I began my project Being Borges with a question that felt deceptively simple: what happens when language becomes literal?
For most of literary history, language has operated through indirection. Words evoke images, but they do not produce them. A sentence unfolds in the mind; a metaphor opens a field of possibilities. Meaning is constructed in the space between text and reader. But with the emergence of generative artificial intelligence, this relationship has shifted. Language is no longer only descriptive – it is operational. It can be used to instruct a machine to produce an image. In this sense, language becomes literal through the visual.
Being Borges emerges from this shift. It is a 12-part series that takes Jorge Luis Borges and Margarita Guerrero’s The Book of Imaginary Beings as its point of departure and asks what is at stake when literary description becomes visual output. The project is not concerned with illustration in the traditional sense. Rather, it explores translation across systems: from Spanish to English, from text to image, from human imagination to machine interpretation.
In doing so, it situates itself within a broader artistic lineage. As Lev Manovich suggests, generative AI does not mark a radical break with art history but rather continues a modernist and postmodern tradition of recombination, in which new works emerge from the accumulation and reinterpretation of existing cultural artefacts. In this sense, Being Borges is less about novelty than about making visible a process that has always been at the core of artistic production: the reconfiguration, the translation, of inherited forms.
Borges, databases and the imaginary
The Book of Imaginary Beings reads like an encyclopedia of creatures that do not exist. It catalogues mythological entities with the authority of scientific description, blurring the line between fiction and knowledge. Each entry is itself a composite, drawing from multiple literary and cultural sources. The beings the book describes are the result of a long chain of translations and reinterpretations.
In this way, Borges and Guerrero anticipate what we now recognize as database logic. Their book is not linear but accumulative. It gathers fragments from different traditions and organizes them into a system that feels both coherent and unstable. Indeed, the preface is one of the most engaging texts in the book: in it, Borges poses a rhetorical question asking why a giraffe is not in the book. What determines reality from fiction is arbitrary at best, capricious at worst.
Generative AI operates in a similar manner. It is trained on vast datasets – millions or billions of images and texts – and produces outputs by recombining patterns using that data.
When I began working on Being Borges, I realized that Borges and Guerrero’s project and generative AI share a structural affinity. Both construct new forms from existing ones. Both rely on accumulation. Both destabilize the notion of originality.
The difference lies in materialization. The book’s beings exist only in language. AI gives them form.
Writing as prompt
Working with text-to-image models, I quickly realized that prompting is not a technical task but a linguistic one. The structure of a sentence – its syntax, rhythm and emphasis – directly affects the resulting image. Writing becomes a form of instruction.
As a poet, this shift is both exhilarating and unsettling. Poetry has always been a generative practice, but its generativity has been internal. A poem produces images in the mind of the reader, not in the world. With AI, that process is externalized. The image appears on screen, fixed and shareable.
This raises a fundamental tension. Language is often understood as open-ended, inviting multiple interpretations, whereas the image is assumed to fix meaning in a singular visual form. Yet this distinction unravels on closer inspection. Images, too, are unstable: they shift with context, memory, cultural conditioning and the viewer’s gaze. What appears immediate is not neutral. In moving from text to image, we do not move from multiplicity to fixity, but from one kind of instability to another.
What Being Borges foregrounds is not the closure of meaning, but the friction between these systems. When we encounter a description as language, we inhabit it – each reader generating a different internal image. When that same description is rendered visually, we are confronted with a version that is not ours. This confrontation reveals our own expectations: how we read, how we imagine, how we trust – or distrust – text versus image. In this sense, the translation exposes something deeper: that language itself may not be as stable a system of signification as we assume.
Being Borges inhabits this tension. I use Borges and Guerro’s descriptions as prompts, but I do not treat the resulting images as definitive. Instead, I place them alongside the texts that generated them, as well as alongside my own poems, creating a triangulation of meaning. The work does not resolve the relationship between text and image; it stages their mutual instability, allowing meaning to emerge in the space between them.
Meaning, like beauty, is in the eye of the beholder.
The impossibility of translation
At the heart of the series lies the impossibility of translation.
Each work follows a three-part process. I first generate images using Borges and Guerrero’s original Spanish text. I then repeat the process using Norman Thomas di Giovanni’s English translation. Finally, I write a new poem – an interpretation, or perhaps a misinterpretation – and use that as a prompt to generate additional images.
The results are never equivalent.
The Spanish prompts tend to produce images that feel more distant, more atmospheric. The English prompts often yield more narrative, more human-centred compositions. My poems generate something else entirely – images that are often more abstract, more fragmented.
These differences reveal that translation is not a transfer of meaning but a transformation of it. Spanish and English are not interchangeable systems; they carry different histories, different connotations, different relationships to the world. AI makes these differences visible.
At the same time, the images reveal the biases embedded in the datasets used to train these models. Western visual culture is overrepresented, and certain forms of representation are privileged over others. The machine does not simply translate language – it interprets it through a specific cultural lens.
In this sense, Being Borges is also a critique. It exposes the infrastructures that shape AI- generated images and invites viewers to question the neutrality of these systems.
Translation as method
Translation, in my practice, is not limited to language. It is a broader methodological framework.
In other projects, I have translated poetry into choreography, treating the body as a site of inscription. In Paperwork and Speech Patterns, I use AI as what I call a “creative conspirator”, building systems that visualize performed verse. Audience responses are transformed into visual structures, making visible the emotional and interpretive labour of reading.
Across these works, translation becomes a way of thinking. It allows ideas to move between media: from text to movement, from movement to image, from image to system. It foregrounds the instability of meaning and the role of the reader or viewer in constructing it.
My work often involves translating poetry into forms such as motion, silence, or digital systems, thereby expanding the field of what a poem can be. In this expanded field, the boundaries between disciplines dissolve.
Literature becomes visual; performance becomes data; language becomes structure.
Recursion and authorship
Borges’s work is filled with recursive structures – labyrinths, mirrors, infinite regress. I wanted Being Borges to embody this logic.
In works such as Scylla, the process becomes explicitly recursive: text generates image, image prompts poem, poem generates image again. Each iteration transforms the previous one. There is no origin point, only a series of translations.
This recursion destabilizes authorship.
Who is the author of these images? Borges, who wrote the original text? Di Giovanni, who translated it? Myself, who selected and reconfigured the prompts? Or the algorithm, which produced the image based on its training data?
Rather than assigning authorship to a single agent, Being Borges distributes it. Authorship becomes a network rather than a fixed position – an interplay among writer, translator, artist, dataset and machine. This distribution is made visible within the works themselves: my poems appear in red, a colour traditionally used to correct, annotate and intervene.
Writing in red signals that my voice enters the text as a response, not an origin—an act of revision rather than invention. It marks the poems as intertextual gestures, embedded within and in dialogue with Borges and Guerrero’s writing and its translation.
At the same time, the work insists that this dispersion is not entirely new. Human creativity has always been intertextual, shaped by citation, influence and reinterpretation. We have always written with and through others. What AI does is render this condition visible – externalizing a process that has long been internal, and in doing so, challenging the myth of singular authorship, a myth Borges himself enjoyed unravelling.
Error and the uncanny
One of the most compelling aspects of working with AI is the presence of error.
Early in the process, the images often contained distortions – anatomical inconsistencies, impossible perspectives, unexpected juxtapositions. These were not failures to be corrected but openings to be explored.
Error introduces the uncanny. It disrupts the illusion of realism and reveals the image’s underlying mechanisms. It reminds us that what we are seeing is not a direct translation of the text but a constructed interpretation.
In Being Borges, I often preserve these moments of instability. They align with Borges’s interest in ambiguity and contradiction, but they also point towards what I call analogue generativity: the idea that meaning is not fixed in the text or the image but emerges through the encounter between reader and work. Each reading becomes its own universe; each viewer constructs – and is constructed by – the work in turn, depending on their personal history and current context.
This becomes particularly evident in The Sylphs, where the image’s apparent lightness and immateriality resist stable interpretation. The text on the bottle is almost legible, yet not. The figures seem to dissolve as much as they appear, inviting the viewer into a continuous act of rereading. In this way, the instability produced by error does not diminish meaning – it multiplies it, returning us to the open-endedness we associate with language and revealing that images, like texts, are never complete in themselves.
From image to body
Although Being Borges is primarily visual, it is also performative. Very much like poetry.
In presenting the series, I perform the poems I write, moving through the images and texts as a guided tour. This reintroduces the body into the process. The voice becomes another medium of translation, carrying the text back into time and space.
This movement – from text to image to performance – reflects my broader practice. I am interested in how meaning is constructed across different modes of experience. The image part of a larger system.
Conclusion: the stakes of literal language
What is at stake when language becomes literal?
On one level, the answer is technical. AI enables new forms of image-making, expanding the possibilities of artistic production. But the stakes are also conceptual. The transformation of language into image alters how we think about meaning, authorship and interpretation.
Being Borges does not attempt to resolve these questions. Instead, it stages them. Does language ever become literal?
By placing Borges and Guerrero’s texts, AI-generated images and my own poems in dialogue, the series reveals the instability at the heart of translation. It shows that meaning is not confined to a single medium but emerges through interaction – between languages, between systems of signifiers, between humans and machines.
Text-to-image AI makes visible something that has always been true: language is generative. It produces worlds, whether in the imagination or on the screen.
What changes is not the nature of language, but our awareness of its power. And perhaps that is the most significant shift of all.
Recommended citation: CABALLERO, Ana María. Being Borges: when language becomes literal Mosaic [online], May 2026, no. 207. ISSN: 1696-3296. DOI: https://doi.org/10.7238/m.n207.2604
Deja un comentario