Trained from 0.14 loss down to 0.06 on the full un-pruned SD 1.5 checkpoint.

Images tested with juggernaut used 0.5 Lora setting

Up-scaled images where based off of text to image and where up-scaled at 0.5 de-noise with a lora value of 0.33-0.5 (Native not highrez fix)

The full text to images where left as example seeds but would improve quality drastically by up-scaling image to image with the settings above.