Generative ML-AI imagery in Art: Representation, Bias and Creativity

The purpose of this research is to investigate and evaluate accessible Generative Models for their use in the development of original performance work and supporting “world building” material to assist in artistic creation and community engagement with an art work.

Using free or open source models and consumer tools, prompts and training methods will be critically evaluated with reference to both

i. The generative model’s output and how closely it matches the expectations of the artists


ii. Limits to representation, or bias that can be observed in the output (with particular reference to, but not limited to, gender and ethnicity bias)

Taking the original libretto of “The Haunting of Alvin Cohan”, a dramatic masque composed and written by David Adams, as a starting point, descriptive snippets of text will be used and manipulated to observe the output of the chosen models.

Where possible, models and parameters will be compared and contrasted. Training and opportunities to host and manage open source models and the relative accessibility for and satisfaction of the user’s artistic intentions will be critically evaluated.

The outcomes of this research will be shared on the project’s social media channels and professional profiles. The artists will also critically evaluate the characteristics of engagement on these platform, including but not limited to the mood, tone, and frequency of measurable interactions with the content, as well as the impact of engagement with platforms, content and audiences on the artist’s and any feedback effects observed in the development of the work.