Theano implementations of two generative models for MNIST:

• An adversarial autoencoder (Makhzani et al.)
• A generative adversarial network (Goodfellow et al.)

Both perform reasonably, but not particularly well. To run the code, you'll need to install Lasagne, as well as this small Theano library I wrote.

Samples from the adversarial autoencoder:

Samples from the GAN: