Abstract

This article critically examines the design philosophies of contemporary creative AI, contrasting large-scale, prompt-driven generative platforms with open-ended, modular toolkits. We argue that the former, which we posit as a ‘black box’ system, while promoting convenience, configures artists as passive consumers of the outputs, limiting agency, transparency and deep engagement. As a counterpoint, we explore the Fluid Corpus Manipulation (FluCoMa) project, an open-source toolkit that empowers artists as active craftspeople. Our comparison draws on three vignettes from the project’s workshop curriculum used in teaching and dissemination: building a sound classifier, an interactive visual sound-bank map, and training a neural network that translates low-dimensional controls to higher-dimensional controls. This analysis teases out a wider space of possibilities for creative AI, advocating for an ecosystem where integration between high-level and immediate generative tools and lower-level machine-learning frameworks is viable, fostering a continuum of practice. To propose a vision for the future, we point at success in this and other open-source projects that we see as championing a sustainable model of development and artistic inquiry.