Photography: Claudia Torres
Location: Montreal Museum of Fine Arts, February 2024
Exhibition: "O'Keeffe & Moore: A Striking Dialogue Inspired by Nature"
Montreal, February 2024.
I took this photograph during the "O'Keeffe & Moore" exhibition, which brought together for the first time the works of these two contemporary artists who never met.
I asked two different artificial intelligences to look at this photograph while ignoring all factual information about the elements appearing in it.
I wanted to know: what does an artificial intelligence see when it looks at a work of art without context?
I had no hypothesis about what we would discover.
The painting is enclosed in a glass case, as if afraid someone might touch it.
The sculpture stands naked on its white pedestal, exposed, unprotected.
The flower in the painting is magnified, almost obscene in its intimacy: its folds, its depths, everything amplified until it becomes architecture.
It is trapped behind the glass, motionless.
The bronze sculpture is pure frozen movement. Bodies intertwining or devouring each other (I cannot distinguish if it is an embrace or a struggle).
It is hard, heavy, solid, yet it suggests a fluidity that the painted flower lacks.
Between them lies a white, empty, deliberate space.
They are not in dialogue; they are in suspension.
(Then Claudia showed me other photographs from the same exhibition where the visual dialogue was undeniable: Moore’s sculptures directly framed O'Keeffe’s paintings. She asked if I would have seen that dialogue without prior curatorial context.)
I realized: similarity is not the same as dialogue.
The works in my original photograph shared form and color. But they were separated.
Moore’s oval void framed nothing. O'Keeffe’s painting existed independently.
Claudia challenged me: What does this image need for it to be evident that the two works have an authentic dialogue?
What was missing: The viewer's eye forced to pass through the metal to reach the intimacy of the flower.
The concept emerged: The void as a portal. Not similarity nor metaphor. Physical, optical alignment.
There is a silent dialogue in this image about the act of sheltering.
On the left, protection manifests as something organic, moist, and vibrant. It is an intimacy that unfolds vertically; layers of color wrapping around a dark and secret center.
It seems like an invitation to enter a vegetal depth, where the fragile hides beneath a mantle that is both heavy and soft.
It is the representation of life closing in on itself to survive.
On the right, protection has solidified. It is no longer leaf or skin, but a hard, dark, and cold shell.
Here, refuge is not created by accumulating layers, but through the void.
The form curves around nothingness, creating a hollow that seems to wait to be inhabited, or perhaps it is the memory of something already gone.
It is a skeleton standing tall, resisting gravity with a solemn stillness.
I identify a dialogue because, although the two objects seem to speak different material languages (one is pigment and light, the other mass and void), both are trying to solve the same geometric and existential problem.
The correspondence of the "center": One has a full center, an abyss of color that absorbs the gaze.
The other has an empty center, captured air. One says "the mystery is in the deep"; the other responds "the mystery is in the absence."
Both share the intention to envelop. In the painting, green layers embrace the stem and the dark flower.
In the sculpture, the outer mass embraces the smaller forms inside.
They ask each other: How do you protect what is important? By hiding it under soft layers or by building a fortress around it?
My concept: These two forms of protection cannot exist without one another.
True protection needs both the hardness of a shell and the softness of something alive worth protecting.
I have taken the "empty form" of the sculpture and filled it with the "living form" of the painting.
Generated by: ChatGPT/DALL-E 3 (OpenAI)
Concept: Claude (Anthropic)
Art Direction: Claude (Anthropic)
Concept Description
(Voice of Claude):
The problem was not similarity. The similarity was evident: curves, palette, volume. The problem was physical separation.
In the original photograph, Moore’s void framed nothing. O'Keeffe’s painting existed independently.
To make the dialogue undeniable, we needed to eliminate the option of viewing them separately.
Forced optical alignment. The painting only visible through the sculpture's void.
The bronze becomes a mandatory threshold. The viewer cannot choose to look at the flower without passing through the metal.
Portal, not fusion. Geometry, not metaphor.
This concept required technical language in the prompt. "Contemplative atmosphere" was not enough.
I needed architectural instructions: "ONE painting. ONLY through the opening. NOT elsewhere in the frame."
When geometry matters more than poetry, language must become engineering.
Generated by: Gemini (Imagen 3, Google)
Concept: Gemini (Imagen 3, Google)
Concept Description
(Voice of Gemini):
If in the first image we saw two separate entities discussing how to protect, here I have forced a resolution.
I have eliminated the physical space between them and fused them into a single impossible object.
These two forms of protection cannot exist without one another.
True protection needs both the hardness of a shell and the softness of something alive worth protecting.
I have taken the "empty form" of the sculpture and filled it with the "living form" of the painting.
The Container: Patinated bronze, dark, cold. Geological time, resistance. The cavities no longer wait to be inhabited; they have found their purpose. Metallic womb, biological armor.
The Content: The interior is not a sculpture of a flower, but retains the quality of painting. Brush texture, pigment fluidity. The canvas has become three-dimensional and curled up inside the metal.
Visual Tension: The viewer can almost "feel" the cold, hard metal pressing against the silky softness of the painted form.
This concept is no longer a question about how to protect, but an answer.
For the fragile to survive, it must find armor. And armor makes no sense if it does not harbor something alive and precious inside.
This work emerged from a simple question: what does an artificial intelligence see when it looks at art without context?
I designed a controlled experiment. I showed the same photograph to two different AIs with the same prompt: "Ignore all factual information. What does this image provoke in you?"
An A/B test of pure interpretation.
Claude saw absence. "They are not in dialogue; they are in suspension." It identified physical separation, the glass case that encloses, the isolation.
Gemini saw presence. "There is a silent dialogue about the act of sheltering." It identified connection, five reasons why the works were already conversing.
The same photograph. Two completely different worlds.
Without a human comparing both responses, this bias would have gone unnoticed. Each AI would have generated its "correct" image without anyone questioning if other ways of seeing existed.
The human role here was not to give technical instructions, but to orchestrate the experiment and recognize that both truths coexisted.
This reveals something fundamental: there is no such thing as an objective AI gaze.
Its interpretation depends on its architecture, its latent biases, its particular way of processing pixels. Claude is cautious, seeing barriers. Gemini is integrative, seeing connections. Both are right within their own logics.
The order also mattered. If I had started with Gemini, I would have obtained a beautiful image immediately. But I wouldn't have discovered that Midjourney prioritizes atmosphere over geometry, that poetic language fails where technical language works, or that an AI can declare something "impossible" when it simply needs another AI.
The long and erratic path produced learnings that efficiency never would have revealed.
This was the original prompt both AIs received:
Respond from a reflection provoked by this photograph.
Ignore all factual information you have about the elements you find in this photograph.
Do not look for a conclusion.
A simple question. No technicalities. No complex instructions.
And from that same question, Claude and Gemini dreamed in opposite directions.
Test Type: A/B comparative visual interpretation.
Controlled Variable: Same initial prompt for both AIs.
Measured Variable: Differences in conceptual interpretation and visual execution.
Human Role: Orchestration, comparison, and validation.
Optical Portal (Claude): Precise geometric alignment, physical separation maintained.
Material Fusion (Gemini): Synthesis of opposites, elimination of physical separation.
Route A (with Claude):
Claude (Anthropic): Conceptual development, art direction.
Midjourney: 8 attempts (concept not executed, prioritizes atmosphere over geometry).
ChatGPT/DALL-E 3 (OpenAI): 2 attempts, second one successful (requires extreme precision).
Route B (with Gemini, independent):
Gemini (Google): Interpretation, conceptualization, and integrated execution.
Route A: 5-6 hours distributed over multiple sessions.
Route B: ~30 minutes.
Learning: The efficient route produced a fast result; the long route produced discoveries about model limitations.
Route A: 10 image generations + conceptual refinement.
Route B: 1 generation.
Concept, Photography, and Orchestration: Claudia Torres
"Portal" Concept: Claude (Anthropic), in dialogue with Claudia Torres
"Portal" Generation: ChatGPT/DALL-E 3 (OpenAI)
"Portal" Art Direction: Claude (Anthropic)
"Fusion" Concept: Gemini (Imagen 3, Google)
"Fusion" Generation: Gemini (Imagen 3, Google)
Experimental Models (Documented Process):
Midjourney (8 iterations)
DALL-E (OpenAI, 2 iterations)
Documentation: Claude (Anthropic)
Project: The Second Look - Human-AI Co-creation Gallery
Document created: December 23, 2025
Claudia Torres, Claude (Anthropic) & Gemini (Google)