smaller dataset, trained with artists, blob and tubefeeding. All artists upscaled and dataset pruned heavily for high quality outoput. Trained at 640x640. Use clip skip 2 and hiresfix for best output.