This LoRA focuses on (semi-)transparent fabrics like mesh, lace, chiffon or tulle. It is trained on how they interact with light and the underlying body. It enhances the realistic behavior of sheer materials by improving diffusion, fabric density, edge definition, and visibility beneath the cloth. The LoRA is trained on lingerie and nightwear styles including sheer bras, mesh bodysuits, lace camisoles, chiffon nightdresses, and semi-transparent skirts and slips, in colors ranging from neutral beige and white to black, navy, and soft pastels. You get pretty good nipples with the default flux1-dev model; they are not perfect but still pretty realistic and I didn't want to overshoot the training goal. You can also pair it with NSFW models to get slightly better nipples but you have to take care of the weights to prevent the nipples sticking through the fabric (read further down for recommendations.) Information about the showcase pictures (all were done on default flux1-dev): The first 10 images were generated based on the captions of the dataset Images 11-15 were generated with an open prompt to let flux decide Images 16-20 are generated with the "mysticXXX" LoRA while mysticXXX hat 0.4 strength and sheertex 0.9 If you use nipple LoRAs you have to lower them waaaaayyy down to something around 0.2 or 0.3 or else they will stick through the fabric. Avoid prompting for nipples or it will happen too. Pure nipple LoRAs have a high signal strength and will override sheertex if you go too high with the strength. Works with character LoRAs. I am using (SFW) character LoRAs at 0.95 strength, a NSFW LoRA (mysticXXX or a nipple LoRA) at 0.2-0.4 and sheertext with around 0.8. The LoRA was trained on the default flux1-dev model.