One Model to Edit Them All: Free-Form Text-Driven Image Manipulation with Semantic Modulations.