FYI: Model seems to perform better with the following configuration on the generator

´´´#declare shape of the noise input
random_normal_dimensions = 2048
#build the generator model
generator = keras.models.Sequential([
keras.layers.Dense(1024, activation=“selu”, input_shape=[random_normal_dimensions]),
keras.layers.Dense(512, activation=“selu”),
keras.layers.Dense(28 * 28, activation=“sigmoid”),
keras.layers.Reshape([28, 28])
])
´´´

1 Like

You need to provide more context like better in terms of generating images or lower loss or stability during training.