AI image generation tools are transforming the beauty industry by offering new possibilities for visualizing skincare massage types through realistic image generation from text-based prompts. This study aims to explore this potential by selecting six massage types based on their practicality and visual representation potential. Three skincare professionals with over 10 years of experience identified experts Swedish Massage, Foot Massage, Bamboo Therapy, Aromatherapy, Thai Massage, and Stone Therapy, and images were generated using DALL·E, an AI-based image model. A total of 24 images (4 per framework) were generated for analysis. Images were created by applying Khizer Abbas framework types, R-T-F(Role-Task-Format), T-A-G(Task-Action-Goal), C-A-R-E(Context-Action-Result-Example), and R-I-S-E(Role-Input-Steps-Expectation). The application of these frameworks resulted in distinct image generation focuses, with each framework emphasizing different key elements. RTF highlighted the therapist’s role and treatment environment, while TAG effectively depicted massage techniques and relaxation states. CARE prioritized the treatment setting through example-based visuals, creating a harmonious representation of the environment and session, whereas RISE focused on step-by-step procedures and professional expertise, ensuring a structured and comprehensive visual portrayal of massage treatments. AI models faced limitations in depicting physical contact and specialized roles, but reconfiguring settings and tools improved image generation. Each framework emphasized specific elements, such as the therapist’s role, techniques, environment, and effects, allowing for a comparative analysis of visual representations across massage types. The main contribution of this study is the comparative analysis of four visual prompt frameworks applied to AI-generated skincare massage images. Further research on effective prompt-writing standards and methodologies for the beauty industry could expand AI’s applicability in beauty, education, and design fields.
Park et al. (Tue,) studied this question.