FOR PRIVATE USE ONLY. This resource is intended to reproduce the likeness of a real person. Out of respect for this individual, only work-safe images and non-commercial use is permitted.
This model contains photorealistic detailes and likeness of 90% + trained with most recent pics + basic face expressions.
This is one of my first flux models, I will update and improve the model slow by slow
recommended base model: https://huggingface.co/black-forest-labs/FLUX.1-dev/tree/main
Trigger word (optional): j0di
Description
FAQ
Comments (4)
What is up with the clip skip of 2? Isn't it always possible to train a lora for good results at clip skip 1? This is the first time I see a lora using this.
Maybe Fuslie next? Or the other offlinetv girls
What you do with ControlNet to get better results? Im genuinely curious
I've never had a lora fight me so hard. I have a standard prompt I use for celebrity loras. I had to delete 95% of the prompt, had to then move both trigger words to the front of what was left, turn off all over loras, and turn this lora all the way up just to get the lora to start making a face that looks like a person, and even now the results aren't great. I even tried switching to the checkpoint recommended. Maybe I'm doing something wrong but I've used dozens of loras now with the same setup and this is the only one that's fought me this hard.