Overview
A female bodies expanding into serious thickness. This model focuses mainly on thigh/hip growth, breast expansion and ass expansion, but can also do other things. Just play around with it.
There is no trigger; use following descriptions:
Her thighs grow (massively)
Her breasts grow (massively)
Her buttocks grow (massively)
Her belly grows
Her height increases (dramatically) / her head hits the ceiling
Her clothes rip / tear / shred / tear open / rips across / leaving exposed
Adding my T2V helper LoRAs will improve the quality and/or growth rate significantly. Also, using the TripleKSampler can greatly improve the animation quality, if you're using speedup LoRAs.
Further descriptors can enhance the effect. Sometimes it can take a few tries to get it just right. These keywords can also be used in the negative prompt to weaken an effect.
For more have a look at the example images/videos.
LTXV 2.3 - V1
Coming in July... it’s really hard to train this one. I had some success, but I’ll retrain for LTX 2.3 Sulphur.
WAN 2.2 I2V - V2.9
Training model: WAN 2.2 I2V
LoRA strength: 0.5-1.0
Sampler: euler / euler a
Use TripleKSampler with 1/4/4 steps or no lightning LoRAs at all for quality.
I highly recommend using my T2V helper LoRAs linked below. Use their low-noise component for improved texture quality or their high-noise for better shapes and bigger expansions.
A woman in a bra standing in a bedroom. Her breasts grow massively; her breasts become gigantic. Her bra stretches tight. She looks down shocked.Description
FAQ
Comments (7)
I've been trying to find a lora with wan2.2 where the person bursts out a building from growth but that's rare to find material in itself. do you have any tips for me to get started training my first wan2.2 lora? I'm thinking in general I need to curate the dataset with like 5 seconds worth of those scenes.
But would I need to have the resolutions all be the same? thinking 50 5 second videos dataset should be enough?
Thanks in advance for any advices!
That’s very ambitious! 50 clips would certainly be enough. They don’t have to be all the same resolution, WAN can handle variation there.
The real problem will be temporal consistency, physics and preservation of details, because WAN wasn’t trained at all on this. It needs to learn several concepts from scratch (size difference, major destruction, fantasy physics).
Your captions have to be precise, your clips varied and high quality.
thanks so much @cupra you've inspired me to try my best! :)
@growthwind you can do it!
A really nice lora you have there. Nice to finally see some good expansions loras.