Algorithmic Visual Poetry: When the Prompt Becomes the Work of Art

At the beginning we only looked at images generated by AI. Today, the art world is discovering that the real magic lies in the text that creates them. "Prompts"

At the dawn of generative art, collective wonder was entirely focused on the final result. Faced with the first, astonishing digital canvases produced by systems like Midjourney or DALL-E, the public and critics questioned the quality of the pixels, the composition of the image, and the supposed "death of the illustrator." The text typed by the user to obtain that image – the prompt – was considered a mere technical tool, a disposable means, an invisible bridge between human intention and the machine's execution.

Today, we are witnessing a resounding paradigm shift. The attention of the art world, galleries, and academia is moving from the generated image to the text that generates it. The prompt is claiming its own ontological dignity, evolving from a simple command line into algorithmic visual poetry, an executive score, and a primary creative act.

In this in-depth piece for the Scenarios and Reflections column, we will explore how textual instructions are themselves becoming the object of artistic exhibition. We will analyze the convergence between the ancient practice of ekphrasis and modern prompt engineering, to demonstrate a radical thesis: the prompt is no longer the means to arrive at the image, but is the conceptual work itself. The algorithm is not the artist, it is only the brush; true art resides in the syntax of the textual incantation.

1. From Keyboard to Gallery: The Exhibition of the Command

The consecration of the prompt as an autonomous work of art is no longer a theoretical speculation, but a museum reality. A fundamental breaking point occurred in New York with the solo exhibition of artist MetaCher, documented by EIN Presswire as the first exhibition to conceptualize AI prompts as pure art ("Prompt IS Art"). In a gesture reminiscent of Sol LeWitt's minimalism and conceptual art, MetaCher physically removed the generated images from the gallery walls. On display were exclusively the "prompt cores" (the raw text strings, introduced by the command "/imagine…"). The statement is peremptory: the heart of the work, the creative genius, resides in the textual instruction, while the visual result is merely a mechanical byproduct that can even distract the viewer.

Similarly, the exhibition JUXTAPOSITION 2026 explored the conflict and collaboration between humans and the algorithm. By placing traditional artworks alongside AI-generated images, the curators transformed text into an instrument of critique. After the reveal of the work, each prompt was visually displayed next to the AI output. Text ceased to be an industrial secret and became an exhibition object in its own right, allowing the public to measure the distance between complex human poetic intention and cold algorithmic execution.

2. Digital Ekphrasis and Executive Scores

To understand the literary value of the prompt, academia is invoking classical rhetoric. An illuminating article published in ArteActa interprets prompting as a contemporary and inverted form of ekphrasis. Traditionally, ekphrasis is the verbal description of a visual work of art (such as the description of Achilles' shield in the Iliad). With generative AI, the process is reversed: language evokes and creates the image. Text becomes the absolute medium between human imagination and the machine's rendering.

In this framework, prompt engineering ceases to be a purely computer science skill and becomes a new literary creative ability. Studies published in Taylor & Francis analyze how writing input for generative models requires exquisitely humanistic skills: a sense of composition, knowledge of art history, mastery of mood, and the ability for visual synthesis.

A systematic analysis conducted on arXiv ("Prompt Analysis of Generative AI Art") demonstrates that complex prompts for Text-to-Image systems function exactly like executive scores. Just as a composer writes notes, rests, and dynamics (allegro, fortissimo) on a staff to guide an orchestra, the AI artist writes subjects, stylistic modifiers, and lighting parameters to guide the neural network's latent space. In this light, as emphasized in the papers on conceptual art, writing prompts means establishing the rules of the aesthetic game, legitimizing textual instruction as pure conceptual art.

3. The Anatomy of Algorithmic Poetry

How does one write algorithmic visual poetry? Operational frameworks like ArtPrompt (published on Emergent Mind) or the guides from The Prompt Home for surreal and conceptual art show us that the prompt possesses a rigorous metric and syntax.

A classic structured formula involves the concatenation of: Subject + Surreal Concept + Texture/Material + Lighting & Mood + Camera/Quality. This structure is not a cold alphanumeric code, but a mosaic of visual, emotional, and technical elements. The process of iterative refinement of a prompt – adding a comma to change the weight of a subject, inserting words like "cinematic lighting" or "melancholic atmosphere" – is psychologically and technically identical to the work of limae labor (poetic revision) that a poet applies to their verses.

The link between traditional poetry and prompts is bidirectional. Research published in journals of the Nature circuit and on arXiv (such as the PoemToPixel framework) has explored the generation of images from genuine literary poems. By extracting emotional features, rhetorical figures, and imagery from the verses of historical poets, Artificial Intelligence translates the deep semantics of the text into a detailed visual prompt. This demonstrates that the prompt is capable of conveying not only geometric descriptions, but also complex metaphors, emotional rhythms, and semantic translations of the highest level.

4. The Taxonomy of the Contemporary Prompt

In light of this evolution, contemporary art criticism must equip itself with a new vocabulary to classify generative text. We must distinguish four evolutionary stages of textual instruction:

  1. The Instrumental Prompt: This is the basic use. A functional and pragmatic description (e.g., "A red cat on a blue sofa") used exclusively to obtain an image in the shortest possible time. There is no artistic pretension.
  2. The Prompt as Score: The text becomes a complex technical instruction. The author specifies the execution, the style (e.g., "in the style of Greg Rutkowski"), the lighting constraints ("volumetric lighting"), and the aspect ratio. The goal is to bend the machine to an exact predetermined vision.
  3. The Prompt as Visual Poetry: The text transcends literal description. It uses figurative, rhythmic language, oxymorons, and abstract concepts ("the whisper of time crumbling on a facade of neon marble"). The text is beautiful to read even before being processed by the algorithm.
  4. The Prompt as Conceptual Work: The definitive level. The generated image is deliberately hidden or destroyed. The text is printed, framed, and exhibited in a gallery. The work is the intention itself.

This taxonomy responds directly to the provocation raised by academic debate ("Is Writing Prompts Really Making Art?"). If twentieth-century conceptual art taught us that the idea is worth more than manual execution, the prompt is the natural apotheosis of this principle in the twenty-first century.

Key Operational Takeaways (for Curators and Artists)

  • Elevate the Text (For Curators): Future generative art exhibitions should never display the image alone. The prompt must accompany the digital canvas (or replace it), printed with the same typographic dignity dedicated to critical captions. The public must be able to read the incantation that evoked the form.
  • Archive the Iterations (For Artists): The creative process of prompt engineering does not reside in the first attempt, but in hundreds of semantic variations. Artists should preserve and publish their "prompting diaries" (the logs of textual modifications), as they show the labor, the authorial choice, and the aesthetic research far more than the final result.
  • Protect the Syntax (For Intellectual Property): In the legal debate over AI copyright, the defense of human authorship must shift from the generated image (which is often non-patentable) to the prompt itself. A 100-word text string, complex, evocative, and structured, is to all effects a literary work subject to copyright, regardless of the pixels it generates.

Conclusions: Reading the Pixels, Looking at the Words

The textual prompt was born as a crack to peek inside the "latent space" of Neural Networks. In a few years, that crack has become a mirror in which our ability to dream through language is reflected. When a prompt merges rigorous artistic intention, engineering constraints, and deeply evocative language, it ceases to be code and becomes literature.

By revealing the creative process and the delicate choreography between human control and algorithmic chaos, textual instruction transforms into a self-sufficient aesthetic object. Having reached this conceptual frontier, the art world is forced to ask itself a radical question: if the true beating heart of the work, the seed of inspiration and authorship, resides exclusively in the linguistic command that generates it… does it still make sense to exhibit the reassuring visual result, or does the prompt itself finally deserve to be read, contemplated, and criticized as a new and supreme genre of poetry?

Bibliographic References and Sources

Article by the Editorial Team of La Bussola dell'IA