FinalCut 4k Fantasy
This FP32 training is a full custom clip with a full FP32 finetune.
High natural language recognition.
Great background details with realistic placment
Fine Detail at 4k with up-scaling image to image workflow
Limited Text recognition slightly below SDXL FP32 base
So when attempting to train a T5xxl at FP32 and ending up with a mostly lobotomized model after a month, CPU training as GPU would have cost 2-4 million dollars I decided to release the model that was CLIP-G and CLIP-L instead