Zalando uses machine learning to explore the potential of algorithms in clothing design. The project is an experiment to built a neural network that can create virtual fashion, which can then be produced in real life and eventually purchased by customers. The project was present at Bread&Butter, a German based fashion event.
Google’s TensorFlow, an open-source platform for machine learning, is the basis of the experiment. Furthermore, Project Muze can be described as a neural network -meaning that it is a computer system or algorithm modeled on the human brain and nervous system. The software was attributed a set of aesthetic parameters and exposed and trained to preferences such as colors, textures and styles from more than 600 fashion “trendsetters”. In addition, it is equipped with data from Google Fashion Trend Reports and data about trends on the Zalando webpage itself. The machine-learning algorithm is then supposed to connect preferences to people with similar interests and create designs accordingly.
I create multiple designs on projectmuze.com:
At the beginning I was asked some personal question, for example it asked me about my favorite art period and mood or what is my favorite type of music. After that I was ask to enter my age and could then draw a design on a mannequin. What first seams very intriguing soon became disappointing. The designs were somewhat strange and the logic behind questionable. For example, it is possible to answer even more question after creating your first piece to change the color and pattern to match your individual taste. After having entered blue as my favorite color, my former designed was then changed to the color green?! After answering another question about my spirit animal the color of the design changed again to a bright purple/pink. The software does not filter all your preferences to create one piece but rather adapts it after each question to a new preference.
Summarising, the software gives a very interesting view on how fashion design might look like in the future. However, at this point the software is still not feasible and far from creating real world wearable pieces.
References:
Zalando SE. 2016. Project Muze. [ONLINE] Available at: https://projectmuze.com/en. [Accessed 25 September 2016].
Paul Sawers. 2016. Google and Zalando launch Project Muze, a machine-learning experiment for 3D fashion design. [ONLINE] Available at: http://venturebeat.com/2016/09/02/google-and-zalando-launch-project-muze-a-machine-learning-experiment-for-3d-fashion-design/. [Accessed 25 September 2016].
Sarah Perez. 2016. Google’s new Project Muze proves machines aren’t that great at fashion design. [ONLINE] Available at: https://techcrunch.com/2016/09/02/googles-new-project-muse-proves-machines-arent-that-great-at-fashion-design/. [Accessed 25 September 2016].
Dear Anna Lena, thank you for your interesting blog! I did not know Google and Zalando were currently launching this experiment, and I think it is definitely something that will be developed and used a lot more in the future.
Your blog got me curious and I decided as well to create a design on projectmuze.com – the questions seemed rather odd to me to create a design, and the design shown was not at all what I had in mind, but it was quite interesting to see how the algorithm made a design with just a few simple questions and a drawing. I will definitely try again once it is a bit more developed 🙂
Hi Lena,
Likewise, thank you so much for posting your blog! I agree with Ananda and have a similar position on this topic. The project seems like an idea that can be very smart and viable in the future, if developed correctly by Google and Zalando. 🙂
Both companies are very data-driven and will become even more so in the upcoming years. Seeking insights into user behaviour and preferences to create future business value, has become part of the daily routine of Google and Zalando. As such, working on Project Muze together seems like a logical step forward. Aiming to use this data and information to create new outfits and designs.
After having gone through the UX of creating a design on projectmuze.com, I fully agree with both Ananda & Lena. The questions to create the outfit were rather abstract and the ultimate design did not really correlate with the mood and the answers that i gave. Instead, it displayed an outfit that was quite abstruse and different to what you would expect to find on any current online shop.
Although the algorithm might still be in the starting steps, I believe that Project Muze is valuable for both Google and Zalando. It builds a fundament for the fashion industry to digitally create future outfits and designs through user data.
Hi Lena,
great blog post! Thanks a lot for sharing it with us.
In my opinion, we are living in “algorithm-for-everything” bubble. Many companies are trying to employ different set of data patterns to reveal the way people think. As in some ways it works really well (for instance administrative work or navigation systems) in some it may not be the best idea.
I think the Project Muze currently falls into the second category. Fashion, coming from arts, is hard to quantify, designing process could be compared to balancing on the fine line between practicality and aesthetic value.
After looking into the topic, I discovered there was similar try to the Project Muze. Researchers from the University of Toronto created an algorithm that tells you how “fashionable” you look in your outfit by comparing it against the range of “fashionability factors”. The insights reach further than just judging your outfit – as the researchers put it: “The garment itself being fashionable is also not a perfect indicator of someone’s fashionability as people typically also judge how well the garments align with someone’s ‘look’, body characteristics, or even personality.” (Wired, 2015)
The question remains, how can you say if something is fashionable or not? With the diversity in our beautiful human kind, tastes and preferences vary.
I would like to remind you wise words from the times of ancient Rome “de gustibus non est disputandum”. 🙂
Hi Lena,
Thank you for this post; as a frequent visitor of online shopping platforms such as Zalando I found it really interesting to learn about this project. I tried it out myself, and although most of the questions asked at the start were not directly connected to fashion design, I believe that it is an interesting way for companies like Zalando to learn more about consumer preferences. In the long term, the concept of machine learning in fashion design may help Zalando to offer even better products to its customers.
I spent some time looking up other designs people created, and what was particularly noticeable across all these designs was their lack of ‘wearability’. I assume that this is partly due to the algorithm still being relatively new, but also because the concept of what is ‘beautiful’ in fashion design is highly subjective and constantly changing. There were odd color combinations, and strange textures and objects that sometimes did not even seem to be attached to the clothing itself. As the algorithm improves over time, I believe that both Zalando and its customers will be able to reap great value from being able to directly incorporate customer tastes into the product offer. What I also find particularly great about this project is that Google’s TensorFlow platform also allows people to experiment with machine learning, which may perhaps contribute towards higher acceptance of this technology in the long run.
I’m looking forward to seeing how this initiative progresses and what the impact will be on the online shopping experience!