I haven't been satisfied with the Female body lora's for Zimage mainly in their ability to get anatomy right in the view from behind and glute creases, vulva, ect. In an experiment to remedy that and just create a general nude Lora I curated 380 images at 1200x800 resolution and Trained the Lora with the de-distilled option in AI toolkit. Due to the large image sample I trained at 15000 step initially but found that still failed at rear anatomy. So I re-trained at 20000 steps then collected the last 4 iterations. I don't think this issue will be fully resolved until the they release the Base model. The results are interesting and even useful is specific cases. But unfortunately the Lora will impact the overall image results. However I find its fun to play with the lora power and even mix it with other lora's . So I decided to share a few version for anyone that wants to play with the Lora variations the sweet spot seems to be at 19000 steps . play with the power but the higher you set it the more it degrades creativity in your image. due to the llama image tagger I used, its helpful to use the word vulva instead of vagina, I recommend you try blending it with Midjourney Luneva Cinematic Lora at a low power like 0.10 then run this lora at anywhere from 0.40 - 0.8