Coca-Cola
LATAM AI PRoduction
The Coca-Cola LATAM team had a unique problem where they already had photo shoots of a variety of talents, and they wanted to use AI to create whole new scenes featuring those talents. They tasked us to use these assets to create brand new scenarios without a actual photoshoot.
The project used a custom AI system built on Nano Banana to iteratively create whole new scenes.
As Senior Director of Technology & AI, it was my role to build the production application, generate half the content and lead a team of creatives in crafting a wide range of new assets for the client.
Process
Generative AI had come a long way since my previous AI production project for Marriott. Whereas it used to be that you had to be a creative technologist to work something like ComfyUI, Nano Banana was really easy to use, allowing creatives to create quick comps of each shot.
It was up to me and the team to convert the compose into high fidelity outputs ready of social.
The process is designed to be iterative so that we can slowly build up an image, and incorporate client feedback as it comes.
The pipeline featured a custom Web Application, 3D product renders, and Photoshop.
Creative comps where converted to ambient occlusion maps as a high quality 4k base. This AO map was used as input, along with talent face images, wardrobe inputs, and food input to create the desired composition.
3D renders of client approved 3D models where composited onto the resulting shot, which was fed to the AI for a round of compositing, and liquid / condensation rendering.
REGENERATED IMAGE
CREATIVE COMP
Bespoke Web Application
While there are many different generative AI platforms for advertising out there, they are always missing something, or they are slow to update to the latest technologies.
As a creative technologist, with the help of AI assisted coding, I was able to create a bespoke web application for this project that was tailored to our process, the assets we received, and the type of feedback the clients gave.
The interface was designed for modern editing based models, specifically, Nano Banana. These models can take many images as input, as well as specific visual and text based instructions.
The application had the following features: Image Manipulation, Painting, Layers, Prompt Enhancement, and Iterative generation.
