnonymousAAA Try lowering the lora strength value to like 0.3
this lora might help speed things up https://civitai.red/models/2840146/faster-harder-shake-harder-or-h3-motion-boosterref2va-update?modelVersionId=3228867
After some testing with a different lora having this issue as well, I think prompting clothing may stop the clothes from disappearing and you might be able to pair a breast/body type lora at like 0.8 STR with this lora at 0.8 to stop the body from enlarging
Just curious since I've tried something similar with mixed results/not very good, you use aitoolkit and train with videos? So you keep the i2v part unchecked? o.o;
I am unsure sorry. I trained it on civitai MiniMax H3 base model
its fantastic! But seems to give a lot of the time the same body shape/stomach.
tested on different strengths
Sorry I am a bit late. It sounds like I could've over trained it slightly but not sure exactly (I am noob), I am looking into it. Perhaps I will make a V2 to fix this. Ty for the feedback : )
@bhopping No problem! You are doing amazing work :) I found turning the strength to around 0.45 and vis to around 0.70 gave a nice balance. Keep it up!
@vdsvsdvsd3433 Nice btw I guess the VIS knob is a multiplier for the strength, so I think your strength is technically set to 0.315, I totally thought it was something else
A masterwork!
thx for the buzz!
@bhopping You’re welcome!
Nice work ♥️ You can do slop bounce for as s s? 👌👌
Currently baking in the oven as we speak. Stay tuned!
Converts images to realism at high weights currently.
I might've over trained it slightly. I didn't know that was a thing, I am looking into it. Ty for the feedback!
@bhopping You are most welcome, and thank you for the lora! Maybe you can use animated images at low weights as part of your training set if you are lacking data.
Unfortunately does next to nothing that I can see, in R2V. You can prompt heavily for it, but the model already has good 'physics' baked in, so prompting it isn't really a feature of this LoRA. Definitely different than the W2.2 version, which was very much just "it just works"
Dude, you are a boobies sorcerer...
Your contribution to female sloppiness science is beyond all what could have done these crazy chinese searchers... Beijin should hire you... You just fixed the main problem with Minimax H3. Love you man
Hi, Can you tell me which checkpoints and models were used to generate the image (not the video)?
2B? That was just from T2V model, some of the training used 2B videos. Is that the one you are referring too?
@bhopping Thanks for getting back to me, I was under the impression that you generated the images using a separate checkpoint and then turned them into videos with MiniMax. I never thought it was possible to create work of this quality using just a T2V model. Could you tell me which T2V model you used? I'd like to give it a try, so it would be great if you could briefly point me in the right direction.
@ssfsd4624623523 I am using minimax_h3_fl2va_pruned_fp8_scaled but I think most ppl are using INT8 over FP8 for more speed if you are on CUDA 13.0+ from what I heard. I just haven't updated my stuff in a long time so I don't reap the benefit of INT8 yet