Minimax H3 Example Promts

The H3 model is quite versatile and this lora replaces also the Bouncy Walk and the Bouncing Forward Bends Lora try Prompts like:

• She is doing a sexy dance her breast is bouncing up and down and from left to right

• She is bending forward deeply and her cleavage gets visible her breast are swinging and bouncing from left to right

• She is walking into the direction of the backward moving camera

Minimax H3 V0.7 FL2VA UPDATE

This is the exact same training set with the exact same step count as the Ref to video model but trained on the FL2VA Model. In my personal opinion I think the Ref2VA Lora is doing also a great job even in FL2VA but I just wanted to ensure that we also have a proper trained FL2VA one. So its totally up to you wich version you prefer...

Minimax H3 V0.7 REF2VA UPDATE

This version improves the softness and the side bounce effect. Let me know what you think in the comments!

Minimax H3 V0.3 REF2VA UPDATE

First version released. Lets check it out ;)

LTX 2.3 V2.5 UPDATE

I have trained with a bit more variety, to get a more stable effect.

Hope you like it, let me know in the comments :)

Also check my updated base workflow with the new Ltx 2.3 1.1 distill lora

https://civarchive.com/models/1598638?modelVersionId=2864162

LTX 2.3 UPDATE

Yes made my first attempt on training a LTX lora. It clearly adds some nice bounce, but im not sure if its there yet. Let me know in the comments

In doubt use my base workflow:

https://civarchive.com/models/1598638?modelVersionId=2831320

WAN

This lora is for animating bouncing breasts with Wan. Since I felt there was a lack of it here ;)

Used I2V-14B-720P as base.

Since ppl were asking about my training worklfow:

I have used 7 videos with about 4 second length, with faces cut off. I have described what happens e.g.

A girl with blonde extensions is standing in a hotel room, she wears black lingerie, her breasts are bouncing shaking or wobbeling up and down, her hands stand still, her face is cut off

Then I used Musubi-Tuner with default parameters and trained about 32 epochs, with my 4090 it was done in about 35 min.

That's all :)