The LORA is intended for Flux2klein 9B. It does what is obvious. It pulls the pants down of a person. The idea of the LORA begun as an editing LORA, but to my surprise it works also for text to image generations. Main prompt for editing mode:<ul><li>Pull the pants down </li><li>Pull the pants down, revealing his briefs (you can define the color of the briefs or underwear and you may also add more text as corrections.)</li></ul>Prompting for text to image<ul><li>Define who is the character</li><li>What he is wearing</li></ul>And then write "he has pulled down his pants, revealing his briefs". If you find a stronger prompt, let me know. In the dataset there were only men, AI-generated, but it probably works for women too (pull the pants down, revealing her panties). It is my first LORA and in beta form. Most of the time it does what it is suppose to do, but I am still not content with some limitations. I mixed a lot of different kind of pants, and this was probably the problem. There are limitations which you can maybe overcome with some extra prompting or a different seed: 1. Not enough folds in the pulled down pants. 2. When hand is in pocket, prompt for hand on the thigh, otherwise the hand will be in the skin. 3. Extreme poses were not included and they ... need patience, if they even get successful. 4. Waistband can be confused in sweatpants and you may see a zipper, just keep generating, or add "pull the sweatpants down. The pants have a rubber waistband and a cord." Use the LORA with weights 1.0-1.3 I have seen better results with euler_ancestral and 8 steps, but euler and 4 steps is also ok. If you love the concept, let me know, so I can work more on the project. I would love your constructive feedback and I am happy to help as much as I am able to. LORA training is new to me and I waisted days just to have this result. But I have learned a lot. Flux2-klein training is a new territory and there not many experience in the community. I would love to se your creations ... the successful ones ... LOL

Files with SHA256: 26b9622ca6866e406a4a368d00ddc61295285c83bc54802a00b2d3d91790c9c1