Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

It's a Feature, Not a Bug: Measuring Creative Fluidity in Image Generators

Aditi Ramaswamy, Melane Navaratnarajah, Hana Chockler

TL;DR

This paper attempts to define and empirically measure one facet of creative behavior in AI, by conducting an experiment to quantify the "fluidity of prompt interpretation", or just"fluidity", in a series of selected popular image generators.

Abstract

With the rise of freely available image generators, AI-generated art has become the centre of a series of heated debates, one of which concerns the concept of human creativity. Can an image generation AI exhibit ``creativity'' of the same type that artists do, and if so, how does that manifest? Our paper attempts to define and empirically measure one facet of creative behavior in AI, by conducting an experiment to quantify the "fluidity of prompt interpretation", or just "fluidity", in a series of selected popular image generators. To study fluidity, we (1) introduce a clear definition for it, (2) create chains of auto-generated prompts and images seeded with an initial "ground-truth: image, (3) measure these chains' breakage points using preexisting visual and semantic metrics, and (4) use both statistical tests and visual explanations to study these chains and determine whether the image generators used to produce them exhibit significant fluidity.

It's a Feature, Not a Bug: Measuring Creative Fluidity in Image Generators

TL;DR

This paper attempts to define and empirically measure one facet of creative behavior in AI, by conducting an experiment to quantify the "fluidity of prompt interpretation", or just"fluidity", in a series of selected popular image generators.

Abstract

With the rise of freely available image generators, AI-generated art has become the centre of a series of heated debates, one of which concerns the concept of human creativity. Can an image generation AI exhibit ``creativity'' of the same type that artists do, and if so, how does that manifest? Our paper attempts to define and empirically measure one facet of creative behavior in AI, by conducting an experiment to quantify the "fluidity of prompt interpretation", or just "fluidity", in a series of selected popular image generators. To study fluidity, we (1) introduce a clear definition for it, (2) create chains of auto-generated prompts and images seeded with an initial "ground-truth: image, (3) measure these chains' breakage points using preexisting visual and semantic metrics, and (4) use both statistical tests and visual explanations to study these chains and determine whether the image generators used to produce them exhibit significant fluidity.

Paper Structure

This paper contains 14 sections, 7 figures, 11 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Background & Experiment Construction
  3. Existing Image Generation Metrics
  4. Defining and Measuring Fluidity
  5. Computational Resources & Justifications
  6. Chain Construction
  7. Image Guidance, Glitches, & Breaking Calculations
  8. Analysis Tools
  9. Results & Analysis
  10. Statistical Results
  11. Analysis
  12. Limitations
  13. Conclusions
  14. Technical Appendices

Figures (7)

  • Figure 1: The chain length frequency distributions for all three control group combinations.
  • Figure 2: A box plot of the distribution of chain lengths for the different combinations of models
  • Figure 3: The different combinations of models plotted on a scale of fluidity (the direction is fluid $\rightarrow$ faithful, with higher values representing more faithfulness) using the Kullback-Leibler divergence.
  • Figure 4: Example chains produced by "ground truth" image 0011, labeled "truck", where the top left is the "ground truth" image and the rest are generated.
  • Figure 5: Example chains produced by "ground truth" image 0004, labeled "broccoli", where the top left is the "ground truth" image and the rest are generated.
  • ...and 2 more figures

Theorems & Definitions (2)

  • Definition 1: Fluidity
  • Definition 2: Breaking Point/Chain Length