Conceptual

Program Generation for Automated Slide Design from Natural Language

An approach to automated presentation-slide generation in which a language model translates a natural-language instruction into an executable Python (python-pptx) program that, when run, produces an editable slide. Compared to end-to-end image generation, program generation gives precise control over text, images, layout, and color and yields user-editable output. Supported by a high-level slide API and by iterative self-refinement in which the model revises its own code from a rendered slide snapshot.