This targets a waist that is 0.4 units of width per 1.0 units of hips, from the front or rear perspective. If you want lower than 0.4 you'll have to train your own model, but this should be easily better than anything else out there for the newer models.Keyword is "tinywaist".I saved it in 50 step intervals so if you think less or more steps could be better I can upload those too.Wan 2.1, LTX-2 and Qwen get very close to the 0.4 target at strength 1.0Klein and ZIT get to about 0.5-0.55. Since it met my ideal of 0.55 I stopped training as there were diminishing returns. Either above 1.0 strength or a LoRA with a stronger dataset might get that lower.Actually, shoutout to @adabot007. His Z Image waist slider well into the negative combined with mine at full positive strength gets it below 0.5! Also makes them a bit taller. I guess that solves the waist problem.The main purpose here is so that Wan will not blow up the waist every time there's a transition or the waist gets hidden by legs etc, or just gradually over time. But also so Qwen and Klein can make next scenes too without butchering the character proportions. It works on all versions of Qwen Image and Image Edit.Also seems to make Wan women more attractive overall and improves their hair for some reason.Trained on nude images. Credit to freckledvixon's Vixons Optic Muse checkpoint, SkinnyWoodPlank's Thin Waist SDXL LoRA, and BalaenaPBC's Narrow Waist LoRA for making excellent training images.Hands may be a bit messed up due to limitations of SDXL based training images. I should have made them raise their arms. There wasn't exactly that great of a waist slider on newer models, so...Please make some really good transhumanist porn with these.Also all potential Buzz proceeds will be used to make LoRAs to make better Missionary POV videos of princesses with tiny waists. Maybe some yandere themes too.Next stop: "All the way in" Missionary POV LoRAs for all five

Flux Klein 9B, LTX-2, Qwen, Wan 2.1, ZIT Waist Sliders

zit_tinywaist_000004050.safetensors

Files with SHA256: 599298dd44514e96700e8f3645089b5830a884a9c747b9dd4aa4f8287a913eaf