Key Points
We combine posture prompts with text instructions to generate images of digital human applicants for interviews. Through subjective assessments by participants acting as interviewers, we investigate how the generation conditions relate to first impressions.
Abstract
Imagine an applicant in a virtual interview whose appearance is created by a computer. Such a computer-created person is called a digital human. We investigated whether instructions about posture, combined with additional written instructions, change an interviewer's first impression of an AI-generated image of this applicant.
Instructions given to an image-generating AI are called prompts. We created applicant images under different conditions: with or without a posture prompt, and with or without an additional text prompt. Participants viewed the images while imagining themselves as interviewers in a virtual setting and reported their first impressions.
Images generated using both kinds of prompt received better first-impression ratings from participants acting as interviewers than images generated without these prompts.
Publications
- Wataru Ganaha, Shota Hioki, Masashi Nishiyama,
First impression assessment of digital human applicant images generated with posture prompts and text prompts, Proceedings of 27th International Conference on Human-Computer Interaction (HCII), vol. DHM, pp. 185 - 199, June 2025. [Slides] - 我那覇 航, 日置 翔太, 西山 正志,
姿勢プロンプトとテキストプロンプトとの組み合わせにより画像生成された面接志願者の第一印象の主観評価,
HCGシンポジウム, A-1-3, pp. 1 - 8, December 2024. [Slides]
