Image Generation can be done in different ways. These are the simple examples neural networks to do the following:

Dataset : MNIST Dataset

AutoEnoder - Image ---> Latent ----> reconstructed Image

Variational Auto Encoder- (Reviewing)

1. Using Xavier Initialization.
2. Brief theory source - http://kvfrans.com/variational-autoencoders-explained/

GAN (Generative Adversarial Networks ) - (Working)

1. Using Xavier Initialization.