A CHI ’24 conference paper by Renee Shelby, Shalaleh Rismani, and Negar Rostamzadeh that investigates how artists who use text-to-image (T2I) models make sense of these systems through “folk theories” of use, harm, and harm reduction. Based on three workshops with 15 artists from 10 countries, it identifies how artists position T2I as both an artistic medium and a pragmatic tool, frame creativity as exceeding basic model affordances, and articulate harms as emerging not only from the technology but also from efforts to “polish” glitches or restrict functionality. The paper argues that transparency and distributed governance are central to harm reduction, offering a grounded lens for responsible AI that is informed by real creative practice.