How to create an autoencoder model where the input shape = output shape

Question

I am working on an autoencoder model. My dataset contains images of shape (256,256,4). I changed them to grayescale before inputting them into the model while training. My label