1&∞CHAIRS
2023
Generation Model: v1-5-pruned-emaonLy
Prompt: a single chair on a white background
Sampling Method: DPM*+ 2M SDE Karras
Sampling Steps: 30 CFG Scale: 7
Initial images of chairs were generated using the Stable Diffusion Ai model, based on the prompt: a single chair on a plain background.
This dataset, of photo-realistic images of a wide variety of chairs, was then used to train the Stable Diffusion model again, extending its knowledge capacity of what a chair on a plain background can look like.
This process of re-training the model on its own generated imagery was repeated again and again. Until, at the 6th iteration, instead of photo-realistic images of chairs, as seen in the initial step, the model produced colorful digital noise in which any resemblance to the represented subject, a chair, would fade completely.
In data science, such phenomena are often referred to as data-cannibalism. Through the necessity to augment datasets and due to AI image and data generation’s increasing and insidious prevalence, more and more new Ai systems will be trained on synthetic datasets, produced by generative Ai models, thus posing ontological challenges and poisoning future datasets.
The epistemic accuracy of the popular Ai models, through a process of necessary and incidental feedback loops produce an echo chamber of auto-generated and consumed data where the domain ontology of a subject and its visual representation decay into non-figurative abstraction, at least for a human eye.
Tokyo | Berlin | Vienna
Contact Email
mail[at]kraft.studio
Press Materials:
Copyright
Kraft Studio © 2023
Egor Kraft
Artist, Director & Founder
Anna Kraft
Production & Communication
Artem Konevskikh
Ai Research & Development