General AI topics / Image generation

Generative poetry with images 061523

by VR · 06/15/2023

Generative poetry with images

While the predominant “prompt engineering” approach to generative art treats deep neural networks (AIs) as tools to be finely calibrated, I have treated them more like collaborators—whose abilities, predilections and foibles can be plumbed. As mentioned in another post, my primary esthetic focus is a contemporary Surrealist approach that uses the constructs of these profoundly alien AIs to break the symmetries of surface reality.

My typical workflow involves a three part process:

I feed the AI a prompt that is typically vague or even cryptic. Often these prompts assume the form of a koan or mashup, where seemingly unrelated concepts are forced into association.
Next the AI parses this poorly defined, or even contradictory, information though its layers of neurons to finally generate an image. Sometimes this image maintains only marginal relevance to the original prompt information.
As the final step—and the one that makes for poetry—I apply my very human associations, biases and desires to create a textual title. As with titling all artworks, this provides a context and a framework for viewing the image and imparting “meaning” to it.

Following are some examples. Each shows the textual image prompt, the generated image, and my textual title for that image.

Prompt: Visualize the wave function of a quantum paradox.