Stefano Corazza tells Sam Charrington why AI design needs fast iteration and precise control. He describes a pipeline that expands a prompt, builds a structured scene description and renders pixels, with opportunities for human edits along the way.
Stefano Corazza discusses training data, synthetic examples and the limitations of benchmarks focused on one-shot generation. He describes using structured rubrics and human evaluation to assess design qualityEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria., while noting areas where the model needed improvement.
Stefano Corazza explains how a design agent uses tools to edit layered compositionsAgent tool use lets an agent call bounded external capabilities so it can retrieve information or change state beyond generating a response. and describes inference optimizations across model and infrastructure. Product performance and cost claims reflect the vendor’s account.
Watch on YouTube
