I cannot stress this enough: this is not possible
in a literal technological sense: you cannot train an AI and expect even comprehensible results with under ~1 million pieces of data, nobody can supply all that with the variance needed. You can only achieve it through wanton scraping