I think I got confused because you displayed the discriminator's accuracy separated into real_accuracy and fake_accuracy. That's good!

I think if you plot the accuracy with each epoch, you will see mode collapse like here: https://machinelearningmastery.com/wp-content/uploads/2019/07/Line-Plots-of-Loss-and-Accuracy-for-a-Generative-Adversarial-Network-with-Mode-Collapse.png

(that would explain the perfect Discriminator's accuracy for fake images at the end and why your model only generates the similar type of bonsai tree image then)