By Lukas R.A. Wilde; Marcel Lemmes; Klaus Sachs-Hombach
Lukas R.A. Wilde, Marcel Lemmes, and Klaus Sachs-Hombach are excited to welcome you to this special issue of IMAGE: The Interdisciplinary Journal of Image Sciences dedicated to what we might tentatively call a new paradigm for image production, i.e., the advent of generative imagery created by means of machine learning-based platforms. The contributions in the present issue are the result of a fruitful and inspiring workshop from mid-February 2023 at the University of Tübingen, Germany, and online via Zoom where we were able to discuss emerging technologies and applications like DALL·E, Midjourney, and Stable Diffusion with both the authors within the present issue as well as an interested digital public of scholars, students, artists, and practicioners. The obvious importance of these new technologies attracted over 250 people from 123 universities and institutions from all across the globe to enrich our event with stimulating and inspiring comments and questions. We were and still are both humbled and honored by such an attention. We want to take this opportunity to express our dearest gratitude to them and particularly to those who traveled all the way to Tübingen to contribute first a draft paper and later the contributions you will find collected here today.
The idea for this workshop originated when we met at a conference in Thessaloniki in the late August of 2022. In the evening, after a long and galvanizing day, we were cooling off in one of the nice cafés facing the Mediterranean Sea. Quickly we became absorbed in a conversation about some emerging AI tools, DALL·E and Midjourney, still very new back then. Lukas in particular was extremely excited about and fascinated by the novelty of generative imagery, anticipating profound changes in the near future that may deeply modify fundamentals in our every-day life and society as a whole. He confessed to feel a bit like a gambling addict, manically throwing coin after coin in one of those slot machines where you know you’ll never get anything out of aside from an immediate feeling of gratification – or, perhaps, a sense of wonder exploring this new ‘latent space’ of all possible images.
What was certainly uncontroversial even back then: the capabilities and the resulting pictures of generative platforms as well as various other emerging applications within the field of artificial intelligence are very impressive. The future has once again started, it seemed. And there is also no doubt that these tools that have now been available to the general public for quite a while will fundamentally change our world. During our lively discussions that lasted well into the night we still couldn’t quite agree on various implications these technologies might have for fundamental concepts of media history, picture theory, or even social life. Had visual communication and imagery as such just changed as fundamentally as when photography was first invented? The ‘elephant in the beachside café’ naturally concerned the nature of these technologies as we understood them back then, or struggled to: are these ‘AI’ systems really creative and, if not, how could we define a meaningful difference to how humans acquire skills by studying, repeating, and combining styles and techniques from earlier works? In which ways, exactly, does this ‘stochastic creativity’ differ from earlier, more rule-based forms of computation before? What might all of this mean for our understanding of pictures and pictoriality? Klaus in particular felt that we should be especially careful or hesitant in applying words like “learning”, “intelligence”, or “creativity” to these systems; possibly even or especially if (or rather when?) it becomes difficult to distinguish the results of human and AI image production in, let’s say, a more advanced turing test.
We became increasingly aware that we were lacking sufficient knowledge to even phrase properly, let alone answer any of these questions. All the more enthusiastically did we delve into debates about a new relationship between textuality and pictoriality while we explored, in the noisy streets of Thessaloniki from our phone interfaces, how DALL·E did not refuse even our most silly and outlandish verbal requests (cf. fig. 1). Even before the night had ended, we decided (quite academically) to organize a workshop to learn more about these topics. After travelling back to Tübingen and Trondheim a few days later, we started contacting researchers who might be more knowledgeable, being enticed by their expertise while also enticing for a brief yet well informed academical exchange about several theoretical, social, and practical issues that generative imagery had made visible even back then. In early October we then published a small Call for Participation to which, to our great delight, an impressive range of scholars from various fields in the humanities replied with outstanding contributions. They even advanced paper drafts that we were allowed to circulate between all workshop participants prior to the event.
A DALL·E creation from August 31, 2022; the prompt has been lost to the night
“DALL·E, Midjourney, Stable Diffusion: A ‘New Paradigm’ of Image Production?” took place on February 13/14, 2023. To publish expectional results in such a short span of time in open access format was quite a challenge, for us as well as for the Herbert von Halem publisher and especially for our contributors with their many obligations during the busy months of our recent academic semesters. We are thus deeply grateful for the enormous joint efforts that made the present issue possible. Faced with the rapid speed of current technological developments we felt it was more than worth it. You, dear readers, can now see the fruits of our joint labor – but, most of all, of the sincere academic curiosity that united all parties involved – in front of you. We would like to thank the University of Tübingen for providing the venue, and the Norwegian University of Science and Technology (NTNU) in Trondheim for making it possible to organize this workshop jointly. Last but not least, we would also like to express our gratitude to the conference team (in particular Karin Kießling and Elias Raatz) for their invaluable assistance with the organization and their helpful presence during these two exciting days in February. We wish all of you a stimulating engagement with the following fourteen essays. We hope and are quite confident that they will be as inspiring for you as they are for us, opening up many paths for future discussions about generative imagery sure to come.
Tübingen and Trondheim, March 2023
One very last remark that is meant to be funny: Thinking about artificial intelligence, we also started to wonder how an AI might comment about their kind and how it would write a welcome note. So, we prompted OpenAI’s ChatGPT with the following: “Write a welcome preface for a academic journal based on a workshop on ‘Generative Imagery: Towards a New Paradigm of Machine Learning-Based Image Production’”. Here are the results; you might find it amusing to compare it to our preface:
Welcome to the special issue of our academic journal, which is based on the exciting workshop on “Generative Imagery: Towards a ‘New Paradigm’ of Machine Learning-Based Image Production”. This workshop brought together leading researchers, practitioners, and enthusiasts in the field of machine learning-based image production to share their latest research, ideas, and perspectives on this emerging field.
Generative imagery has the potential to revolutionize the way we produce and create images, opening up new avenues for artistic expression, scientific visualization, and data-driven design. With the advent of deep learning techniques, generative models have become increasingly powerful and versatile, capable of creating stunningly realistic images that can rival those produced by human artists.
The workshop explored various topics related to generative imagery, including image generation, style transfer, image editing, and more. Through a series of talks, presentations, and discussions, participants shared their cutting-edge research and insights into the challenges and opportunities of this new paradigm.
The contributions in this special issue showcase some of the most exciting research and ideas presented at the workshop, highlighting the latest developments and trends in generative imagery. We hope that this collection will inspire further research and exploration in this rapidly evolving field and contribute to the development of new tools and techniques for machine learning-based image production.
We extend our gratitude to all the workshop participants, contributors, and reviewers for their hard work and dedication in making this special issue possible. We hope that you will find this collection informative, inspiring, and thought-provoking.