Colloquium on "Objects of Interest and Necessity" – HIAS

Nicolas Malevé (Sciences Po, Paris) & Christian Ulirk Andersen (Aarhus Institute of Advanced Studies)

Hamburg Institute of Advanced Studies, Nov 25, 2025

AGENDA

1) Introducing AI imaging (the interface Draw Things)

2) Hardware (the GPU and how to access it through a network)

3) Currencies ("Kudos" in Stable Horde)

4) Organisations of AI imaging (Hugging Face, in particular)

5) The question of 'autonomy' and why 'infrastructures' are not trivial.

8-10 minute presentations, followed by two-three questions (8-10 minutes)

... Based on book, open access PDF

... "Guestbook" at the bottom of this page

QR code for the "Objects of Interest and Necessity: a tour guide to autonomous AI imaging" project page. Download the book, follow the wiki (see 'Colloquium on "Objects of Interest and Necessity" – HIAS' page)

"Draw Things": Introducing AI imaging

Draw Things" is a free, AI-assisted image generation app for Mac, iPhone, and iPad that uses Stable Diffusion to turn text prompts and images into art. It is known for being able to run offline on devices for privacy

It includes a extensive control over many Stable Diffusion settings ("basic" and "advanced") and variety of features for professional-level image generation and editing - as opposed to the simple presets of e.g. DALL-E (OpenAI)

EXAMPLE prompts:

"Disney-Pixar's Woody from Toy Story in a saloon, holding a beer, drunken look on his face, realistic textures. The art style is cinematic, with realistic lighting and shadows, smooth brush strokes inspired by Pixar's aesthetics for animated movies. Ultra-realistic textures, ultra sharp details, and vibrant colours, 8k, masterpiece."

Negative prompt: blurry, cluttered background, low resolution, unrealistic lighting, harsh shadows, flat lighting, low-resolution, watermark, text, logo, text overlays, extra limbs, distorted anatomy, bad proportions, other characters, distracting props

Base model: SDXL Base (v.1.0) // good for anything – large scale, automated

LoRA: animte/pixar-sdxl-lora // 'tweaking' the base model - small scale, community made

Trigger word: Pixarrrable

LoRA "The Incredible Hulk (2008)." By the user 'BigHeadTF'

Screengrab of the LoRA page on the civit.ai platform. Civitai.com is a platform for images and models (LoRAs) and features a mix of fan-fiction, role-playing game characters and stereotypical (incel-like) gender representations.

What we have seen is what we call the "pixel space" of AI imaging. There is much more to AI than just the 'algorithms' and statistical "diffusion" models that generate the images.

"Latent space" - which we're not going to talk much about. But notice the pipeline of models, their different purposes, scales and communities behind.

A screen shot of the interface to create a LoRA in Draw Things. It displays selection of some (of the many) parameters that users can define in the training, as well as the upload of images for the training dataset. The images also needs to be annotated individually.

Hardware (the GPU and how to access it through a network)

Image from the post A complete anatomy of a graphics card: Case study of the NVIDIA A100. [1]

Ironwood, new Google TPU, centralized architecture of TPUs

TPUs are stacked in huge datacenters

Edge deployment

AI Horde, a horde of graphic cards

Artbot demo (pre-recorded)

Currencies ("Kudos" in Stable Horde)

The funding mechanism of open infrastructures imposes limits to their development. As the material base is mutually shared by the contributors, the Horde as a whole considerably reduces its costs. In this manner, it is able to operate with minimal funding such as an occasional grant. Its economical structure is further extended by an internal currency. The Horde uses a system of incentives that takes the form of Kudos (DB0 2022a) rewarding those who dedicate computer time to the project by giving them priority in the queue and the means to generate their own images faster. Importantly, Kudos are also conceived as a mechanism against appropriation. If someone builds a commercial interface on top of AI Horde, the commercial app will not be able to siphon all resources for its purpose. The more it will use the Horde without giving back, the slower the network will respond to its queries. Unless the commercial actor gives computer time to the Horde, thereby feeding back into the project.

A partnership with LAION for the production of better algorithmic aesthetic predictors (Thoukydidis and hlky 2023). The quality of a predictor is key to image generators as their value lies in their ability to provide users with environments to generate images that correspond to their taste. Producing good aesthetic predictors is a complicated task. In general, predictors are trained on large collections of images where microworkers rank them according to a set of aesthetic criteria. Even if the click workers are badly compensated, the annotation of large data collection remains costly. And the annotators often struggle with the aesthetic criteria they have to apply. Therefore, communities of users who are engaged in the production of images are highly valued by engineers both for their expertise and their interest in getting predictors that reflect their taste. AI Horde’s partnership with LAION can be understood as a convergence of interests. Horde users evaluate images produced by Stable Diffusion models and rank them in order of preference. And LAION is responsible to turn these human judgements into data to feed aesthetic predictors. Further, to encourage participation, AI Horde distributes Kudos to those providing such ratings. In accordance, users who rate images have priority access to the Horde’s queue. And Stability occasionnally donates access to its GPUs to the Horde in exchange.

Hugging Face: The organisations of AI imaging

Where is the model?

https://huggingface.co/animte/pixar-sdxl-lora (the LoRA used before)

Hugging Face is imply put, a collaborative hub for AI development – targeted image generation, speech synthesis, text-to-video. image-to-video, image-to-3D, and much more.

It attracts amateur developers who use the platform to experiment with AI models, as well as professionals who use the expertise of the company or take the platform as an outset for entrepreneurship.

Screenshot of huggingface.co, 2017. Made by French entrepreneurs Clément Delangue, Julien Chaumond, and Thomas Wolf. The company was entirely focused on building a new chatbot app for teenagers.

Screenshot of huggingface.co, 2025. There were (in 2021) 10,000 forks (i.e., branches of development projects) and around 5,000 companies using Hugging Face in one way or another

Huggingface' community section reveals how users are predominantly with high developer expertise and knowledge of how AI models work, and how to employ or experiment with them. Mainstream platforms offer very few parameters for experimentation - whereas Hugging Face enables a richness.

Capital interests

The company has moved from providing a service (a chatbot) to becoming a major (if not the) platform for AI development – for speech synthesis, text-to-video. image-to-video, image-to-3D and much more.

Business model: Monthly fees - from "Pro" to "Enterprise" – with priority support and restricted access for secure product development. Users of the platform today include many start-ups. You can also find and experiment with 'flagship' models such as the Chinese DeepSeek, OpenAI's GPT-models, etc.

Venture capital: Received their first round of investment of $1,2 million in 2017. In 2021 they received a $40 million investment to develop its "open source library for natural language processing (NLP) technologies. Investors today include Google, IBM, NVIDIA

Partnerships: Hugging Face is able to host and distribute – in VERY large scale – the use of datasets and AI models. This is a highly sought expertise. E.g. Microsoft and Amazon Web Services.

A diagram by the European Business review representing Hugging Face business model. The excessive (and growing) market value of Hugging Face reflects, in essence, the high degree of expertise that has accumulated within a company that consistently has sought to accommodate a cultural community, but also a business and enterprise plane of AI. Managing an infrastructure of both hardware and software for AI models at this large scale is a highly sought expertise.

Democratizing AI?

By making AI models, datasets and also processing power widely available, it can be labelled as an attempt to democratise AI and delink from the key commercial platforms, yet at the same time Hugging Face is deeply intertwined with numerous commercial interests. Hugging Face is not only for amateur developers. On the platform one also finds an 'Enterprise Hub' where it offers, for instance, advanced computing at higher scale with a more dedicated hardware setup and 'Priority Support'. For this more commercial use, access is typically more restricted. This may also explain the investment of e.g. Meta in the company (as en entrepreneur hub). It is suspended between more autonomous and peer-based communities of practice, and a need for more 'client-server' relations in model training, which generally is dependent on 'heavy' resources (stacks of GPUs) and specialised expertise - perhaps explaining the investment by Amazon Web Services. Other investors includeGoogle, Intel, IBM, and NVIDIA.

The question of 'autonomy' and why 'infrastructures' are not trivial

Abandoning the marvels of the digital future:

Why 'disconnecting', 'decentralising' and 'autonomy'?

Different motivations:

A) A dissatisfaction with the content and experience

B) A (Marxist) dissatisfaction with the ideological construct (i.e., the separation of labor from capital + society's self-creation independent of capitalist control)

The digital (AI) imaginary - in Silicon Valley design

Cover of design researcher/guru Don Norman's book - advocating for interface design where the user can freely focus in the task (rather than the tool), and ultimately embeds 'people' in an interactive experience (rather than a 'user' of a tool).

Microsoft's vision of a digital future. A world where the digital (AI) surrounds us, we immerse ourselves, experiential AI - very electric blue vision.

"The ‘twin peaks’ of American ideology – realism and individualism"

Gregory Ulmer: (1991) the new interfaction/interface industry is a direct continuation of Hollywood cinema’s drive for “occultation of the production process in favour of a consumption of the product as if it were ‘natural’ […]. The ‘twin peaks’ of American ideology – realism and individualism – are built into the computing machine (the computer as institution),”

“The Californian Ideology”

Richard Barbrook and Andy Cameron’s critique of Silicon Valley: “The Californian Ideology” (1995) is combining libertarian rhetoric and technological enthusiasm :

“Across the world, the Californian Ideology has been embraced as an optimistic and emancipatory form of technological determinism. Yet, this Utopian fantasy of the West Coast depends upon its blindness toward – and dependence on – the social and racial polarisation of the society from which it was born.”

"Texan Ideology"

Fred Turner, the Texan Ideology

Fred Turner, Texan Ideology (2025)

"Vibe revenue"

Magnificent Seven vs the rest of the economy

Vibe revenue

On the ruins of AI?

Dependencies and chosen dependencies

AI - suspended between different planes

+++++++++++

Themes (suggestions):

Intro to our project - "autonomy" and "decentralisation" of AI (and other knowledge infrastructures)

How to imagine with objects

Maps and cartography

Pixel Space: Interfaces to AI imaging

Pixel Space: Visual cultures of AI imaging (where do images come from - where do the go?)

Latent Space: LAION, Models, LoRAs (how are images collected and what happens to them?)

The material plane: The infrastructures of GPUs - earning and spending currency.

The organisations involved: LAION, Hugging Face, etc. – the grey zones of autonomy