April 4-6, 2024 • Hyatt Regency • Lexington, KY
Innovations in Health Communication
Innovations in Health Communication
Abstract: Feasibility of AI-Generated Imagery for Targeted Health Communication Messaging
Background: AI-generated imagery has reached the forefront of media, with large amounts of news coverage. While it may be considered to be a fun electronic toy, there are also practical implications of AI-generated imagery in the field of health communication, such as extremely targeted and tailored imagery. However, research in the role of AI-generated imagery in health communication is lacking.
Purpose: The purpose of this study is to determine the feasibility of using OpenAI’s DALLE-E 2 to create highly customized images for use in health communication material. The study utilized DALL-E 2 to generate tailored and customized images for 12 different news articles. Generative AI allows for highly customized images to be rapidly generated without a large number of resources or equipment.
Methods: As part of a larger study, AI generated imagery was created for each of the 12 (3x2x2) experimental factors. The factors were emotional (fear, hope, neutral), homophily vs heterophily, and fact vs fear. Each of the 12 experimental factors had an associated news article with two images, one of a billboard and another of either a person’s face (narrative) or a generic scene (fact). The face or scene images were generated by DALL-E 2.
Results: Dall-E v2 was able to successfully generate the images. The model’s scene-based images were superior to those of people’s faces. Some of the 12 required multiple generations, with several modifications to the image generation prompt. DALL-E 2 did, in most cases, struggle in the generation of face(s), with many being deformed or extremely unrealistic. Overall, DALL-E 2 was able to provide usable images for the news article.
Conclusion: AI-generated imagery is a feasible method of highly customized media creation for health communication material. It provides a practical and adaptable tool, with AI-generated imagery having the potential to be extremely innovative in targeted and tailored health communication. The proliferation of open-source AI image generators, such as Stable Diffusion, and updated closed-source, proprietary image generators, such as DALL-E 3, will continue to improve the capability of image generation. One marked improvement in DALL-E 3 is the generation of realistic faces, which the previous generation of image generators struggled with. The improvements will allow for more lifelike images with unparalleled customizability, in addition to ease of use of the generators, with fewer required prompt modifications.
Purpose: The purpose of this study is to determine the feasibility of using OpenAI’s DALLE-E 2 to create highly customized images for use in health communication material. The study utilized DALL-E 2 to generate tailored and customized images for 12 different news articles. Generative AI allows for highly customized images to be rapidly generated without a large number of resources or equipment.
Methods: As part of a larger study, AI generated imagery was created for each of the 12 (3x2x2) experimental factors. The factors were emotional (fear, hope, neutral), homophily vs heterophily, and fact vs fear. Each of the 12 experimental factors had an associated news article with two images, one of a billboard and another of either a person’s face (narrative) or a generic scene (fact). The face or scene images were generated by DALL-E 2.
Results: Dall-E v2 was able to successfully generate the images. The model’s scene-based images were superior to those of people’s faces. Some of the 12 required multiple generations, with several modifications to the image generation prompt. DALL-E 2 did, in most cases, struggle in the generation of face(s), with many being deformed or extremely unrealistic. Overall, DALL-E 2 was able to provide usable images for the news article.
Conclusion: AI-generated imagery is a feasible method of highly customized media creation for health communication material. It provides a practical and adaptable tool, with AI-generated imagery having the potential to be extremely innovative in targeted and tailored health communication. The proliferation of open-source AI image generators, such as Stable Diffusion, and updated closed-source, proprietary image generators, such as DALL-E 3, will continue to improve the capability of image generation. One marked improvement in DALL-E 3 is the generation of realistic faces, which the previous generation of image generators struggled with. The improvements will allow for more lifelike images with unparalleled customizability, in addition to ease of use of the generators, with fewer required prompt modifications.