In wek 2 vid2, it shows 3 graphs and third one with a perfect model is called high variance, why is that? is 0% error bad?

It depends. If you get 0% error on your training dataset, then the question is what accuracy do you get on the test data? If the test error rate is also 0%, then you’ve got the best possible scenario. If it’s 10% or 20%, then you’ve got an overfitting problem.