trained to make smaller logos/icons from a smaller dataset for v1, results have been good for prompts that don't get too complex.