Generating realistic looking human face images aligned with natural language descriptions using modified version of NVidia's StyleGAN.
None of these people are real. They are generated by network according to given descriptions.
You can find more detailed explanations in the presentation under "presentation" folder.
Some examples with descriptions. Changes in descriptions are marked as red. Note how generated image changes according to given description.
More than one image can be generated with given description by changing noise in intermediate layers:
