The open-source initiative, Editable Design, is committed to tackling a key challenge in the design realm: while AI-generated images boast visual allure, they often fall short in terms of modifiability. Conversely, designs crafted through code offer flexibility but may lack aesthetic finesse. Editable Design bridges this gap by employing visual models to guide design direction, structured code for precise expression, and an Agent to seamlessly orchestrate the interplay between the two.

The project's workflow begins with harnessing visual models to establish design guidelines. Subsequently, it generates standalone assets, reconstructs the layout using HTML, and populates it with authentic text. Following several rounds of refinement, it yields an editable design file. This empowers users to effortlessly tweak text, reposition layers, and even replay the design process for deterministic verification.

Presently, the project showcases 14 comprehensive case studies, encompassing a diverse array of visual design tasks. This marks a significant leap forward in AI design, transitioning it from mere pixel-based output to a structured delivery phase that seamlessly integrates with subsequent production stages.