Art therapy is an excellent treatment for managing mental health, especially focusing on self-reflection, self-efficacy, and self-esteem. However, professional psychotherapy remains under-represented, and many individuals are deterred from pursuing art therapy due to a lack of artistic skills. This article introduces MIRA, a generative AI agent designed to extend the Squiggle Game into a self-support tool for everyday use. By combining sketch-based collaborative drawing with reflective feedback, MIRA lowers the barrier to creative engagement while opening up new opportunities for self-support.

We conducted a controlled study with 30 participants, comparing three conditions: an AI agent (MIRA), human-only curved drawing, and independent drawing. Analysis revealed that participants in the AI condition reported significantly higher levels of self-reflection, self-efficacy, and self-esteem compared to those in the control condition. Interviews revealed that users perceived MIRA not only as a skill compensator that alleviated performance anxiety, but also as a co-creator who encouraged more ambitious expression and a reflective partner who induced deeper reflection. Participants also emphasized the novelty and engagement of the step-by-step generative process, describing it as both engaging and healing. At the same time, the findings revealed a mixed attitude among users toward AI interventions: some participants valued the objective insights of AI, while others questioned its depth of empathy. These complex perceptions highlight the importance of carefully positioning AI agents as supportive partners rather than replacements for therapists.

In summary, this study presents the first empirical analysis of a self-support AI agent adapted from Squiggle Games. Our work demonstrates how generative AI can expand the inclusiveness of art therapy by lowering barriers to entry, supporting reflective practice, and fostering creative confidence. Finally, we draw implications for the design of reflective AI systems that st