Reading trends with Artificial Intelligence

On March 10, the students of the Mastere Spécialisé in Management presented their reading of the Fall-Winter 2023 women's ready-to-wear season as part of the annual "Fashion Week" workshop. Coordinated by Benjamin Simmenauer, the exercise was divided into two parts: the analysis of the fashion show of a brand that was assigned to them; the transcription of the week's "trends" using Artificial Intelligence.

A new way to represent trends

While trend analysis often crystallizes around the more or less precise census of clothing forms perceived during fashion weeks, the challenge was to renew the exercise by calling upon an Artificial Intelligence generating images such as DALL-E2, Midjourney or Stable Diffusion. After identifying the key trends of the week, the students were asked to create an image that would represent them by writing a "prompt", a textual description guiding the AI.

Promising results

The results are promising. Seven images were finally generated under the direction of Denis Bonnay (lecturer in philosophy, University of Paris Nanterre). They allow us to identify quite accurately the formal properties of the week, between formal wardrobe with sometimes disproportionate shoulders (Saint Laurent, Balmain), transparency and nudity (Ann Demeulemeester, Givenchy) and blackness (Dior, Alexander McQueen).

The critical analysis of these productions (carried out by the students at the end of the exercise) shows that each silhouette reflects both:

1/ the biases of human observation and interpretation: the groups freely focused on the aspects of the fashion shows that seemed most salient to them, which explains some of the differences between the images generated

2/ the capacities and limits of the algorithms used, which need to be tamed to be used in a truly efficient way: in some cases, the prompts (instructions) required several dozen iterations before generating an image that was deemed satisfactory.

Image-generating AI is not the only possible application of AI in fashion. Their ability to generate rather "realistic" silhouettes already raises different questions: can a non-creative person now be a designer and think up a collection with the help of prompts? How can we still distinguish a "real" silhouette from a "fake" one? Can AI help generate new shapes? These are just some of the questions that the students of the Master's program were able to confront during the Fashion Week workshop.