UWU_Lustful_Clarity
Details
Download Files
Model description
ABOUT THIS LORA:
Designed to help my UWU XL model perform a bit better when prompting for explicit adult content, but it also helps improve retro fantasy styles and overall any fantasy styles. No tags needed just prompt like normal for my main model. Much like my main model this was trained on natural language descriptions and tags as well as Florence 2's natural language vision model. (So you can say things like "In this photo I see a sci-fi city filled with people") which should slightly leverage your ability to prompt with LLM descriptions.
THIS IS NOT A FINAL VERSION!
While it does help with intimate scenarios its not perfect and as such please do pair with other loras! This Lora can get you in the ball park and occasionally one shot a perfect image but not always. It still needs further training.
USAGE:
A good strength to use is .5-.75 as a baseline.
I would use this lora in a main initial generation and then pass it back into a kSampler with a denoising strength of .3-.5 with any other Loras you'd like to apply as far as positions or actions go.
Style Loras + UWU Lustful Clarity --> Position or pose Loras --> Detailer Loras --> Upscaler
MODEL BEHAVIOR:
This lora does have an extreme bias with the Male+Female characters, as does my model in general. Mainly the faces. This is because the dataset contains images featuring my own artwork and photo collections. Out of the 70k images used for my Main model an additional 5k where generated depicting specific actions and then prompted via PromptGen 2. Resulting in better adherence to prompting for lewd or sexual actions.
TRAINING:
The Lora was trained on 5k images, and every 6k steps reduced by 1/4th until about 1k images were left. About 1k of the images are sexual in nature, 2k or so are actual photos, 1k are illustrations/anime styled, and the remaining are personally generated images.
ALL IN ALL:
Results may vary, and I haven't tested it on any other models. This Lora will absolutely result in very different images/generations probably within any SDXL model.
SAMPLE PROMPT:
contrasting, flat shaded minimal high fantasy work painted by Frank Frazetta and Rowena Morrill, 1970s, mythos of the damned, 1923 Weird Tales, haunting horrific natural structures, scary and twisted, nightmarish, in the ashes of the dead, decayed in hell, stalking, otherworldly, godless presence, unsettling, cursed, reaching forward, rebirth of the dead,
In this picture I can see a epically painted masterpiece,
A illustration digital painting, dark fantasy, Martian landscape, moonlit, barbarian standing on a pile of corpuses, a scantily clad barbarian girl is kneeling, by an immense fearsome monster in the background, iridescent scales, fog covers the landscape
camera_angle: from above, art_style: illustration, image_composition: middle




















