Can an AI Do Balenciaga Better Than Balenciaga?
By Julia Krantz, Magic Fabric
This post was originally published by Julia Krantz on Magic Fabric, a blog about virtual fashion and 3D technologies.
Will designers be replaced by algorithms in the future of fashion?
The thought of a fully automized society where robots and AI take over simple chores is not a futuristic utopia. It is, in fact, becoming more of a reality. But can machines also do more complex work that requires both imagination and artistic thinking?
Robbie Barrat, artist, and AI researcher, plays with the idea in his project, where he trains a neural network with image data from five different fashion shows from the luxury fashion house Balenciaga. The result is videos generating new models with outfits based on the training data. The garments, shapes, and textures might not be directly translatable into new pieces. Yet it shows on Artificial Intelligence’s ability to create combinations, shapes, and textures a human eye might never have come up with.
To date, Robbie does not seem to have gotten himself recruited by the ancient fashion house. However, he clearly offers radical perspectives on how artists and designers might be able to make use of AI as a tool in their daily work to innovate their creative work process.
The project in itself evokes curiosity around how similar approaches could shake the fashion industry to its core. Would we accept paying large numbers for designer products completely generated from datasets? Or will the touch of a human become a stamp of quality in the near future?
Follow Robbie’s work process on Twitter
Also check out Julia’s post on Magic Fabric, Machine Learning – the future of Design?
The high res video generation network has started training! Still lots of artifacts...— Robbie Barrat (@DrBeef_) August 29, 2018
Not sure what part of Balenciaga's outfits the AI is getting this from, but right now it really loves to make what look like flickering neon scarves/turtlenecks - I want one in real life 🧣✨ pic.twitter.com/HRNpff4btL