TRIGGER WORD: pd_style
This is my first LoRa ever created. I have no experience in LoRa-training at all and used Claude to help me.
Will make more LoRas in the future (specially specific styles that I either can't find on civitai, when I'm not satisfied with the already existing options or when there are requests in the comments, so feel free to make suggestions)
For further Information look at the specific version info ("About this version" drop-down on the right side)
Please leave some reviews and suggestions in comments so i can make a better version.
Will be removed if requested.
Description
This version is somewhat more stable and gives better results (used 2000 steps instead of 1500 and train_batch_size = 3 instead of 1). Some outcomes may still be disorted and I recommend to use some kind of finger-improvement like a LoRa for fingers when using this LoRa (and maybe some face detailler).
Due to the low number of valid training images available online, the LoRa is still slightly overfitting, so use LoRa strength between 0.7 and 0.9.
I will probably make even more and better versions and maybe tweak the captions of the training-images i used (because right now, i just used an automatic tagger lol), but only if i feel motivated to do so (could be tomorrow, could be in a week, could be never, depends on engagement and feedback with this LoRa i guess)
Please leave some feedback and post images with this model so I can figure out on what to improve! :D