Every day, billions of ad impressions are seen across Facebook, Instagram, and other Meta platforms. Each ad contains an image or video asset chosen by one of the over 10 million advertisers on the platform.

Media is uploaded, selected from a media library, and then customized for each impression. Media is cropped for specific aspect ratios of different ad placements, given custom text properties, and more.

This project was a fundamental shift in how advertisers interact with the media before publishing ads to their audiences, shifting from single image or video model to a flexible media model.

The problem

Meta’s ads ecosystem was evolving toward a more dynamic future. Machine learning systems could identify high-performing creative combinations, but media customization flows treated each creative asset independently. Advertisers were required to manually edit assets one by one, and duplicate effort across multiple ads and placements. They were forced to think in terms of static outputs, which isn’t how the ad delivery system works.

As advertisers uploaded more assets, the workflow became increasingly difficult to manage. The system could not fully leverage performance signals across a broader creative pool, and advertisers optimized individual assets manually, while delivery systems optimized performance automatically behind the scenes.

The interaction model wasn’t built for future AI-assisted optimization systems, and introduced friction precisely where automation was becoming most valuable.

The opportunity

The mental model shifted from “edit this image” to “add creative ingredients the system can use to create new recipes”. 

This required re-thinking everything about media interactions in ad creation flows.

  • Media selection and uploading
  • Creative customization
  • Media > placement mapping
  • Balancing control and automation

Because advertisers could now upload and manage pools of media assets, all in a single ad, the system could more effectively select dynamic media variants and combinations of creative assets.

Instead of manually defining every output, advertisers could enable the system to continuously identify higher-performing creative combinations.

Design goals

Reduce workload complexity

Enable advertisers to manage multiple assets without exponentially increasing customization effort.

Preserve advertiser trust

Automation needed to feel understandable and controllable, not opaque or complex.

Improve creative performance

Enable the delivery systems to identify more performant combinations of media, formats, and placements.

Future-proof creative editing

Build a foundation that could support the creative customization options of increasingly adaptive ad systems.

The new model for customizing ad media

Outcomes

For advertisers

Enhanced media customization workflows

The transition from single-media editing to a multi-media customization system reduced operational friction across ad creation workflows. Advertisers could manage pools of creative assets instead of manually editing individual images and videos across placements, creating a more scalable and flexible workflow for a modern campaign creation that mirrors how the system uses creative assets.

For Meta

System optimization enablement

The redesigned architecture enabled Ads Manager to dynamically optimize creative delivery using pools of advertiser-provided media assets. This allowed the system to identify higher-performing combinations across audiences, placements, and formats, establishing foundational infrastructure for more adaptive, machine-learning-driven ad optimization systems.