Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

No Longer Trending on Artstation: Prompt Analysis of Generative AI Art

Jon McCormack, Maria Teresa Llano, Stephen James Krol, Nina Rajcic

TL;DR

The paper analyzes over 3 million prompts and their generated images from diffusion-based text-to-image systems to understand how language shapes AI art and its cultural impact. By combining NLP, topic modelling (MPNet, UMAP, HDBSCAN, cTF-IDF), and image captioning (BLIP), the study tracks prompt usage from mid-2022 to late-2023 across three large datasets. It finds prompting concentrates on surface aesthetics, reinforces Western and mainstream stylistic norms, and reveals recreational use patterns (e.g., Christmas, colouring books) with biases in artist-name references. Despite visible biases and homogenization, the authors argue that current AI imagery does not pose a direct threat to human art and call for deeper exploration of agency, authenticity, and the social implications of prompting practices. The work provides empirical, scalable insights for researchers, artists, and platform designers into how prompts encode cultural values and influence visual culture at scale.

Abstract

Image generation using generative AI is rapidly becoming a major new source of visual media, with billions of AI generated images created using diffusion models such as Stable Diffusion and Midjourney over the last few years. In this paper we collect and analyse over 3 million prompts and the images they generate. Using natural language processing, topic analysis and visualisation methods we aim to understand collectively how people are using text prompts, the impact of these systems on artists, and more broadly on the visual cultures they promote. Our study shows that prompting focuses largely on surface aesthetics, reinforcing cultural norms, popular conventional representations and imagery. We also find that many users focus on popular topics (such as making colouring books, fantasy art, or Christmas cards), suggesting that the dominant use for the systems analysed is recreational rather than artistic.

No Longer Trending on Artstation: Prompt Analysis of Generative AI Art

TL;DR

The paper analyzes over 3 million prompts and their generated images from diffusion-based text-to-image systems to understand how language shapes AI art and its cultural impact. By combining NLP, topic modelling (MPNet, UMAP, HDBSCAN, cTF-IDF), and image captioning (BLIP), the study tracks prompt usage from mid-2022 to late-2023 across three large datasets. It finds prompting concentrates on surface aesthetics, reinforces Western and mainstream stylistic norms, and reveals recreational use patterns (e.g., Christmas, colouring books) with biases in artist-name references. Despite visible biases and homogenization, the authors argue that current AI imagery does not pose a direct threat to human art and call for deeper exploration of agency, authenticity, and the social implications of prompting practices. The work provides empirical, scalable insights for researchers, artists, and platform designers into how prompts encode cultural values and influence visual culture at scale.

Abstract

Image generation using generative AI is rapidly becoming a major new source of visual media, with billions of AI generated images created using diffusion models such as Stable Diffusion and Midjourney over the last few years. In this paper we collect and analyse over 3 million prompts and the images they generate. Using natural language processing, topic analysis and visualisation methods we aim to understand collectively how people are using text prompts, the impact of these systems on artists, and more broadly on the visual cultures they promote. Our study shows that prompting focuses largely on surface aesthetics, reinforcing cultural norms, popular conventional representations and imagery. We also find that many users focus on popular topics (such as making colouring books, fantasy art, or Christmas cards), suggesting that the dominant use for the systems analysed is recreational rather than artistic.
Paper Structure (15 sections, 2 figures, 8 tables)

This paper contains 15 sections, 2 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. Text-to-Image Generation
  4. Prompt and Image analysis
  5. Datasets
  6. Initial Processing
  7. Dataset Statistics and Initial Analysis
  8. Analysis
  9. Use of Artist Names
  10. Topic Analysis
  11. Coda: What are all these images actually of?
  12. Summary and Conclusion
  13. Limitations of the study
  14. Findings
  15. Acknowledgements

Figures (2)

  • Figure 1: Visualisation of the MPnet embeddings of 1700 prompt specifiers and the 40 topics identified using the HDBSCAN clustering algorithm.
  • Figure 2: The 8 most popular topics for upscaled images in MJ2023