Our new architecture analyzes and decomposes images into a code-like intermediate representation called layouts — an internal visual language that captures the composition and structure of any image.
Our intermediate image representation is designed for transparency and control. Rather than hiding the model's understanding behind a black box, we expose its internal representation, or code, to enable direct manipulation of visual elements. Users can now move, resize, add, remove or replace objects with granular control.
Our intermediate image representation is designed for transparency and control. Rather than hiding the model's understanding behind a black box, we expose its internal representation, or code, to enable direct manipulation of visual elements. Users can now move, resize, add, remove or replace objects with granular control.