Conference Paper Draft
Phantom Mirror: Iterative Prompting and the Image as a Site of Thought in AI Phototherapy
Abstract
This paper proposes a new framework for understanding AI image-making through a practice-led methodology termed AI phototherapy. Moving beyond prompt-based authorship models, it argues that meaning in AI-generated imagery emerges through iterative, pattern-based interaction rather than linguistic instruction. Drawing on Walter Benjamin, Victor Burgin, and Jacques Lacan, the paper introduces the concept of the psychic snapshot: an image produced under affective pressure, prior to narrative stabilisation. It contends that AI image-making constitutes a shift from linguistic to visual-perceptual meaning construction, in which the image becomes a site of thought rather than representation. The paper situates this practice within contemporary debates on authorship, agency, and Creative AI, proposing that iterative prompting establishes a new regime of perception grounded in recognition rather than intention.
1. Introduction: Beyond the Prompt
Recent discussions of AI art have largely centred on authorship, prompt engineering, and the aesthetic status of generated images. While these concerns are valid, they risk mislocating where the work actually occurs.
In my own practice, developed through a project titled Phantom Mirror, AI image-making is not reducible to the prompt or the final image. Instead, it unfolds as a process of iterative engagement in which meaning emerges through interaction with generated patterns.
This paper argues that AI image-making must be understood not as a tool-based extension of existing artistic practices, but as a reconfiguration of perception and meaning-construction. It proposes that the decisive moment in this process is not production but recognition.
2. Methodology: AI Phototherapy and Iterative Prompting
The methodology underpinning this work is termed AI phototherapy.
It operates as follows:
- An emotional or experiential disturbance initiates the process
- A linguistic prompt is used to open a generative field
- Multiple images are produced
- The artist moves rapidly through these images, rejecting most
- One image produces a moment of recognition
This process is not linear. Once the initial prompt has been entered, the work shifts from language to visual iteration, where images generate further images through variation.
The key methodological principle is dialectical refusal: meaning emerges through the rejection of images that fail to resonate, rather than through the refinement of a predefined idea.
This aligns with the broader Creative AI shift from artefact to process, where artistic practice is understood as a relational system rather than a fixed output .
3. Benjamin: The Image as Flash
The structure of this process can be understood through Walter Benjamin’s concept of the moment of danger.
Benjamin argues that memory does not emerge through narrative reconstruction, but appears as a flash under conditions of disruption. Meaning is seized rather than constructed.
AI phototherapy operates precisely within this structure.
Images are not produced through calm reflection. They emerge in moments of emotional intensity, where language is insufficient. The resulting images function not as representations of the past, but as interruptions within the present, where something becomes visible that could not previously be articulated.
4. Burgin: The Psychical Object
While Benjamin provides a temporal model, Victor Burgin provides an ontological one.
Burgin distinguishes between the digital image as a distributed object and the psychical object, which exists through its effects on the subject. Meaning is not contained within the image but produced through affective and relational engagement.
In iterative prompting, images become psychical objects at the point where they:
- interrupt the flow of iteration
- produce affective arrest
- demand attention without explanation
The majority of generated images are discarded because they fail to produce this effect. The image that remains is not the most technically refined, but the one that acts upon the psyche.
5. Lacan: The Image at the Limit of Language
The final theoretical framework is provided by Jacques Lacan.
Lacan’s concept of the Real describes that which cannot be symbolised within language. It appears as rupture, excess, or impossibility.
AI image-making, in this context, can be understood as operating at the limit of the Symbolic.
The prompt belongs to language, but quickly becomes insufficient. The iterative image field begins to approximate something that resists articulation. The resulting images do not capture the Real, but circle it, producing forms that feel meaningful without being fully interpretable.
This explains the persistent uncanny quality of AI imagery: it occupies a space between the familiar (Imaginary) and the unsymbolisable (Real).
6. The Psychic Snapshot
From the convergence of these frameworks emerges the concept of the psychic snapshot.
A psychic snapshot is:
- produced in real time under affective pressure
- generated before narrative stabilisation
- recognised rather than constructed
- understood only retrospectively
It is not an image of something.
It is an image formed by something acting on the psyche.
This concept extends both photographic theory and psychoanalysis. It reframes the image not as representation, but as event.
7. From Language to Visual Meaning
The most significant implication of this methodology is the displacement of language.
Traditionally, meaning is constructed through a linguistic sequence:
experience → language → interpretation
In AI phototherapy, this sequence is reconfigured:
experience → image → recognition
The image becomes the primary site of meaning-construction. Language remains as an entry point and a reflective tool, but it no longer structures the process.
This aligns with the proposition that AI image-making introduces a visual-perceptual layer of meaning, capable of operating where language fails .
8. Agency and Authorship
This shift necessitates a rethinking of agency.
Authorship is not located in the prompt, nor in the system, nor in the final image. It is distributed across the process.
However, the role of the artist remains specific:
- navigating the iterative field
- sustaining attention
- recognising affective alignment
Agency becomes a function of attunement rather than control.
This complicates dominant narratives of AI art that focus on ownership or authorship. It suggests that the artist’s role is not diminished, but transformed.
9. A New Regime of Perception
Taken together, these elements constitute a new regime of perception characterised by:
- non-linear temporality
- pattern-based iteration
- recognition as epistemic event
- images as sites of thought
AI image-making is not simply producing images. It is reorganising how meaning is constructed and perceived.
This requires a new literacy.
AI images cannot be read as photographs or paintings. They must be understood as nodes within a dynamic perceptual field, where meaning emerges through interaction rather than representation.
10. Conclusion: The Image as Thought
This paper has argued that AI phototherapy establishes a practice in which the image becomes a site of thought.
Drawing on Benjamin, Burgin, and Lacan, it has shown that:
- images emerge under conditions of disruption
- function as psychical objects
- operate at the limits of language
The concept of the psychic snapshot captures this convergence.
Ultimately, the contribution of this work is methodological.
It demonstrates that AI image-making can function as a form of thinking through images, particularly in contexts where language is insufficient.
The image is no longer:
- a representation of thought
- or an outcome of thinking
It becomes:
the place where thinking happens.