READ EACH VERSION DESCRIPTION
Recommended settings:
1024 x 1024 resolution (most consistent)
Juggernaut XL (or similar photorealistic model)
Use Natural Language prompts
0.8 Strength
No refiner
DPM++ 2M Karras
>30 steps
Note: upscaling may degrade film grain
Trained on many high-resolution 35mm and medium format film images
The produced film grain may look bad if files are stored as .jpg, so .png is recommended
Most training images were shot with Portra, Ektar, or Superia X-TRA filmstock
Description
Version 1.0 is finally here! I have worked very hard to ensure that I trained on high-quality, well captioned images, but I still am not an expert! If anyone has any SDXL LORA Training tips, let me know!!!
I am trying to reduce the bias in my model, but it's been very difficult (for example, asking for a "handyman" will produce mostly white men).