As a sociotechnical practice at the nexus of humans, machines, and visual culture, text-to-image generation relies on verbal prompts as the primary technique to guide generative models. To align desired aesthetic outcomes with computer vision, human prompters engage in extensive experimentation, leveraging the model’s affordances through prompting for style. Focusing on the interplay between machine originality and repetition, this study addresses the dynamics of human-model interaction on Midjourney, a popular generative model (version 6) hosted on Discord. It examines style modifiers that users of visual generative media add to their prompts and addresses the aesthetic quality of AI images as a multilayered construct resulting from affordance actualization. I argue that while visual generative media holds promise for expanding the boundaries of creative expression, prompting for style is implicated in the practice of generating a visual aesthetic that mimics paradigms of existing cultural phenomena, which are never fully reduced to the optimized target output.

中文翻译：

---

超越魔力：在视觉生成媒体中推动风格作为可供性实现


作为人类、机器和视觉文化之间融合的社会技术实践，文本到图像的生成依赖于口头提示作为指导生成模型的主要技术。为了使所需的审美结果与计算机视觉保持一致，人工提词器进行了广泛的实验，通过提示风格来利用模型的可供性。本研究侧重于机器原创性和重复性之间的相互作用，解决了 Discord 上托管的流行生成模型（第 6 版）Midjourney 上人类与模型交互的动态。它研究了视觉生成媒体用户添加到其提示中的样式修饰符，并将 AI 图像的美学质量作为可供性实现产生的多层结构来解决。我认为，虽然视觉生成媒体有望扩大创意表达的边界，但激发风格涉及产生一种模仿现有文化现象范式的视觉美学的实践，这些美学从未完全简化为优化的目标输出。