This is a style lora, It was created with a synthetic curated blend of images, meticulously hand picked. Things to note:<ul><li>It was created mostly for NSFW stuff, you can still use it for SFW it should have enough training data but it's not what is specializes in.</li><li>Only MSRD is necessary for the style, Trained at 1024x1024.</li><li>It does seem to favor woman (all that good nsfw stuff), no humans seems to work well.</li><li>I use at 1.0 and 30 steps, 5CFG for version 4.2 to start with.</li><li>Let me know if you notice anything or think something needs improved on.</li></ul>V4.2 notes:<ul><li>Eyes are much better added different kinds, style is more prominent</li><li>Slightly overbaked so the style would be stronger, very occasional disfigurations sometimes occur</li><li>Style is even more prominent on illustrious, trained on v0.1 for flexibility</li><li>Play around with the weight for it, I was having fun all across the board even all the way up to 2.0</li></ul>V3 notes:<ul><li>Eye issues seems to be better, if you are still having issues try adding an eye prompt to your positive prompt. Exp: red eyes, blue eyes, green eyes, glowing eyes etc..</li><li>Lower CFG is king, it'll allow this lora to do it's job better, lower CFG = better lora/pictures</li><li>For whatever reason, you may need to add monochrome to your negative prompt now, standing too if you get sick of generic standing poses</li></ul>V2 notes:<ul><li>This is an earlier version that isn't cooked as harshly as the other one. I added it as a temp fix for people to see which works better for them for now till I can make another model.</li><li>Also for some reason it doesn't seem to be super compatible with every checkpoint, even though it was trained on Illustrious, not sure why there, will also try to fix later.</li></ul>

MSRD Styl3

Added some more eye data, and other data, changed optimizer, changed back to network 64

Files with SHA256: 6057d17ceb645d0df794b721c466e79c0e0bba4d8c4206f7a5486e98142ac7de