Training modern text-to-image (T2I) models often feels inaccessible, overshadowed by the perception that it requires massive infrastructure and highly complex engineering pipelines. We wanted to explore the opposite direction: how far can we go with a deliberately simple recipe…