Konferenzbeitrag
Transferscope – Making Multi-Modal Conditioning for Image Diffusion Models Tangible
Vorschaubild nicht verfügbar
Volltext URI
Dokumententyp
Text/Conference Paper
Zusatzinformation
Datum
2024
Autor:innen
Zeitschriftentitel
ISSN der Zeitschrift
Bandtitel
Verlag
Gesellschaft für Informatik e.V.
Zusammenfassung
The significance of artificial intelligence (AI) is progressively amplifying for designers, especially within the domain of human-computer interaction. For design students, a foundational comprehension of machine learning (ML) algorithms is indispensable to navigate and utilize this technology in both theoretical and applied contexts - in order to leverage it within design proposals, and also within the design process. Generative AI Tools have rapidly entered creative processes of designers and artists alike, and have been heavily adopted by lay people. They have been praised for democratizing high-quality image creation. However, there are still concerns about the limited artistic control and steerability they provide , especially for professional creatives. This raises questions about how well these tools can be integrated into carefully developed creative workflows, given the constraints on composition and detail. Additionally, text2image algorithms are highly competitive with more manual creation and visualisation techniques in terms of speed and fidelity, while lacking opportunities for deliberation and fine-grained control. As a physical artifact, Transferscope attempts to tangibly introduce professional designers and students to generative AI powered workflows that facilitate creative control, while maintaining the option to leverage serendipity-driven iteration uniquely made possible by the instant-availability provide by image generation models like stable diffusion.Transferscope serves an educational purposes within an experiential teaching approach, and has been designed to work within exhibition and classroom settings alike.