LORA trained on 35 publicly available images of Margot Robbie.
Trained on base SD 1.5, should work with most models.
This model comes with an optional trigger word h4rl3yqto replicate Harley Quinn from Suicide Squad.
Use ADetailer if details are lost and strength of 0.8 should work on most prompts.
Consider leaving a review if you like the model. All feedback is appreciated.
Description
FAQ
Comments (1)
friend, i have finally gotten around to testing your model, i'm uploading a few samples to show you: it's overfitting. I think you need a version with at least 200 steps less to make this model work like you intend it to.
Notice what it does to the background. Or how it disobeys my prompt.
All key features of a model with either too few pictures or too many steps, but given how it distorts the background, I think it's overfit. At lesat 200 steps less, if not more.
it shouldn't be necessary to use a lora at a lower strength than 1, that will make it worse, always, because at 1 it does everything it's supposed to, lower, it stops before. but at 1, it destroys the image.