Audience appraisal of AI generated visual media across viewers and scenes

How do audiences relate the authenticity, emotional meaning, trustworthiness, and acceptability of AI-generated visual media? For cultural exhibitions and other uses of reconstructed pasts, emotional appeal and perceived credibility need not coincide, and audience responses may differ across scenes. We examine these questions through an appraisal-oriented account of AI-generated reconstructions of urban everyday life. Study 1 ( N = 513) used Mode-A composite PLS-SEM with 24 items representing eight constructs to estimate viewer-level evaluation associations. Ten of eleven hypothesized paths were supported. Nostalgic response was associated with acceptance intention, but its hypothesized positive association with trust was not supported after perceived authenticity and AI traces were considered. Study 2 independently compared baseline and treated versions of 12 scenes using a nonrandomized design (primary N = 780). The overall perceived-authenticity difference was not significant (0.010, 95% CI [−0.033, 0.052], p = 0.656). Perceived AI traces were higher in the treated group in the fixed-scene analysis (0.060, 95% CI [0.025, 0.096], BH-adjusted p = 0.007), but the interval crossed zero in a post hoc scene-level condition random-slope model (0.060, 95% CI [−0.098, 0.219]). Condition-by-scene interactions were significant for all eight constructs. Together, the studies show why the appraisal of AI-generated pasts should distinguish emotional reception from trust and examine particular proposed scenes rather than rely on a single average authenticity rating.