← Close Project

PORTRAITS IN LATENT SPACE
Spectral self-portraits in ongoing translation

Type: Artwork/ Research
Focus: Spectral self-portraits in ongoing translation
Year: Since 2022

Ongoing since 2022, this work documents the accelerating evolution of generative AI through self-descriptive image prompts. Each iteration reflects not only the state of the models but a refracted version of the self: layered, partial, speculative. As AI develops, so too do its projections—at once intimate and alien. This is not a portrait, but a dialogue across time, code, and representation.

Emerging from the spectrality of latent space, these images visualize—on a deeper level—the ongoing reflection required when engaging with a technology of unprecedented power: a space where identity is neither fixed nor fully visible, but continuously negotiated through algorithmic interpretation—just as the power structures behind it remain opaque and shifting.

“We have surrendered to the benefits of the digital, at the very moment that constant, escalating revelations simultaneously demonstrate their monstrous potential for betrayal.”
— Rem Koolhaas et al., Elements of Architecture (Köln: Taschen, 2018), IL.

In early 2025, the Chinese startup DeepSeek disrupted the global AI landscape with a powerful open-source model developed at a fraction of the cost of its Western counterparts. Just weeks later, OpenAI unveiled its premium-tier offering—locked behind a $200/month paywall—signaling a future in which unprecedented cognitive power is increasingly concentrated in the hands of the wealthy. This asymmetry exposed more than economic disparity: it revealed the ideological core of contemporary AI—where power, access, and influence are not just technical conditions, but deeply political ones.
In the same month, OpenAI Chairman Sam Altman appeared alongside U.S. government officials at the White House launch of Stargate, a $500 billion initiative toward artificial general intelligence (AGI). This convergence of private and state power—aligned with a far-right administration actively dismantling democratic checks and diversity mandates—raised urgent concerns about the consolidation of unprecedented technological power within systems where democracy functions more as a symbolic form than a lived practice. The political affiliations are not incidental; they shape the very architectures of intelligence we interact with, and the ideological frameworks embedded in their design.

Architecture bears political and social responsibility. It demands a critical reflection on the tools we use and the power structures embedded within them. As with anything in this complex world, these structures are often opaque, layered, and difficult to fully perceive.

This process of reflection is ongoing—marked by constant recalibration.

The architectural profession must remain alert to the ethical, technological, and systemic frameworks that shape both its tools and its outcomes, in a sustained effort to understand—and reimagine—the world we live in.

To engage with AI as a creative tool is to engage with its latent conditions. This work visualizes that engagement: a recursive process in which critique and creation are inseparable. As the models shift, so too must our understanding of authorship, agency, and representation.

The technology marks a quantum leap in possibility—particularly in grasping the hyper-complex correlations of a globalized and digitalized world, and in aiding decision-making within it. But by profoundly adding to this complexity at a pace that defies comprehension, its utilization becomes paradoxical and self-contradictory. Its presence is irreversible. The question now is not whether we use it, but how—and whether we can remain attuned to what remains invisible beneath its surface.

In September 2025, image generation reached a level of precision that allows facial features to be iteratively adapted with enough detail to recreate likeness. From this point onward, the model in this project is instructed in my actual appearance and capable of recreating it with accuracy. The resulting images are not filtered versions of existing imagery but original creations produced by the trained model. Whereas filters — whether conventional or AI-based — transform existing material, this generative model synthesizes entirely new imagery from learned representations, marking a technological shift from modification to creation.

09/2025

09/2025

07/2025

07/2025

07/2025

07/2025

07/2025

07/2025

03/2025

03/2025

03/2025

04/2024

04/2024

04/2024

04/2024

02/2024

11/2023

09/2023

09/2023

04/2023

01/2023

01/2023

01/2023

08/2022

08/2022

08/2022

08/2022

07/2022

07/2022

07/2022

07/2022

05/2022

05/2022

05/2022

05/2022

05/2022

05/2022

05/2022

05/2022

05/2022

05/2022

03/2022

03 /2022

03/2022

03/2022

03/2022

03/2022

← Close Project

↑ Top