V3 VOID-

TRIGGER WORD IS NOW "handvoid", specify that the hands are touching them and where

after 3-4 days of experimenting and quadruple the dataset size, I have an update for this model! much more consistent and better quality. YMMV when using finetuned checkpoints. It works very well with OG Z-Image Turbo

--

V1/2-

I am back with an amazing LoRA because I finally figured out how to train locally!

So basically, this model allows Z-Image to make many mysterious hands actually grab and touch a woman (and probably man, not sure). I've made loras of this concept last year with Flux 1 and SDXL, and they didn't work very well but we've (and I) have come so far!

Keep LoRA strength to 0.8, if it looks too cartoony (its trained on mostly 2d art after all), lower to 0.7. if need more grabbing, up to 0.85 if its not cartoony-looking

Trigger word is "handgoo", specify that the hands are touching them and where. example prompts in images above!

notes:

• the model has a tendency to make the woman grab her own breasts if you don't specify her arm position. i pose them with arms up and it works just fine