Abstract: Embodiments of the present disclosure provide techniques for refining visual assets using generative artificial intelligence models. An example method generally includes receiving an indication of a visual asset for modification and an input natural language prompt specifying a design objective to be applied to the indicated visual asset. Visual components in the indicated visual asset and properties of the identified visual components to be modified are identified based on the specified design objective. One or more control panels are populated with one or more controls for modifying the identified visual components to achieve the specified design objective. An input is received in at least one of the one or more control panels, and the indicated visual asset is modified based on the received input.

Abstract: Embodiments of the present disclosure provide techniques for generating candidate visual assets using generative artificial intelligence models and a selection of a visual asset based on which the candidate visual assets are generated. An example method generally includes receiving an indication of a visual asset for modification and an input natural language prompt specifying a design objective to be applied to the indicated visual asset. The design objective is translated to a plurality of transforms to be applied to the indicated visual asset. A candidate set of visual assets is generated based on applying the plurality of transforms to the indicated visual asset, and the candidate set of visual assets are output for selection in a user interface.