Name

_Results

close [x]

Ümüt Yildiz

Germany

Artist Statement

My journey with AI began not with aesthetics in mind, but with disruption. In the early days of image generation, I worked with StyleGAN—models primarily designed for realism, harmony, and the smoothing out of visual anomalies. But I wasn’t interested in perfection. I was drawn to the seams, the glitches, the tension between control and collapse. I used those models against their original purpose—deliberately pushing them into territory where they broke, bent, or revealed something they weren’t meant to. I didn’t want idealized faces or flawless scenes. I wanted ghosts in the machine, systems turned inside out. That same spirit drives my current work with intelligent image-making systems. I treat AI not as a tool for enhancement, but as a collaborator in critique. Today’s models are more advanced, more “real,” but that realism often conceals deeper fictions—political, emotional, and ethical. My work confronts this by blending hyperrealism with contradiction. I use photorealistic textures to depict absurd, painful, or socially charged scenes—forcing the viewer to ask not just is this real?, but why does it feel real—and who benefits from that illusion? AI, to me, is a paradox: a mirror that doesn’t reflect unless you bend it. I train and direct these models, but my authorship lies in where I point them, what I withhold, and what I insist they reveal. My creative code is shaped by systemic distrust, emotional fatigue, and a refusal to let machines simply replicate the world as it is. Instead, I use them to expose, twist, and archive the world as it hides itself. My artistic truth is this: I don’t make AI art to explore technology—I use technology to expose what we’ve accepted as normal. My images are not meant to be consumed passively. They are meant to stick, to sting, and to remind us that even automated output can carry the weight of human intention—if we choose to wield it that way.

Published in >
The AI Art Magazine, Number 2
Free Speech Liar, AI generation, 2025
Free Speech Liar, AI generation, 2025.
Ümüt Yildiz, Free Speech Liar, AI generation, 2025

Description

I created Free Speech Liar as a response to the violent absurdity of our current political moment—where truth is fractured, language is weaponized, and “free speech” is no longer a right, but a performance. The figure at the center of this work is deliberately contradictory: a masked rider, labeled “LIAR,” waving a distorted “FREE SPEECH” flag while passing through fire on a white horse. The symbolism is blunt, because the crisis is blunt. Around the world—from authoritarian regimes to so-called democracies—free expression is under siege. But more insidious than overt censorship is the way language is hollowed out and repurposed by those in power. We’re told we are free to speak, even as truth is drowned in noise, distraction, and carefully crafted misinformation.

Process

I don’t see AI as a neutral medium. It reflects the logic of the systems that built it—optimization, efficiency, aesthetic coherence. My work breaks that logic. I aim to create images that feel real but make you uneasy. Images that don’t resolve, but disturb. Free Speech Liar isn’t just about a figure on a horse—it’s about the spectacle of false authority, the co-option of democratic language, and the burning ground beneath us all. This is not protest art in the traditional sense. It’s an indictment. Of the media. Of digital passivity. Of our willingness to watch fire and call it light.

Tools

I first began generating training data in ComfyUI, where I created a series of politically charged, semi-surreal images using custom-trained models. These images—filled with contradictory symbolism, hyperreal textures, and layered socio-political critique—formed the initial visual dataset. This dataset was then used to fine-tune a custom instance of Krea’s K1 model via Krea Train, which allows the upload of up to 50 reference images. This second phase of the process transformed the K1 model into a mirror of my aesthetic and thematic vision—while retaining its technical strengths in photorealism, lighting control, and texture fidelity. Using this fine-tuned K1 instance, I iteratively developed Free Speech Liar. I crafted precise prompts embedding real-world political signifiers, manipulated composition and color using Krea’s live editing tools, and refined each iteration with intentionality. The final image was upscaled for clarity, with ambient noise and tonal treatment added to give it the visual texture of decaying media—haunting and seductive.

Image credit:
Essay by
The AI Art Magazine, Number 2
Published in:
The AI Art Magazine, Number 2,
September 2025
WHY?
The magazine reveals what happens when humans and machines get creative together. We’re throwing the doors wide open believe it's going to be a big party—come in and let’s dance.
Media coverage
WHO?
Christoph, Franziska, Ingo, Jason, Jan, Jens, Katrin, Mike, Simone, Tito
© 2025 polardots.studio