Author(s): Jason Logan, Dijana Handanovic & Andrew Kudless
“Re-cognition” is a term coined by Professor Douglas Graf in his 1986 essay, “Diagrams.”1 This paper uses Graf’s essay to frame an argument for the use of artificial intelligence (AI) as a collaborative sketching and conceptualization tool. Graf argues for a loose, hybrid, and flexible form of diagrammatic practice as an alternative to what he called the “anxiety of origin,” and a concern for “the ‘proper’ manifestation of type” that characterized much of postmodern formal analysis. Rather than valuing recognition, Graf argues for the genera¬tive capacities of “re-cognition” through transformational diagrams. G raf supports his argument through a series of revisionist diagrams that remove a building “from the firm position of artifact to the more uncertain territory of artifice.” He goes on to write, “[a]s the object of analysis, the building is reduced to information, information is reduced to invention, and invention to interpretation.”2 Like Graf’s notion of re-cognition – to come to know or understand again – this paper considers the “re-cognitive” potential of AI diffusion models. It focuses on the similarly loose, hybrid, and flexible nature of using diffusion models to digitally sketch, transforming images from artifact to artifice by leveraging large datasets of information for inven¬tion and reinterpretation. This paper documents a process using diffusion models, in coordination with neural network control structures like ControlNet, as a generative sketching and conceptualization technique. Where many text-to-image models are difficult to control, ControlNet allows designers to use images, drawings, and depth maps to guide the diffu¬sion model’s output. This collaborative sketching technique augments the text prompt process, allowing greater control over the images generated by the AI model. This paper considers the value of a digital workflow that, like a hand sketch or a transformational diagram, is immediate and unburdened by the specific inputs and accuracy required by other digital tools.
Volume Editors
ISBN
978-1-944214-49-4