Type | |
Stats | 283 24,509 |
Reviews | (22) |
Published | Sep 7, 2024 |
Base Model | |
Hash | AutoV2 F68DF34B9E |
The goal of this lora is to provide a level of control over the components of a diaper. While flux understands that a pair of shorts have left and right sides, it doesn't understand that the sides of a diaper and the center of a diaper are different components that shouldn't naturally blend or share colors. It doesn't understand that the tapes aren't just colorful lines on the front. It doesn't understand that prints can be confined to the landing strip part of the diaper. It doesn't know what a wetness indicator is or that designs can fade when wet. The aspirations for this lora are to teach those concepts. It's not there yet.
By design, a diaper, without any other description, tends to be a standard white diaper. Overly descriptive prompts focused solely on the diaper will likely result in a cropped image. That is normal for Flux, but the lora training likely reinforced that.
This version of the lora does a passable job at:
Overall Colors
Side Colors, independently
Overall Prints
Four Tape Diapers (Clear Tapes with Blue Ends)
Thematically Appropriate Diapers
This version of the lora does NOT do a passable job at:
Front Only/Landing Strip Prints
Intentionally Two Tape Diapers
Intentionally Six Tape Diapers
Any Tape Descriptions
Partially Obscured Diapers
Specific ABDL Brands
Side Specific Prints
Bulges and Swelling
Wetness Indicators
Partially Dissolved Prints
Avoiding Diaper Wedgies and Cameltoes
Complex Poses (Might Be a Flux Issue)
The community is still learning a lot about training and prompting Flux. For all I know, this lora could do better for some of those other things with better prompting techniques, samplers, schedulers, and guidance scales. Please comment if you learn something useful or if you notice some overfitting. I have aspirations of moving things from the 'Not Passable' category to the 'Passable' category in subsequent versions.