Generative Algorithm Based Digital Runway


How might fashion designers collaborate with an AI?

This creative technology project explored a future scenario in which fashion collections are generated via algorithmic selection. Designers would feed quantitative preferences into software, which would interpolate those choices into a fashion collection.

[Given the amount of machine learning algorithms running behind the scenes in online retail, this idea is really not so far fetched.]

To maintain the artistic spirit of the project, the creative direction called for more abstract, geometric designs that had recognizable clothing qualities but did not try to be real clothing.

The main design challenge was figuring out how designers would create designs. And, given the art direction and the technical limitations in a short timeline, how we could create an ‘algorithm’ that creates sufficient range of variation.

I organized the design experience into 4 steps:

  1. Creating the cut of the garments by manipulating sliders that control the shape.
  2. Embellishing the garments with geometric shapes that can be placed across the surface.
  3. Choosing materials and uploading patterns.
  4. Selecting looks that are generated by the Rose algorithm.

ROSE is written as a basic genetic algorithm. It creates 18 looks at each generation from the features that have been selected and the two looks that were chosen in the previous generation. The user progresses through 5 generations, and the looks generated become gradually more extreme.

Some generated collections are below.

While this was largely an artistic project and ROSE was purely conceptual, one does have to wonder– as product algorithms become more commonplace and robust, what becomes of the role of the designer? Could a computer design things better matching to individual consumer tastes and shopping habits?