Results are very distorted - what am I doing wrong?

Are there some flags I should set with the executable (py) to generate quality similar to the examples in your paper?
Finally set everything up to work error-free on a Lambda Labs server powered by A100 GPU and the output is worse than FOMM using square vid/png files..

I would appreciate any and all help please!
Thank you