Nike — AI R&D
Made
Speculative visuals and motion concepts built around Nike as the subject — self-initiated R&D, not commissioned work.
Needed
A look at how generative AI could speed up ideation and give future campaigns repeatable creative pipelines.
Worth
A tested mix of node-based AI, 3D and prompt engineering, plus a body of concept visuals to pitch the approach with.
The problem
Brand storytelling still runs on long production cycles. The research asked how AI could accelerate ideation while keeping the art direction recognizably Nike.
What I built
A pipeline that combined node-based generation in ComfyUI, Adobe Firefly and 3D in Blender, so concepts could move from sketch to motion test quickly.
How it was made
The motion is a soft body simulation in Blender, blocked out so the movement stays mine rather than something a model guesses at.
The depth map comes out of Blender itself — z-depth and mist off the render layers, not depth a preprocessor guessed back out of a finished frame. It carries geometry and animation and no look at all, which is exactly what Wan 2.0 wants as a reference to follow. ComfyUI is the graph that drives it there.
The look arrives separately, as a style frame transferred on top. One thing has to line up: the first frame of the depth map and the style frame have to be the same shot. When those two drift apart, the model starts inventing.
My part
I led the research: the concept, the creative direction, the Blender simulation and its depth passes, and the style frames the look was transferred from.
3D exploration
Raw passes, straight out of Blender. The middle one is the file itself.
Style references
Stills made in Firefly, Nano Banana and Flux to find a look before any of it moved. Open any one full size.
Made with
- Tools
- Blender
- ComfyUI
- Wan 2.0
- Adobe Firefly
- Nano Banana
- Flux
- Art direction & AI
- Artem Maslov