Hybrid Artefacts
Transforming Abstract Drawings into Generated Architecture
With the rise of generative tools, there has been sudden confusion regarding the basic understanding of what Art & Design Artefacts are and may become in the near future. In these turbulent times, it has become apparent that some creatives, eager to use these generative tools, are beginning to create things that cannot be accurately categorised within the known realm of Art & Design Artefacts. A new sub-category of artefact has emerged: non-physical objects that are dematerialised yet appear real, forming a distorted spatial realm. These artefacts, which do not exist in pure reality, are partly human, partly fictional, echoing the sounds of an imaginary Piranesian playground.
The generated output initially appears as simple sensory impressions, lacking the original authorship of a human creator. Subsequently, a {Human} touch — automatic drawings, coded images with unknown {hidden content} — is incorporated and used as a synthesising guide. Through pre-defined, repetitive instructions, the AI image generator transforms the abstract drawings into an ensemble set of images. The result is a carefully selected body, a curated slice of the whole, shining through these abstract, non-representational experiments.
{Human} drawings and {Artificial}generations seem unrelated. Drawings are human, and generations are artificial. But when we synthesize the two, we find ourselves in a contradiction. What seemed distinct and separate has suddenly fused into a chaotic, yet occasionally pleasing, entanglement of forms. The decoded output {generated data} reveals instances of artistic expression which, at irregular intervals, transform the abstract input into more elaborate organic three-dimensional structures. These are Hybrid Artefacts, born and co-authored at the intersection of human intuition and machine logic, occupying a liminal domain that reconfigures traditional notions of creative agency.
Central to this research is the calibration of surprise, operating at the edge of generative chaos and predictable control. By systematically varying parameters such as classifier-free guidance (CFG) and sampling steps, while holding others constant, the research identifies a narrow operational band where, among predominantly chaotic generations, key frames emerge. These are moments where transformations achieve spatial coherence while remaining visually unpredictable. Identifying and extracting these rare coherent generations from large batches relies on a trained eye, personal intuition, and subjective aesthetic judgment. Through iterative testing and pattern recognition across multiple generational batches, recurring formal qualities emerge that allow for increasingly intentional navigation of the latent space.
To further narrow the range of latent space in which the generations occur, a personalised ranking {tier list} of generations produced, prioritising the most peculiar results. The parameters and settings intuitively discovered to produce these novel images are then visually coded into a virtual machine: a series of sequentially operating nodes through which the transformations materialise. The resulting script serves as the basis for a more effective and streamlined workflow while still maintaining a desired degree of chaotic behaviour {hallucination}.
A series of exploratory case studies contextualizes the initial findings within the broader state of the art: (1) Constructing speculative narratives around the self-referential Hybrid Artefacts. (2) Replicating the experiment using externally sourced abstract drawings by Hermann Finsterlin and Friedrich Kiesler as input. (3) Employing a double transformation process where abstract drawings are first converted into images of artificial 3D models, then transformed again into actual 3D models materialized as physical 3D-printed objects.
Through systematic curation, individual artefacts converge into coherent series, together they form a unified body of work called Artificial ~ Nature. The research's generative workflow produces organic forms that appear naturally grown rather than digitally constructed. Across their diversity, these artefacts share underlying structural patterns: recurring spatial features and morphological relationships that create a coherent visual language. This collection points toward a future of design that emerges from the fertile intersection where human creative and artificial intelligence meet, the very ground upon which future worlds are built.
(1) Compositional Affects in Generated Images: An Examination
(2) Morphological Transformers: Automatic Drawing into Synthesised Visual Representations
(3) Metamorphix: Curated from a Sampling Set of Personalized Generations
(4) Framework Definition: Exploratory Case Studies
(5) Factor of Surprise, Selection - Analysis and Progression
[Exhibit A] Artificial ~ Nature
Latent space: High-dimensional mathematical space where AI models encode and organize learned information, with similar concepts positioned closer together, allowing the model to navigate between different possible outputs by moving through this abstract territory.
Classifier-free guidance (CFG): A parameter that controls how much the model should follow your text prompt when generating an image.
Sampling Steps: The number of iterations used to denoise a random noise image into a final picture.
Keywords: multidimensional, latent spaces, compositional affects, prompt playing, looping problems, purifiers, deep learning models, generative process, diffusion, distortion, compression, biases in datasets, compilations, fine-tuning, transformers, infinite possibilities, curation, originality.
Ⓜ Marcel Moonen, PhD researcher, Design Based Doctorate, TU Berlin, 2023 - 2026, Hybrid Artefacts [EDUCATIONAL PURPOSE ONLY] M. Production