How discriminator determine whice one is real image and whice one is fake?

Question

How we train the discriminator actually ? It wants 0 and we pass both real and fake image together , the real image will closer to 1 so the loss will high . So does it mean