Now we feed this generated image to the discriminator, which returns the probability of the image being real. The discriminator loss is given as follows:

Here:

• implies the real image, , conditional on the text description,
• implies the generated fake image