This AI Paper from Harvard and Meta Unveils the Challenges and Innovations in Developing Multi-Modal Text-to-Image and Text-to-Video Generative AI Models