What is a good approach for an autoencoder on video data?

Question

I built an autoencoder for image data. Following a simplified version:

    wx = 28
    hx = 28
    n_channel = 3
    latent_dim = 10
    n_filter = 16

    x_in =