BACKGROUND
Mental health issues are prevalent and challenging both on a personal and societal level. Art therapy has been a recognized method of treatment, offering patients an expressive and non-verbal way to process emotions. With advancements in digital technologies, integrating artificial intelligence (AI) into therapeutic practices has the potential to enhance accessibility and effectiveness. However, the application of generative AI within art therapy remains underexplored, necessitating further investigation into its technical and ethical challenges.
OBJECTIVE
This study aims to develop and showcase a novel technical design that integrates generative AI into art therapy, specifically focusing on the use of text-to-image models. The goal is to provide a supportive tool that enhances patient expression and creative customization, while preserving the role of the therapist in guiding the therapeutic process.
METHODS
We define a simplified art therapy session process and demonstrate how it can be augmented with generative AI. Our technical implementation leverages edge detection, sketch-to-image models, and inpainting techniques to enable patients to refine their artwork through text prompts. We qualitatively evaluate the system using three illustrative examples: a digital sketch, a painted image, and a photograph of a sculpture.
RESULTS
The system successfully generated, refined and adapted versions of the exemplary artworks for all three cases, demonstrating its ability to maintain the essence of the original input.
CONCLUSIONS
The integration of generative AI into art therapy offers significant potential for enhancing accessibility and expressive possibilities for patients. While this proof-of-concept highlights technical feasibility, future research is required to assess therapeutic efficacy, user acceptance, and address ethical concerns, such as bias in AI-generated content. Our approach offers a foundation for further investigation into combining AI with traditional therapeutic practices, supporting the development of extended applications in clinical settings. Our implementations are publicly available at https://github.com/BFH-AMI/sds24.
