Exactly like my first SDXL lora but flaccid, it seems to have less trace on the overall characters, but at the same time the genitalia are even more dfficult to get, this is more a proof of concept and a tool to be use for training, and merge, it have been trained on the base model SDXL 1, it work best on this, and need may be good in near futur with a more "anatomycal" base models, as for the last one the dataset have been separateed between "Photo" and "Artwork", and the small size is normal, most of the Penis you will see have been estrogenized.None the less it lack some clarity and quality because of the forsekinor lack off inside the database, so expect this region to be slighlty fused... and confused.to avoid burned pictures use the lora at 0.8 strenght and try to play around if you have the patience (SDXL is still very slow and painful to use on most computers)good luck to all ! this is just a pionneer Lora so do not expect the quality the 1.5 could give us yet, but I hope I am contributing to it, so do not hesitate using it to build better lora and models.Update:way better version after cleaning and adding better pics inside the Dataset. the result are way more consistent, and we gain a lot of definition (not that hard to do with the last results to inprove upon it) . recomendend strenght of the lora "1"

Flac-futabaseSDXL

Easier to work, around, less mangled, more consistency, even though complexe pose are out of order, and we get still far from a reliable Lora, At least for now with our actuals model. still trained on SDXL1.0 base model. I made a more consistent database, integrating in it Gennerations from SD1.5 models and Lora. it help having high res not filtered pics. Also curated some more none clear or blurry pics.

Files with SHA256: 0780d63840258fee31b556b8a52d7f22fa4fffb66dd64d700791ddee18995947