Instagram story viewer> @hyperfiguration> Posts

Julien Bonet

@hyperfiguration

886

followers

911

following

Visual development artist at Authorsfirst AI studio
Photographer exploring AI models
@photo.edition.berlin for Masquerades series
🇫🇷 France

POSTS STORIES REELS TAGGED

Download All

Holidays in latent space
Model: MJ 6.1

Exploring synthetic landscapes through accumulation, variations on the same prompt. by @hyperfiguration

45

5

a month ago

Latent Space Landscapes
Model: MJ 6.1

The third and final series exploring this region of latent space. I’m not quite sure how to describe it, it looks like a psychedelic sci-fi set made of plastic and papier-mâché, as if training debris had washed up on an artificial equatorial beach. by @hyperfiguration

31

11

a month ago

From latent space
Model: MJ 6.1

Variations on a same prompt.

I love Midjourney 6 for its tendency to drift into strange landscapes when the prompt becomes so complex that the model seems to stretch its statistical limits to the far corners of the latent space. by @hyperfiguration

28

4

a month ago

Latent Cartography
Model: MJ 6.1

AI models are being released one after another, and this one is already over two years old. Yet, for the first time, it felt more relevant for this exploration than recent models, much like reaching for an old camera for its specific render. by @hyperfiguration

56

10

a month ago

Computed class
(Les Krims deep remix)
Model : MJ 6.1 2026

In the 1980s, Leslie Krims produced portraits of the middle class in interiors saturated with objects.

He used real subjects and pushed documentary to such an extreme that it became strange fiction.

AI does the opposite: it manipulates immaterial subjects - words, attributes, averages - and tries to assemble them into a credible documentary.

But as soon as it is saturated with data, the statistical nature of a system disconnected from the world becomes visible: it no longer documents reality, but the recesses of latent space. by @hyperfiguration

50

11

9 months ago

AI and aesthetics of violence
(Model : MJ 6.1 2026)

For this series, I provided the AI with journalistic images of the 2023 French riots (and it can be easily verified that it had already been trained on this type of event, even without input images).

Within this journalistic framework, the riot is predominantly represented as a sensational visual event (clashes, fire, vandalism…) rather than as a political process. The rioter almost systematically appears as the aggressor.

This interpretation is reflected in the French government’s report on the 2023 riots: "The opportunism of shop looters appears as a far more prominent element than the expression of anger or rejection of republican institutions. […] a gigantic outlet for release, in which confrontation with law enforcement takes on a festive dimension."

Based on this documentary material, I generated over 3,000 images using creative prompts (both images and text) that pass these photographs through the latent space filter. 
While the AI can perfectly reproduce a journalistic riot image, as soon as creative perturbations are introduced, a recurring pattern emerges: the subject is shifted into a passive posture. Those usually depicted as aggressors become bodies struck, hit, or off balance.

The overload in the prompt produces a causal asymmetry: the event remains, but the agent disappears. The world acts, the body receives.

In a creative mode, the AI refrains from aestheticizing offensive violence and favors suffered violence, as if refusing to transform the rioter into a heroic figure.

This inversion places the institutional narrative in tension with the reality of state violence, revealing the riots as a response to structural violence rather than merely an outlet for release.

@ferisbarkat by @hyperfiguration

35

1

9 months ago

Memories from latent space
Model : MJ 6.1 2025 by @hyperfiguration

64

1

9 months ago

On disembodiment by AI
(Joël-Peter Witkin deep remix)
Model : MJ 6.1 2025

Following my previous post Aesthetics of Censorship, I was asked about the difference between the discomfort produced by the images of Witkin that I remix and that generated by AI.

In Witkin’s work, the image stems from a desire to grasp a manifestation of the divine in flesh, living or dead, marginal or deformed. Discomfort is not the goal.

My images, by contrast, are a rewriting filtered through probabilities, in which censorship generates a different kind of unease.

Where Witkin’s photography engages real people and genuinely mutilated bodies, AI manipulates only signs.

Without lived transgression or moral exposure, the generated images remain fundamentally hollow if their sole purpose is aesthetic, whether in terms of beauty or discomfort.

My images document neither flesh nor reality, but a system of constraints, calculations, and prohibitions. I use AI to document the latent space. by @hyperfiguration

29

0

9 months ago

Lost resort
(Martin Parr deep remix)
Model : MJ 6.1 2025

In the latent space, a cultural reference is detached from lived experience. It becomes a set of effective, deployable visual correlations.

This is not a loss of memory but a memory without a world: the AI does not understand reality, it reduces it to a mathematical abstraction.

The generated images no longer represent a real territory, they bear witness to a strange new landscape: that of the latent space, where visual references circulate, recombine, and exist independently of the world that produced them.

This disembodiment does not extinguish cultural references, it grants them an unprecedented afterlife, transforming them into materials continuously activated and redirected by the new intentions of those who deploy them. by @hyperfiguration

55

2

9 months ago

Memories from Latent Space

Variations on the Same Prompt
(Model: MJ 6.1 2025)

AI are trained on datasets that can contain several billion images, partly sourced from the web, social networks, and our apps.

From our faces or vacation photos, they generate memories without a subject: statistical remnants of private images aggregated into plausible settings.

Beneath the apparent anonymity and strangeness of the latent space, a few human traces remain, a fabricated nostalgia based on real memories. by @hyperfiguration

54

8

9 months ago

Datapainting : faces I II III (2025)

Variations on the same prompt.

Exploration of Midjourney's latent space 6.1.

AI: The Mirage of Creation?

During their training, AI models build an internal representation (latent space), a certain statistical vision of the world based on billions of images.

These are feature spaces with several hundred dimensions where each point is a possible image.

The latent space therefore virtually contains all the images that can be generated by combinations of learned features. As if the images were already there mathematically, and we were extracting those that deserve to be seen.

What if the artistic gesture in the age of AI was as much about observing as truly creating?

The model doesn't contain images, but rather fields of possibilities guided by the prompt. That said, anyone who has seriously delved into generation has already had this feeling of "discovering" a work.

But these territories are ephemeral. With each update, the latent space is reshaped. Creating with AI also means mapping regions that will soon be replaced by new geographies of the possible.

#aiartwork #aiartist #generativeart #promptography #hyperfiguration #datapainting #latentspace #midjourney by @hyperfiguration

50

11

a year ago

About creativity in AI.

Datapainting : fil rouge IV V VI (2025)

(red thread is the driving element chosen by AI in the middle of a complex prompt, over thousands of generations) by @hyperfiguration

82

6

a year ago