Trying my hand at training a Z-Image Turbo concept. The data set was pretty simple and uniform, using the same caption for all images. Don't expect this model to be particularly versatile. Recommended Scheduler/Sampler: DDIM/DDIM_UniformCFG/Steps: 1.0/11 for close to medium shots, try 2.0/20 for wider FOVs.I've tried a bunch of different resolutions and aspect ratios and they all seem to work fine.Faces were either fully or partially cropped out of all the images in the data set, so it should work well with character Loras. I find even ZIT character Loras work well if you bump their strength down to 0.9.You can try combining with other Loras. Most of the samples are using at least one other Lora.The PNGs should have my workflow embedded.Basic caption:Photograph of a close-up of a nude woman's buttocks and vulva. The focus is on her anus and labia, with her buttocks prominently displayed in the center.Check the prompts in the samples for more ideas.

Files with SHA256: a2ee86c49272df513ac40056e612881009126359597e869b0dd650da97a2d74c