Local Diffusion

By Aron Petau • 6 minutes read •

Core Questions

Is it possible to create a graphic novel with generative A.I.? What does it mean to use these emerging media in collaboration with others? And why does their local and offline application matter?

Official Workshop Documentation | Workshop Call

Workshop Goals & Structure

Focus: Theoretical and Playful Introduction to A.I. Tools

The workshop pursued a dual objective:

  1. Accessible Entry Point: Provide beginners with a low-barrier introduction to text-to-image AI
  2. Critical Discussion: Initiate a nuanced political discussion about the ethical implications of these tools and demonstrate conscious decision-making options (such as locally installed tools)

The learning format was designed to be open, practical, and experimental, with emphasis on participants' creative output. Specifically, participants were tasked with working in groups to create a short graphic novel of 4-8 panels using AI. They had to actively modify the algorithm and familiarize themselves with various functions and interfaces.

Workshop Structure

The workshop was divided into two main parts:

Part 1: Theoretical Introduction (45 min)

Part 2: Hands-On Practice (2+ hours)

The "Categories Game" Warm-Up

To overcome initial fears about prompting, participants played a round of "Categories" (Stadt-Land-Fluss). They had to fill predefined prompting subcategories like "Subject", "Color", "Style", and "Resolution" with words starting with specific letters. This game challenged participants to think creatively about prompt construction beyond ready-made sentences found online.

Why Local AI Tools Matter

Consciously Considering Ethical and Data Protection Factors

A central idea of the workshop was to highlight the ethical implications of using AI tools and emphasize the consequences of local computing versus cloud computing. The workshop addressed two essential differences when applying the same AI models and algorithms:

Option 1: Proprietary Cloud Services

Option 2: Local Installation

Option 3: University-Hosted Services

From a data protection perspective, local and university-hosted solutions are far more conscious choices. While UdK services are technically also cloud services with data stored on a server, they represent a significant difference from proprietary services like OpenAI.

Visual Storytelling with Stable Diffusion

Participants engaged enthusiastically in the workshop process. They tried many different prompts and settings, producing results with a great variety of aesthetic and visual narratives. The workshop concluded with a final discussion about:

Technical Framework

With AI becoming increasingly democratized and GPT-like structures integrated into everyday life, the black-box notion of the mysterious all-powerful intelligence hinders insightful and effective usage of emerging tools. One particularly hands-on example is AI-generated images.

Tools & Interfaces Introduced

Learning Outcomes

Participants gained the ability to:

Reflections: The Student-as-Teacher Perspective

Personal reflection by Aron Petau

On Preparation and Challenges

"Preparing a workshop definitely felt like a big task because I felt the need to answer questions about tools that I myself am just discovering. One concern was that I wouldn't be able to answer an advanced technical problem. This ultimately turned out not to be a major issue, probably due to the limited duration of the workshop.

When it comes to the experience with an AI workshop, I believe it takes more than 3 hours to dive into such complex tools together with people. Even by extending the explanatory/theoretical part, I didn't manage to cover all the concepts I had deemed valuable beforehand... Nevertheless, a duration of 3-4 hours seems appropriate for an introductory workshop, as errors in time management accumulate over longer periods and more teaching experience would be needed here."

On Workshop Format and Atmosphere

"I really liked the rather non-hierarchical framework of the workshop, where it was clear that it was more about skill-sharing rather than a lecture format. Especially with practical things like image generation, when I didn't know the effect of a prompt or a parameter – which is the point after all – I could simply try out the effect together with the workshop participants and then examine the results.

The participants seemed to like the chosen format and difficulty level, where not too much mathematics and formulas were conveyed, but rather an intuition for the underlying process. The participants also actively participated in the critical discussion about the ethical use of AI and contributed perspectives from their own fields, which I greatly appreciated."

On Learning Teaching Practice

"During the preparation of this workshop, I had the opportunity to work independently and determine and organize my workshop dates myself. I greatly appreciated this freedom and authority, but a bit stronger pressure on a final deadline would have helped me lose the concerns about the teaching situation more quickly.

Now I'm looking forward to a possible round 2 – a next iteration where we can dive deeper into the depths of ComfyUI, an interface that I absolutely love, while its power also sometimes scares me."

Empowerment Through Understanding

Empower yourself against ready-made technology!

Do not let others decide on what your best practices are. Get involved in the modification of the algorithm and get surprised by endless creative possibilities. Through exploring local AI tools, we can:

While there is not much we can do on the data production side and many ethical dilemmas surrounding digital colonialism remain, local computation takes one step towards a critical and transparent use of AI tools by artists.