Technical Viability of Text-to-Video Model for Specific Applications

I have certain questions regarding technical viability of this "Text-to-Video" models

1. Is it possible to fine-tune this model on "Driver Monitoring Dataset" ? 
2. Can it be used to generate longer videos (lets say 2-3 mins)
3. Is the model able to generate synthetic data?          
4. Model Used - Make-A-Video SD JAX   

@lopho ?