This produces clean, colorful, well defined images of a woman in the theme of a Witch.The model was trained on full figured woman with light and dark skin tones, mostly dressed in white clothing and hat, but Chroma can change these colors easily.Because of the tuning method used, this LoRA is likely to produce nudity and sexual interactions without being prompted to do so and negative prompts may not alleviate this apparition.None of the example images were post processed or scaled, except the occasional gamma correction.The XL version was tested with stylized fine tunes, Illustrious and realistic themed models. Each implementation having its own delights and drawbacks. In order to make up for some of the common issues, I added closeup shots of things like eyes, lips, hands, feet and toes. Some few explicit source images were also used. All of the training images were AI generated.

The Witch

I've chosen to bump this number to 2.0 since it's a retraining.I was pleased with the previous version for Z-Image but the results weren't spectacular, not even as good as the original Illustrious from which I derived the dataset. This one is different, and rank 64. I prefer the detailed images that come from this over the dataset. The captions produce the images almost exactly, with much more crisp details and gentle contrast.The colors are so vibrant I couldn't find a reason to NOT post this. There's not a lot to say outside of what's posted in the main description. I hope the example images give it the presentation it deserves.

Files with SHA256: 519991721a8db6366a4e239dcefaeb42191badb8a7098bb4969a694c2d59dbc0