Experimental Pony training using Full FP8 and Adapt Lion.

1. In my testing 0.75-0.85 is the best lora value to use.

2. The Data Set was only landscape photos so the Lora is SFW possible.

3. Trained against this model - Works with some realistic models and not others.