I Feel We Are Together: How People Perceive Personalized Face-Swapped GIFs in Text-Based Communication

Paper Title

I Feel We Are Together: How People Perceive Personalized Face-Swapped GIFs in Text-Based Communication

Publication Info

• Topic area: AI-mediated self-presentation in text-based communication
• Keywords: Face-swapped GIFs, generative AI, computer-mediated communication, self-presentation, social presence, intimacy, nonverbal cues, identity cues, emotional expression, privacy concerns

Background and Problem

• Problem / challenge: Text-based computer-mediated communication (CMC) lacks rich nonverbal cues like facial expressions, which are critical for emotional and social interactions. Existing visual cues (e.g., emojis, generic GIFs) are symbolic and fail to reflect users’ real identities. While generative AI enables personalized self-representation, its socio-emotional effects in text-based CMC remain underexplored.
• Significance: Enhancing self-expression and relational intimacy in text-based CMC is crucial for improving communication quality and emotional connection, especially in asynchronous settings.
• Motivation and related work: Prior work has shown that nonverbal cues like emojis and GIFs enhance social presence and emotional empathy. However, these cues are abstract and lack embodied identity. Synchronous media (e.g., video calls) provide richer identity cues but impose a high psychological burden. This study addresses the gap by introducing and evaluating face-swapped GIFs (FSGIFs) as a novel, AI-mediated embodied identity cue for asynchronous text-based CMC.

Solution

• Proposed approach: Face-swapped GIFs (FSGIFs), created by replacing the face in a GIF with the user’s face using generative AI, to enhance self-representation and relational intimacy in text-based CMC.
• Novelty:
  1. Empirical validation of FSGIFs as socio-emotional tools for text-based CMC.
  2. Exploration of AI-mediated self-representation through FSGIFs in dyadic interactions.
  3. Analysis of FSGIFs’ socio-emotional effects using social interaction theories.
  4. Design implications for personalized visual cues reflecting users’ facial identities.
• Procedure and key techniques:
  1. Conducted a two-phase within-subjects experiment with 32 participants (16 dyads of close acquaintances).
  2. Phase 1: Measured relational effects (e.g., intimacy, co-presence) during a collaborative travel-planning task using FSGIFs and generic GIFs.
  3. Phase 2: Evaluated emotional effects (e.g., valence, arousal) of FSGIFs versus generic GIFs in a controlled setting.
  4. Conducted semi-structured interviews to explore user perceptions, strategies, and privacy concerns.

Results

• Concrete findings:
  • FSGIFs significantly increased co-presence (M = 5.56 vs. M = 4.66, p <.001), intimacy (M = 5.91 vs. M = 5.25, p <.01), and enjoyment (M = 6.59 vs. M = 6.03, p <.05) compared to generic GIFs.
  • FSGIFs enabled greater selective self-presentation (M = 5.84 vs. M = 4.72, p <.001) without increasing psychological burden.
  • Negative emotions were perceived as less intense and expressive in FSGIFs than in generic GIFs (p <.001).
• Advantage over baselines:
  • FSGIFs bridged the gap between symbolic cues (e.g., generic GIFs) and synchronous media by offering high-fidelity identity representation with low effort.
  • Enhanced relational benefits without the performance burden of video-based cues.
• Experiments / evaluation:
  • Phase 1: Naturalistic conversation task with counterbalanced conditions (generic GIFs vs. FSGIFs).
  • Phase 2: Controlled evaluation of emotional perception using Russell’s circumplex model of affect.
  • Metrics: Social presence, self-disclosure, self-presentation, valence, arousal, expressiveness.
• Limitations and future work:
  • Limited to close-friend dyads; findings may not generalize to other relationships or group settings.
  • Novelty effect may influence initial positive reception of FSGIFs.
  • Synthesis quality issues (e.g., mismatched facial geometry) need improvement.
  • Future work should explore long-term use, diverse relational contexts, and other embodied identity cues (e.g., gestures, voice).

Summary

This study introduces face-swapped GIFs (FSGIFs) as a novel AI-mediated embodied identity cue for text-based communication. FSGIFs enhance relational intimacy, co-presence, and self-expression while maintaining the low effort and controllability of asynchronous CMC. Quantitative and qualitative analyses reveal that FSGIFs are particularly effective in close relationships, though users prefer them for positive expressions and avoid negative contexts. The findings highlight the potential of FSGIFs to balance personalization with privacy and contextual appropriateness, offering new design opportunities for communication technologies. Future research should address synthesis quality, explore broader relational settings, and investigate long-term adoption.