DefinePK

Abstract

This work is focused on the prospects of the revolution of Generative Adversarial Networks (GANs) in the current fashion design, accelerated by the duality between the automation of algorithms and the creativity of humans. As GANs have proven themselves technologically able to produce unique designs, the use of them in the fashion industry begs many questions regarding copyright and intellectual property rights, as well as the existence of human designers.
We present a dual-phase framework:

GAN Training & Design Generation:

Curate a dataset of 50,000+ haute couture and streetwear designs (from 2010–2023) to train a conditional DCGAN (Deep Convolutional GAN).
Implement style-transfer techniques to blend distinct fashion eras (e.g., Baroque embellishments with futuristic minimalism).

Human-AI Collaboration Evaluation:

Conduct workshops with 30 professional designers to assess:

Aesthetic quality of GAN outputs via Likert-scale surveys.
Usability in real-world workflows (e.g., pattern drafting, fabric selection).

Compare GAN-generated designs against human-created ones in blind consumer preference tests (n=500).


Key findings reveal:

65% of GAN outputs were rated "professionally viable" by designers, particularly for ideation phases.
Consumer bias: Human-made designs preferred for "emotional resonance" (p < 0.01), while GAN designs excelled in "novelty" (p < 0.05).
Ethical flashpoints: 78% of designers expressed concerns about job displacement, underscoring the need for hybrid creativity models.

This research contributes to:

Fashion technology: Demonstrates GANs' capacity to augment—not replace—human creativity, with applications in sustainable fast fashion.
AI ethics: Proposes attribution frameworks for AI-assisted designs, addressing IP ambiguities.

Design theory: Reconciles algorithmic randomness with intentionality in creative processes.