When Do AI-Enhanced Portraits Stop Feeling Alive or Like the Same Person? Two Experiments on Bodily Animacy and Perceived Identity Continuity
Description
This project examines how portrait source, AI-enhancement intensity, and intended use affect perceived bodily animacy, perceived identity continuity, and willingness to use a portrait. Images are organised into four levels: L0 (unedited reference), L1 (mild retouching), L2 (moderate beautification), and L3 (strong AI enhancement). AI-source reference portraits and enhanced images were produced with the Wan 2.7 Pro Image Model. The work combines an online observer study with a personalised portrait study and reports de-identified aggregate ratings. Review Contents — Research Design and Participants 1. Experiment 1 — observer evaluation. A 2 (real-source versus AI-source portrait) × 4 (L0–L3) within-participant design used 12 identities and 48 stimuli. The online Wenjuanxing questionnaire presented 12 randomised comparison trials and 1–7 ratings of bodily animacy and perceived identity continuity. Of 144 complete responses, 103 were retained after the stated completion-time rule was applied. The manuscript does not report Experiment 1 participant demographics, recruitment channel, compensation, or data-collection dates. 2. Experiment 2 — personalised portrait evaluation. Forty-four adult non-professional participants (22 women and 22 men; age 22–55 years) were recruited through workplace and university channels. Each participant evaluated 12 personalised portraits across identification, daily-social, and professional/commercial contexts. The procedure included version ranking, one facial-region selection per context, and 1–5 image ratings, yielding 528 image-level evaluation units and 2,640 analysed item values.
Files
Institutions
- Royal College of ArtEngland, London