Anonymization of PII data in documents using diffusion models - I'm in the process of reproducing academic papers. The idea is you can replace sensitive information from financial/medical documents with synthetic analogues without visually altering them, so they can be kept/used for AI training