AI-Aided Design

Semester: Winter 2023/24
Teaching Format: Projekt mit seminaristischem Unterricht
Integriertes Produktdesign (B.A.)

Hochschule Coburg
Prof. Dr. Michael Markert,
Lukas Bebb,
Lisa Bialon,
Anita Dimitrova,
Alina Freund,
Sarah Haust,
Lisa Herbert,
Julia Kipke,
Julia Knötzele,
Julius Liedtke,
Sascha Meus,
Utisi Mwendwa-Wavamunno,
Anton Schmidt,
Johannes Schmidt,
Silas Weber,
Anton Zerbe

  • This article is based on a press release from Coburg University of Applied Sciences by Natalie Schalk, 4 March 2024

What if you could design anything you imagine?

»Imagine designing with the power of artificial intelligence and synthetic media« (ai-productdesign.de)

Artificial intelligence (AI) writes texts, creates lifelike images and is already programming computer program itself. Will AI soon replace human designers? Can it enrich and improve design work? Students on the Bachelor’s degree program »Integrated Product Design« at Coburg University of Applied Sciences explored these questions with Prof. Dr. Michael Markert in the winter semester 2023/24. The result is a wide range of designs that are now being presented in a digital exhibition.

Prof. Dr. Michael Markert und die Studierenden des Seminars Ai-Aided Product Design

»We analyzed various media such as images, sounds, texts and videos. We tried out many current possibilities of machine learning and available applications. We used high performance computing clusters to train our own models, but also our standard laptops and mobile phones to synthesize media.
We also looked at social and ethical issues and discussed how data control, provenance and output consistency could be improved.
We asked ourselves: Can AI collaborate with a designer? Can it do everything the industry advertises? Will there soon be a general artificial intelligence that rules the world?«

To answer these questions, the design professor and his students took a closer look at the heart of machine learning and the artificial creation of synthetic media:

Instead of drawing and designing, the students in Michael Markert’s »AI-Aided Design« project tried out, trained and experimented with various machine learning applications on High Performance Cluster computers, as well as on their own laptops and smartphones. The professor likes to speak of Machine Learning instead of Artificial Intelligence:

»This term is already misleading in German, not only because ›intelligence‹ can also stand for ›information gathering‹ in English. However, it is still just mathematical models that don’t understand anything, but simply calculate statistically probable complements.«

Markert is surprised by the high quality of the seminar’s results: »Almost all of the works show a good understanding of current machine learning technologies. The results are astonishing design drafts ranging from 3D visualizations with innovative Neural Radiant Fields to voice clones and a fashion line generated entirely by AI. They will be shown in an Instagram exhibition – a new work has been posted on the @ai-productdesign channel every day since the beginning of February until mid-March and at https://ai-productdesign.de.

Digital Assistants

»It was an interesting experiment for the students, during which they learnt a lot, especially about the performance of AI and their own expectations when dealing with the new synthetic media,« explains Markert. However, they all came up against their limits relatively quickly, mainly due to their expectations being too high.

»A system that has only been trained with what already exists cannot create anything new in the same way that humans can.«

As digital assistants, however, AI can speed up the design process and relieve designers of a lot of work. This is demonstrated by the experiments of Anton Zerbe and Sascha Meus on the synthesization of textures and geometries. The AI does not generate the complete design of a product but only specific parts of it, such as the visualization of surfaces.

Lukas Bebb has trained a system using 250 of his own drawings – and it can actually generate variations in very good quality and also transfer his drawing style to other representations. Utisi Mwendwa-Wavamunno has analyzed synthetically generated images of West and East African women and investigated which cultural and social prejudices can be found in the generated representations. Lisa Bialon worries about democracy because it was almost too easy for her to clone politicians’ voices and let them say whatever she wanted.

A Differentiated View of Hype

Another lesson learnt was that if you want to achieve good results, you have to be able to express exactly what you want to achieve.

»You also have to keep a critical eye on the results generated, because you are responsible for what you produce. The neural networks are only as good as the data they have been trained with and they often hallucinate or produce meaningless results that seem very plausible at first glance.«

The students have gained insights that go beyond writing prompts, i.e. instructions for the artificial network – resulting in a surprisingly differentiated view of the current hype, as Markert explains.

»Due to its technological complexity, the socially and now also politically charged debate is often characterized by ignorance and even an exaggerated fear of the downfall of humanity. Instead of just talking about it, we simply tried it out.«

Markert is convinced that the job profile of product design will constantly change as a result of new technologies. »The students in Coburg are now well prepared for this.«

»There’s no reason to be afraid; the hype is probably a little exaggerated. Artificial intelligence is a long way from being intelligent, but it will be there now and in the future, changing our design and creative processes.«

The results of the project have been on display since February 2024 in an Instagram exhibition and online at https://ai-productdesign.de.

Works in Focus

How I trained an AI to copy my drawing style by Lukas Bebb
@lukas.bebb

“My project had the goal of training an AI-model (LoRA) on drawings of my childhood. This taught me the process of using a dataset to train an AI to my liking. This is neccessary, because big AI-models like Stable Diffusion (XL), can struggle with displaying very specific concepts on which they have not been trained. This project proved that an individual like me can achieve this goal.”

AI generated NeRFs
by Sascha Meus and Anton Zerbe

@sascherrrr
@anton.zerbe

“The goal of our project was to support the rendering process with AI tools. Our idea here was to use the recently emerged AI technology NERFs to create the most realistic 3D geometries possible. In the process, we started some experiments with this type of AI tool. We recorded various objects in different environments. Although we were able to achieve interesting results in this area. We realized that this technology shows its strength in other fields. We see the potential strengths in the video photography sector in order to be able to decide how the scenery should be displayed during video editing. Drone shots of a static environment in particular could work very well.”

D2V
by Johannes Schmidt

@levenait

“The combination of multiple powerful AI tools into one workflow enables the generation of new product variances while keeping proportions close to the input product.”

Dj AI
by Julia Knötzele

@juliaknoetzele

“Livecoding presents a distinctive method for music creation utilizing code, expandable sample libraries, and live synthesizers. It provides extensive control, facilitating nuanced adjustments. I developed a package that can be installed in your text editor, granting access to OpenAI through. Everything necessary for collaboration is predefined, requiring only the input of short prompts in the editor. Additionally, I implemented a shortcut for seamless access, ensuring minimal disruption during livecoding sessions. With AI integration, the process is enriched by a wealth of knowledge and coding experience, fostering innovative music generation through collaboration. Stay tuned for further insights into this captivating fusion!”

Single word prompt
by Julia Kipke

@juliakipk3

“This project explores how AI systems respond to one-word prompts when generating images. Different AI models are given the same prompt with no other information, and the images they produce are observed to gain insights into how AI algorithms understand and interpret natural language inputs. Despite limited input, AI systems demonstrate diverse creativity. Analyzing these outputs reveals AI’s capacity for creative thinking and varied visual representations.
In total, nine tools were used and given the prompt ‘toaster’ to each generate 50 images. The following slides show just a portion of the 450 generated images, which were analyzed.
Why ‘toaster’? ‘Toaster’ is a relatively specific word, but it still allows for a variety of interpretations. A toaster is also a common and familiar object, which can make it an accessible prompt for testing AI image generation models. Despite its simplicity, a toaster presents some level of complexity in terms of shape, function, and context. This complexity can challenge AI models to accurately capture and represent the various aspects of a toaster in their generated images, providing valuable insights into their capabilities and limitations.”

How can I make use of AI in my creative endeavors?
by Alina Freund

@alina.frnd

“In the early days of artificial intelligence, people often look at the future with concern or curiosity and wonder what the job market will look like. This is why I have decided to examine this topic and tried to let AI take over as much of my design process as possible. Here are my results:
Despite looking good and convincing at first glance, a closer look at Ai-generated images quickly reveals the inconsistencies they have. Many of the designs fail to work/ are not functional in reality, but that doesn’t mean they are automatically unusable. The designs can instead be used as inspiration, on the basis of which many ideas can then be modeled after and developed further. This can be seen as a form of help, but in the end it does not take away completely the design process. That’s why a product designer is still needed, however, they can work more efficiently with AI.”

VintiqueArt Studios – a design process generated entirely by AI using the example of a streetwear brand
by Anton Schmidt

@anton_schmidt8

“This project is about investigating the importance of artificial intelligence for the design process. To work through this on a specific design, I have developed a streetwear brand completely from scratch. From stylistic direction and influences to names, logos and background of the brand to different garments, possible patterns, product renderings and an AI generated video of a runway show, the aim was to have each individual step designed by different tools so that the designer only takes on a selective role.
The result should serve as a basis for discussion in order to summarise how useful AI programmes are already for the design process in the current state of the art, which tasks the tools can take over from the designers, to what extent designers are prepared to give up an unusual amount of freedom of action and which role synthetically generated content could play in the the future of product design.”

Exploring AI as tool to visually find the gaps in the presentation and representation of the diversity of African People and their cultures
by Utisi Mwendwa-Wavamunno

@tisi_mwe.ndwa

“We design the world we live in simply be how we choose to be in the world and that is reflected in what we produce and the biases that result from that. With Artificial Intelligence being yet another reflection of this, I sought to use it, in the form of image generation, as a tool to explore what biases or ‘gaps’ that I, not simply as an individual from a non-dominant culture but Utisi and all that encompasses, could identify. Perhaps by learning to see and recognise the different gaps that will be spotted by different individuals each informed by their own uniquely different experiences, we can spot more ‘gaps’ and start seeking ways to mend them.”

Further Informations

Related Posts