R&D

Experiments from my role as Head of Creative AI: systems thinking applied to creative problems.

R&D 01

F1: Image-to-Video

Can image-to-video AI hold character, hyperrealism and brand accuracy shot after shot?

Dense storyboarding gave the model keyframes instead of guesswork; Nano Banana Pro handled the images, Seedance the motion, and the results were cut like any other film. The full piece lives in Work.

F1 storyboard frame F1 generated frame F1 final shot

Rule: eliminate the guesswork.

R&D 02

Industry Character Photoshoots

How do you photograph workers from a dozen industries without a dozen photoshoots, and without it feeling like stock?

A workflow for generating industry-specific characters and carrying them through multi-shot, hyperrealistic photoshoots with Nano Banana Pro: consistent faces, wardrobe and light across every frame.

Generated character, shot 1 Generated character, shot 2 Generated character, shot 3

R&D 03

W4: Mixed-Media Launch Edit

How far can one edit stretch when traditional footage and AI-generated frames share the same timeline?

A product launch built from interviews, user photography animated as stop-motion cutouts, AI-generated sequences and motion graphics, with AI voice filling the narrative gaps. Production time dropped from 2 weeks to 2 days. The full film lives here.

W4 campaign frame W4 stop-motion cutout W4 final frame

R&D 04

Financial Education Illustration System

How do you illustrate a database of financial concepts accurately, consistently, and in any partner's brand colours?

An illustration grammar of one hero object plus supporting elements per concept, generated at scale, then handed to a product designer who batch-converted flat images to vectors and mapped colours to partner brands through the Figma MCP. I designed the pipeline and the creative; the vector automation was the designer's build.

Financial concept illustration Brand colour mapping

R&D 05

The Content Library

What happens to years of content nobody can find?

Every internal asset processed through AI image analysis into a database, stored and hyperlinked through Google Drive via the Workspace API, and served through a vibe-coded front end anyone in the company can search. Underused content became self-serve infrastructure: 500+ assets and counting.