“I’m happy even though it’s not real”: GenAI Photo Editing as a Remembering Experience

Paper Title

“I'm happy even though it's not real”: GenAI Photo Editing as a Remembering Experience

Publication Info

• Topic area: The impact of Generative AI (GenAI) photo editing on personal remembering experiences.
• Keywords: Generative AI, photo editing, memory, remembering experience, human-computer interaction, identity, cultural bias, memory reconstruction, design implications, responsible AI.

Background and Problem

• Problem / challenge: Existing research focuses on misinformation and false memories caused by GenAI-edited media but lacks empirical evidence on how people use GenAI to edit personal photos and how this affects their remembering experiences.
• Significance: Understanding these practices is crucial for designing responsible GenAI tools that support personal meaning-making while preserving trust in personal archives.
• Motivation and related work: Prior work has framed remembering as a reconstructive and situated activity, but photo editing has traditionally been viewed as corrective or aesthetic. GenAI's capabilities for semantic editing shift this dynamic, enabling users to align photos with subjective memory rather than factual accuracy.

Solution

• Proposed approach: A two-phase qualitative study using GenAI photo editing guided by the Remembering Experience (RX) framework, which includes four dimensions: State of Mind, Social Context, Physical Context, and Memory Content.
• Novelty:
  1. Introduction of "felt memory" as a lens to understand GenAI-mediated remembering beyond factual accuracy.
  2. Identification of a "hierarchy of editability" that distinguishes between identity-critical and context-oriented photo elements.
  3. Empirical insights into how GenAI editing becomes part of the remembering process, enriching or distorting memory.
  4. Design implications for responsible GenAI that balance expressive remembering with risks of distortion and loss.
• Procedure and key techniques:
  1. Participants edited personal photos from 1, 5, and 10 years ago using a GenAI tool, guided by RX dimensions.
  2. Semi-structured interviews explored participants' intentions, perceptions, and experiences with the edits.
  3. Data were analyzed using reflexive thematic analysis, content analysis, and descriptive statistics.

Results

• Concrete findings:
  • Participants generated 133 edit requests, with 12% resulting in unsuccessful edits.
  • Most edits focused on modifying existing content (46.6%), followed by additions (24.8%) and removals (21.8%).
  • Environmental edits were rated highest for realness (M=4.16) and emotional connection (M=3.75), while human face edits were rated lowest for realness (M=2.80).
  • Older photos (10 years) attracted more generative edits, such as additions and extensions.
• Advantage over baselines:
  • GenAI enabled participants to align photos with subjective memory, enhance emotional connection, and creatively reinterpret memories, surpassing traditional photo editing tools' capabilities.
• Experiments / evaluation:
  • Conducted with 12 participants (balanced by gender and GenAI familiarity) in a controlled setting.
  • Participants rated edited photos on realness and emotional connection and reflected on their experiences in interviews.
• Limitations and future work:
  • Small, culturally homogeneous sample limits generalizability.
  • Study conducted in a controlled setting rather than natural contexts.
  • Future research should involve larger, more diverse samples, longitudinal studies, and comparative analyses of different GenAI tools.

Summary

This study explores how GenAI photo editing reshapes the remembering experience by prioritizing subjective "felt memory" over factual accuracy. Participants used GenAI to align photos with personal memories, treating environmental elements as flexible while preserving identity-critical features like faces. Editing itself became a form of active reminiscence, enriching memory but also introducing risks like information loss and cultural bias. The findings suggest design implications for responsible GenAI, emphasizing identity protection, cue preservation, and mindful memory work. These insights contribute to understanding how GenAI can support expressive remembering while mitigating potential distortions.