This model doesn't require face restoration, all the images you see above do not use it. I would recommend changing the LORA weighting between 0.7 and 1.3 depending on the distance away from the camera and how close you want it to be to the original images.
I got people saying that my previous model wasn't flexible enough (even though it was) so I had to create generations with different concepts. Here I have included some different prompts that show how flexible this LORA can be.
This LORA was trained with 1500 steps. 15 images, each with 100 steps. I think I used 1 Epoch. I trained on default 1.5 ema-only.
ALL the images you see above were using the default 1.5 ema-only model, nothing special. So you all got no excuses for bad results XD. Feel free to use with other 1.5 models, im sure they will give fantastic results too.
For any private work, please feel free to contact me for more information:
Follow for more...