Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Seizing the Means of Production: Exploring the Landscape of Crafting, Adapting and Navigating Generative AI Models in the Visual Arts

Ahmed M. Abuzuraiq, Philippe Pasquier

TL;DR

Visual artists face a fragmented landscape when personalizing generative visuals; the paper maps options across three modes—model navigation, model adaptation, and model crafting—while highlighting active divergence and the governance between crafting and traditional development. It argues that model crafting should be treated as a legitimate creative activity and outlines challenges (training costs, interdisciplinary skills) and opportunities for data-efficient workflows and cloud-based tooling. The authors propose design directions for creativity-support tooling, including visual programming and direct manipulation interfaces, to broaden access for non-expert artists. The work aims to reconnect authorship, agency, and craftsmanship with generative AI in the visual arts and to guide future tool design and research.

Abstract

In this paper, we map out the landscape of options available to visual artists for creating personal artworks, including crafting, adapting and navigating deep generative models. Following that, we argue for revisiting model crafting, defined as the design and manipulation of generative models for creative goals, and motivate studying and designing for model crafting as a creative activity in its own right.

Seizing the Means of Production: Exploring the Landscape of Crafting, Adapting and Navigating Generative AI Models in the Visual Arts

TL;DR

Visual artists face a fragmented landscape when personalizing generative visuals; the paper maps options across three modes—model navigation, model adaptation, and model crafting—while highlighting active divergence and the governance between crafting and traditional development. It argues that model crafting should be treated as a legitimate creative activity and outlines challenges (training costs, interdisciplinary skills) and opportunities for data-efficient workflows and cloud-based tooling. The authors propose design directions for creativity-support tooling, including visual programming and direct manipulation interfaces, to broaden access for non-expert artists. The work aims to reconnect authorship, agency, and craftsmanship with generative AI in the visual arts and to guide future tool design and research.

Abstract

In this paper, we map out the landscape of options available to visual artists for creating personal artworks, including crafting, adapting and navigating deep generative models. Following that, we argue for revisiting model crafting, defined as the design and manipulation of generative models for creative goals, and motivate studying and designing for model crafting as a creative activity in its own right.
Paper Structure (9 sections, 1 figure, 1 table)

This paper contains 9 sections, 1 figure, 1 table.

Table of Contents

  1. Introduction
  2. The Landscape of Personalization
  3. Navigating Generative Spaces
  4. Adapting Generative Spaces
  5. Crafting Generative Spaces
  6. Active Divergence
  7. Model Crafting vs. Model Development
  8. Designing for Model Crafting
  9. Conclusion & Future Work

Figures (1)

  • Figure 1: Three modes for creating personal works using generative models: (1) model navigation, (2) model adaption, and (3) model crafting. In the first mode, artists navigate within a fixed generative space to find the aesthetics they like, e.g. by prompting. In the second, navigation takes places within an adapted generative space. In the third, artists (co-)explore the space of generative models (straight arrows) as well as navigating within the generative space of each model (curved arrows).