This paper presents folk theories of text-to-image (T2I) models as understood by artists in their creative practice. Through reflexive thematic analysis of data from a workshop with 15 artists from 10 countries, we highlight three high-level sets of folk theories related to using T2I models, their potential harms, and harm reduction strategies. These theories are: (1) T2I models as an artistic medium with information-rich properties, (2) T2I models as a mundane tool for prototyping and communication, and (3) T2I models expanding creative expression. Theories of harm articulate T2I models as harmed by engineering efforts to eliminate glitches and product policy efforts to limit functionality. Theories of harm-reduction orient towards protecting T2I models for creative practice through transparency and distributed governance. We examine how these theories relate, and conclude by discussing how folk theorization informs responsible AI efforts. The study reveals a wide range of beliefs and normative expectations among ML-artists. They articulate "creativity" as innovative use that exceeds basic model affordances. They frame T2I models as a medium to incorporate into their personal art practice or as a collaboration tool for client-based work. They theorize T2I as something harmed by engineering efforts to eliminate glitches and product policy efforts to limit functionality. This folk theory is directly informed by their beliefs that T2I models are an important medium and tool for creative practice, necessitating their protection. It also shapes their beliefs about harm-reduction, which orient towards protecting T2I models for creative practice through transparency and distributed governance. The study also highlights the importance of understanding the situated knowledge and beliefs of different artist communities. It underscores the need to examine folk theorization across technological use, harm, and harm reduction dimensions. These folk theories can inform responsible AI development by calling attention to fundamental questions of harm and the frictions between the values encoded in algorithmic systems and those held by communities.This paper presents folk theories of text-to-image (T2I) models as understood by artists in their creative practice. Through reflexive thematic analysis of data from a workshop with 15 artists from 10 countries, we highlight three high-level sets of folk theories related to using T2I models, their potential harms, and harm reduction strategies. These theories are: (1) T2I models as an artistic medium with information-rich properties, (2) T2I models as a mundane tool for prototyping and communication, and (3) T2I models expanding creative expression. Theories of harm articulate T2I models as harmed by engineering efforts to eliminate glitches and product policy efforts to limit functionality. Theories of harm-reduction orient towards protecting T2I models for creative practice through transparency and distributed governance. We examine how these theories relate, and conclude by discussing how folk theorization informs responsible AI efforts. The study reveals a wide range of beliefs and normative expectations among ML-artists. They articulate "creativity" as innovative use that exceeds basic model affordances. They frame T2I models as a medium to incorporate into their personal art practice or as a collaboration tool for client-based work. They theorize T2I as something harmed by engineering efforts to eliminate glitches and product policy efforts to limit functionality. This folk theory is directly informed by their beliefs that T2I models are an important medium and tool for creative practice, necessitating their protection. It also shapes their beliefs about harm-reduction, which orient towards protecting T2I models for creative practice through transparency and distributed governance. The study also highlights the importance of understanding the situated knowledge and beliefs of different artist communities. It underscores the need to examine folk theorization across technological use, harm, and harm reduction dimensions. These folk theories can inform responsible AI development by calling attention to fundamental questions of harm and the frictions between the values encoded in algorithmic systems and those held by communities.