At Rise of AI 2025 in Berlin, artist and AI creator Stefanie Krüll traces her own journey with Midjourney since 2022, from glitchy early outputs to today's photorealistic images, and shows recurring series built around high-gloss surfaces, contrasts of color and texture, and a chimera-themed video narrated with lines from Baudelaire and scored using Kling AI and Suno. She describes training her own small Flux model on twelve photos of herself to generate an imagined version of her life from childhood to old age, a project shaped by her father's death at 38, and argues the emotional impact she felt watching it shows AI-assisted work is not devoid of feeling. She walks through a history of computer art, from Michael Noll's coining of the term at Bell Labs through Vera Molnár, Harold Cohen's AARON, Refik Anadol, the GAN-generated portrait 'Edmond de Belamy' sold at Christie's, Mario Klingemann's self-generating portrait machine, and the robot artist Ai-Da, before naming three contemporary generative artists she follows. A Q&A closes on whether AI can create art, with Krüll and an audience member agreeing that the human artist's intent, not the tool, is what makes the result art.
Key takeaways
- Krüll trained her own small Flux model on twelve photos of herself to generate an imagined version of her life from childhood to old age.
- She traces the term 'computer art' to Michael Noll at Bell Labs, coined after a printer malfunction produced patterned lines that fascinated him.
- Harold Cohen's drawing program AARON raised early questions about whether creativity belongs to the artist or to the autonomous program.
- The AI portrait 'Edmond de Belamy,' made by the collective Obvious with a GAN, sold for more than $430,000 at Christie's.
- Krüll argues AI is a tool an artist uses to express a message, not an independent creative agent, unlike the artist.
Detailed notes
From glitchy generations to photorealism
Krüll describes discovering Midjourney in 2022, initially rejecting it as unusable because of glitches and fragmentation, before returning once the tool improved. She shows a morphing video comparing the same prompt across Midjourney versions one through seven (2022-2025), noting the tool became usable for her work around version four and is now highly photorealistic, which she says brings creative possibilities alongside the risk of deepfakes, a risk she distances her own work from by focusing on surrealistic fantasy worlds.
Recurring series: color, texture and the chimera motif
She walks through several series: high-gloss surfaces combining patisserie imagery with body fragments; an earlier matte series of 'powerful strong women' using candy colors against a strong straight gaze; a headdress series symbolizing self-confident women; and a 12-image collage series contrasting near-photorealistic black-and-white figures with vibrant floral and geometric detail. A separate video series built on the theme of chimeras draws on Charles Baudelaire's poem 'Every Man His Chimera,' with narration and Kling AI-generated sound.
Training a personal model to imagine a lifetime
Krüll trained her own small Flux model, from Black Forest Labs, on twelve photos of herself and used a trigger word to generate new images of herself, then prompted an imagined version of her own life from childhood to old age. She says some details, like a lifelong interest in music, are true, while others, like having siblings, are invented. The project is tied to her father's death at age 38, when she was 11, which left her wondering how old she would get; she says the emotional response the generated video gave her, describing it as giving her goosebumps, shows AI-assisted work can carry genuine emotion.
A short history of computer art
Krüll traces computer art to Michael Noll, Frieder Nake and Georg Nees at Bell Labs, recounting how Noll coined the term 'computer art' after a colleague brought him a page of randomly patterned lines from a malfunctioning printer. She covers Vera Molnár, who moved from a self-built 'machine imaginaire' to using computers as a translator for her art; Harold Cohen, whose program AARON generated drawings via a self-built drawing machine and raised early questions about whether the artist or the program is being creative; Refik Anadol, who won the German Design Award in 2017 for 'Infinity Room' and projects AI-generated imagery onto buildings by architects like Frank Gehry and Zaha Hadid; the French collective Obvious, whose GAN-generated portrait 'Edmond de Belamy,' trained on about 15,000 art-history images and titled as a tribute to GAN inventor Ian Goodfellow, sold for more than $430,000 at Christie's; Mario Klingemann's 'Memories of Passersby I,' a machine that continuously generates and observes its own portraits of non-existing people in real time; and the humanoid robot artist Ai-Da, whose portrait series 'AI God' depicts Ada Lovelace and Alan Turing.
Three contemporary generative artists
Krüll names three current artists she admires: an artist known for knitted 'anatomy sweaters' showing internal organs alongside surrealist short videos; im.ai from Singapore, who chains AI video clips using a first-frame-last-frame technique in an illustrative, comic-like style critiquing social problems to reach younger audiences; and Kelly Bush from the US, whose realistic surrealist scenes Krüll compares to Salvador Dalí.
Q&A: is AI art, art?
Asked whether the social-critique imagery is appropriate to show teenagers, Krüll says she would use it, citing positive experience discussing AI ethics with young people in past workshops. An audience member reframes the closing question from 'is AI art art' to 'can AI create art,' arguing that AI itself has no purpose or message, only the human artist using it does, and compares the debate to early dismissals of digital and phone photography as not 'real' photography. Krüll agrees, and adds that trying the tools firsthand reduces fear of them.
Quotes
Information kills fear.
Stefanie Krüll
For a program to be fully creative in a more complete sense, it has to be able to modify its own performance. And that's a very difficult problem.
Harold Cohen
You need to be an artist to create art, no matter what the tool is.
an audience member
Questions answered in this talk
Who coined the term 'computer art'?
According to Stefanie Krüll at Rise of AI 2025, Michael Noll at Bell Labs is credited with coining the term, after a colleague brought him a page of randomly patterned lines produced by a malfunctioning printer; Noll worked alongside Frieder Nake and Georg Nees as early computer-art pioneers.
What was Harold Cohen's AARON?
Per Krüll's talk, AARON was a program Harold Cohen wrote that generated drawings autonomously through a self-built drawing machine, raising early questions about whether authorship and creativity belong to the artist or the program itself.
How much did an AI-generated portrait sell for at Christie's?
Krüll states that the French collective Obvious trained a GAN on about 15,000 art-history images and sold the resulting portrait, 'Edmond de Belamy,' whose title is a tribute to GAN inventor Ian Goodfellow, for more than $430,000 at Christie's.