Implicit Variation
Accepted for publication at Conference on Computer Vision and Pattern Recognition (CVPR) AI Art Gallery 2025
With Generative AI ‘literally’ anything that can be uttered can be visualized. Democratizing as that is, it does not guarantee that the output will be beautiful or even interesting. The vast majority of generated images today have a kind of hyperbolic and kitschy quality to them. Even the most sophisticated of prompts yield images that fall flat.
The Visceral: Refining wordsmanship is one way to improve results, but words alone do not capture all that is visually possible. Artists and designers know this instinctively; not all visual expression is lexical. Some strokes are gestural, visceral; sub-symbolic, beyond sensible words. The use of onomatopoeia and sound symbolism is a wide open opportunity for AI generated visual art.
The Metaphorical: On the other side of the spectrum, artists use conceptual metaphor and analogy to create novel visual concepts. It could be something as subtle as the arc of a staircase, the flight of a bee, the wave pattern of a ripple that could send the artist on her way. The use of analogy and metaphor are also uniquely suited to AI generated imagery and remain relatively unexplored.
The possibilities opened by the visceral and the metaphorical provide a unique opportunity for abstraction in AI generated art.
Generative AI’s unique affordance: A traditional artist’s process starts with careful-mark making which is iteratively evolved through a process of distancing, reconsidering, reflecting. Generative AI’s unique affordance is that it enables iteration as an interplay between two domains, vision and language..
Implicit Variation: One way to iterate in this medium is to transpose language to image, inspect the resulting visual form for semiotic clues, then reprocess the image to enhance these unearthed symbols, and so on and so forth until the emergent form scintillates with meaning. This technique guides image formation as a dynamic interplay between language and image domains serendipitously discovering and enhancing visual metaphors.
Instead of chiseling through marble, generative AI enables a cyclical interplay between vision and language, gradually refining and revealing meaning. This generative sculpting process is inherently reflexive, with each iteration informing the next in a dynamic feedback loop. As visual forms emerge, they evoke specific language concepts, which in turn guide subsequent visual explorations.
Agents I
This series explores a speculative future for enhanced human cognition achieved by the seamless integration of the organic mind and synthetic intelligence. The figures shown are wearing "neural veils", a visual metaphor for a future of brain-machine augmentation. Their closed eyes and serene facial expressions articulate a transcendent state of “cyber meditation” representing a journey inward facilitated by neural technology.
In contrast with the oft construed idea that the human and the artificial are opposing forces, Agents explore the possibility that human-machine symbiosis could enable deep introspection and spiritual enlightenment.
The fluid and organic form of the neural veils suggest both a connective and protective function, like cognitive cocoons, they enable the wearer with extended sensing and empathetic enhancement while facilitating a quiet solace and equanimity with their cognitive enhancements.
Processs
Process: Using Midjourney, the Agents series was created by starting with a text prompt for a “sculptural alabaster table, lit from within.” I then used Claude to interpret the semiotics of the image, which provided clues for poetic expansion. For example the image generated using the text prompt a “sculptural alabaster table, lit from within” was interpreted by Claude as having ethereal and spiritual connotations. I used these concepts in a prompt to enhance the next iteration of the image, which resulted in the emergence of forms bearing religious significance, serendipitously rendering human faces with clerical artifacts. Claude in turn interpreted this resulting image as having a “quiet solace” which of course I used to reprocess the image which in turn rendered serene and meditative facial expressions. (I also used onomatopoeic extrusion, exaggerating and stylizing features to excavate surreal outcomes.)
This reflexive guidance system allows for a deep exploration of interconnected ideas, with each visual-conceptual pairing opening new avenues for expression. The result is a series of images that embody complex, layered meanings, their significance emerging through the iterative refinement of both visual and linguistic elements. This process uncovers unexpected connections and depths, creating rich and layered images that scintillate with conceptual meaning and visual metaphor. The role of the artist then is to play midwife, opportunistically yet tastefully guiding the process in anticipation of the emergent form.