Our Problem:
Banner production is largely manual. Each size, brand, and language variation requires hands-on design work. At the volume Gen Digital operates, this was creating bottlenecks, inconsistencies, and slow turnaround on deliverables.
Our Idea:
Build an agentic banner-generation tool with Claude Code that renders banner designs in multiple formats and languages from a single brief.
To achieve this, I started by training Claude across various sections of the Norton brand guideline and libraries of past marketing templates. After some of trial and error and way too much coffee, Claude was able to read a single project brief and generate banners across multiple aspect ratios and languages at once, intelligently crop photos of people depending on the banner size, and run safety checks for legal, brand, and general design compliance standards.
For feedback, I had Claude build a simple plug-in that pushes the designs to Figma, where teams can review and give comments and have Claude address their feedback—all without touching the layer panel.
Example comments:
> "Move 'campaign,' to the first line of the headline."
> "Let's zoom out a tad on this photo to show more of the phone in frame."
> "Move 'campaign,' to the first line of the headline."
> "Let's zoom out a tad on this photo to show more of the phone in frame."
Once Claude addresses the feedback, it builds out all the exports in an organized folder structure, ready for shipping to the client in all specified delivery formats (JPG and PNG at 1x and 2x size, zipped HTML5).
Final thoughts: The time it took to reach a first pass on designs dropped from a couple hours to a few minutes — sometimes even seconds. Wild.
Final thoughts: The time it took to reach a first pass on designs dropped from a couple hours to a few minutes — sometimes even seconds. Wild.