Macys Flower Show 2025
Macy's
For Macy’s annual Flower Show, a series of 10 animated vignettes were created using a hybrid workflow combining Unreal Engine and AI video generation. Each scene began as a dreamlike floral environment built and lit in Unreal, serving as the visual base for AI-driven animations. By exporting multiple unreal scene render passes, the team enabled granular AI control over composition, scale, and mood. The installation was on display for three weeks, drawing heavy foot traffic and showcasing a new direction in experiential content creation.
Project Breakdown
This project was a milestone for us. The Macy’s Flower Show was the first commercial project our team did that fully used gen AI as the production pipeline. Not as a helper or a concept tool, but as the actual rendering engine from start to finish.
The brief was to create a series of cute, stylized vignettes that also had to be animatable. So not just static images, but semi-consistent scenes that could hold up through motion. On top of that, the project went through multiple rounds of client R&D, revisions, and phases. If you’ve worked with gen AI in production, you know that kind of iterative back-and-forth was basically a nightmare for these tools at the time.
We started with closed models, and it quickly became clear that the lack of control was going to be a bottleneck. Getting a nice image once is easy. Getting a consistent, directable output across dozens of iterations with client feedback is a completely different challenge.
For me as a technical artist, this project laid the groundwork for how I started thinking about gen AI in future work. It taught me that controllability isn’t a nice-to-have, it’s the whole game. It pushed me to develop in-painting intuition and to start thinking about the image as patches rather than pixels, which ended up informing a lot of the workflows I built after this.
We also used Unreal Engine on this project. A few of us built out the scenes in Unreal and generated depth maps that fed into the earlier iterations of the pipeline. That connection between 3D ground truth and gen AI output was something I kept coming back to in later projects.
