Lora to have the subject bounce their pecs.
LTX 2.3
There's not an explicitly defined trigger word, but ever video in the dataset was captioned with "does a pec bounce", so you should use that.
Wan 2.2
Add pec bounce to your prompt.
I typically use Euler Simple with:
High - CFG 1.5 / 5 Steps / 0.6 Strength
Low - CFG 1.0 / 5 Steps / 0.6 Strength
I use this with the Wan Lightning Lora, 2.0 strength on High, 1.5 on Low. It seems like there's a slight desaturation effect when using the Lora that tweaking the Low weights can help with.
I've also tried setting the High CFG to 1.0 and increasing the strength. Sometimes better, sometimes worse.
Example workflow: https://pastebin.com/3C87PMzY
Description
FAQ
Comments (15)
Nice. What I really need is a lora that creates larger pecs.
Like an Edit model where you pass in a pic and say "Make the pecs bigger?"
@cruffsant461 that would be good. Or just a regular image lora trained on larger pecs. But an edit would be perfect I think.
@cruffsant461 I'd love to make larger pecs without also getting ahuge bodybuilder physique
@ForeverNecessary737716 You're probably gonna want to try training a Lora on the specific images you're thinking of, then. Z-Image might be good for that, and you may be able to even train in locally
@cruffsant461 I kinda want to make images that don't exist. Like skinny guy with large pecs.
@ForeverNecessary737716 hmm, you're either gonna have to get very specific with prompting to generate a starting dataset for yourself or maybe you can pass some pics to an edit model and say "only make the pecs bigger" or something
do you think we can have a female pec bounce version please?
hi, it's me again with a few questions:
1. what strength of the lora you'd recommend for the best results?
2. what workflow did you use to get the sample animations?
3. does the lora work well with anime images?
Depends on the image, but I think I typically used 0.6 on High and 0.5 on Low.
I'll attach a pastebin example workflow.
I think it should? Can't remember if I've tried it.
@cruffsant461 ok, thanks