A within-subjects, prospective design was used to assess memory performance based on 3 within-subject variables: exposure condition (consciously cued, unconsciously cued, and control), emotional valence (positive and negative), and test time point (0 hour, 24 hours, and 72 hours). In this design, the participants acted as their own controls and were exposed to all levels of these variables. The study was conducted in 3 phases: Learning, Gaming, and Testing. In the Learning phase, participants learned 36 cue-target pairs using a self-feedback process. The target images had either positive or negative emotional valence. In the Gaming phase, participants engaged in a game designed to induce amnesic shadows, in which 18 of the previously learned cues were exposed to the participants either in a conscious or unconscious manner. Participants were tested on the previously learned cue-target pairs at three time points: (1) immediately after, (2) 24 hours after, and (3) 72 hours after the Gaming phase. The experimental procedure is outlined in Figure 1.

We followed the TREND (Transparent Reporting of Evaluations with Nonrandomized Designs) statement [20] to ensure comprehensive and transparent reporting of this within-subject experimental study. A completed TREND checklist is provided.

Recruitment of participants was done through convenience sampling via surveys targeting students at Yonsei University, Seoul, South Korea, from December 1, 2023, to March 1, 2024, which complied with ethical regulations for research on human participants. The surveys were posted on Yonsei University’s bulletin boards and online community.

Then, the final participants were selected according to the inclusion criteria: (1) undergraduate or graduate students aged 19‐29 years enrolled in Yonsei University, (2) corrected vision of 0.5 or better in both eyes, (3) no known health issues, and (4) Korean language proficiency. The study analyzed a final sample of 56 participants (Figure S1 in Multimedia Appendix 1). Initially, 85 potential participants completed an online prescreening survey. During the online prescreening survey, participants were asked whether they were currently undergoing any form of medical or psychological treatment, and none reported having an active condition. Nineteen participants were excluded on the basis of predetermined criteria (age, language, and vision). The remaining 66 individuals were enrolled after providing written informed consent. For the analysis, 10 participants were excluded because of technical issues, allocation errors, voluntary withdrawals, and incomplete data.

The final 56 participants had a mean age of 23.37 (SD 1.84) years, and 62.5% (35/56) of the participants were female. Their mean corrected vision was 0.99 (SD 0.21) (Table S1 in Multimedia Appendix 1). All participants met the inclusion criteria (aged 19‐29 years, fluent in Korean, currently enrolled in university, and having a corrected vision of 0.5 or better).

From the experiment, participants’ game scores, total feedback duration, EEG data, and verbal responses were collected and processed, as seen in Figure 1. Primarily, verbal responses were processed into the outcome variables: METEOR score, the Gist score, and the BERTScore. EEG data were also collected as the outcome variable. The exposures consisted of 3 within-subject factors: exposure condition (consciously cued, unconsciously cued, or control), emotional valence (positive or negative), and test time point (0 hour, 24 hours, or 72 hours). Two predictors were included as covariates to control for individual differences in engagement and learning effort: participants’ game score and total feedback duration. Several potential confounders were also identified, including stress induced by flickering stimuli, unassessed clinical or trauma history, variability in concentration and gaming experience, a limited number of items per cell in the 3×3×2 design, the absence of a neutral baseline condition, and potential assessment imprecision due to the small independent norming sample.

The outcomes of the 56 participants who completed the entire experiment were analyzed. Statistical analyses were performed using R (version 4.4.0; R Core Team). All data are available in the Open Science Framework (see the “Data Availability” section). The level of statistical significance was set at α=.05, and the CI at 95%.

Two factorial repeated-measures designs were used to examine the effects of the game-based ShIF. In the first analysis, exposure condition (conscious, unconscious, and control) and test time point (0 hour, 24 hours, and 72 hours) were included as 3-level within-subject factors. This analysis focused on the main effect of exposure condition (ie, conscious < control, unconscious < control), the main effect of time, and their interaction (ie, whether the effect of exposure condition varied across days). All post hoc pairwise comparisons were Bonferroni adjusted.

In the second analysis, emotional valence (positive and negative) was added as an additional within-subject factor, resulting in a 3×3×2 design. This model evaluated whether the short-term memory interference observed immediately after game exposure persisted across the follow-up sessions and whether this pattern differed depending on emotional valence. Post hoc pairwise comparisons were Bonferroni adjusted.

To confirm the robustness of these results, repeated-measures ANCOVA was conducted for each analysis using the same factorial structure, controlling for individual differences in task engagement. Participants’ game score and total feedback duration were included as covariates. Missingness occurred only in these covariates: game score was missing for 3 participants (3/56, 5.4%) and feedback duration for 3 participants (3/56, 5.4%), with 1 participant overlapping, yielding 5 out of 56 unique cases (8.9%). Missingness satisfied missing completely at random (χ²₂=2.87; P=.24). Given this covariate-only missingness and the missing completely at random assumption, we analyzed the ANCOVA as a complete-case dataset (n=51) using type III sums of squares with Greenhouse-Geisser corrections where necessary. Post hoc comparisons were Bonferroni adjusted.

To examine the statistical significance of the intervention, we tested whether there was a significant difference between the conscious and unconscious phases in negative images, as well as between the conscious and unconscious phases in positive images. Paired testing was conducted to eliminate individual differences and minimize variations arising from the game progression. Specifically, 40 participants each played the 2 types of games 9 and 18 times in total. All EEG data were segmented within a 2-second time window. Segmentation was performed after preprocessing to avoid filter artifacts at the edges. All studies were performed equally in the 2 groups. To compare the differences between the conscious and unconscious phases, the ith EEG from the conscious phase was paired with the corresponding ith unconscious phase EEG from the same individual in chronological order.

Similarly, to assess the difference between conscious and unconscious states, the EEG from the ith conscious phase was paired with the EEG from the ith unconscious phase for each participant. After data pairing, a 2-sided Wilcoxon signed rank test was performed for each pair. To control for the potential inflation of type I error caused by multiple pairwise comparisons across emotional conditions and EEG channels, a Bonferroni correction was applied.

This study was reviewed and approved by the institutional review board (IRB) of the Severance Hospital Human Research Protection Center (approval number: 4-2023-0638). All procedures involving human participants were conducted in accordance with the ethical standards of the Declaration of Helsinki and the institutional guidelines issued by the Yonsei University Health System, eIRB-provided document (Multimedia Appendix 2). Written informed consent was obtained from all participants prior to their participation. The consent form described the study purpose, the voluntary nature of participation, the right to withdraw at any time, and the use of anonymized data for research and publication. For the secondary analysis of behavioral and EEG data, the IRB approval explicitly permitted reuse of the deidentified data without requiring additional consent. All participant data were anonymized prior to analysis. Personally identifiable information such as names, email addresses, or voice recordings was removed or replaced with coded identifiers. Verbal responses were transcribed and stored using pseudonymous participant IDs, and all EEG recordings were delinked from any personal identifiers. Data were stored on encrypted institutional servers accessible only to the research team. Participants who took part in the main experiment received monetary compensation of 30,000 KRW (approximately US $20) for their time and effort. Those who participated in the supplementary awareness validation test received a proportional amount based on the duration of their participation. Compensation procedures complied with the institutional policy for human subject research. No identifiable images of individual participants are included in the manuscript or supplementary materials. All figures and photographs display either schematic representations or deidentified data. If any image contained potentially identifiable human features, explicit written consent for publication was obtained from the individual prior to inclusion. The study protocol detailing the full experimental procedures is provided in Multimedia Appendix 3, and the statistical analysis plan used in this study is described in Multimedia Appendix 4.

To validate the effect of game-based ShIF, we designed a game structured in 2 phases: an FP, in which the character’s movement corresponded to the arrow keys, and an RP, in which the character’s movement was inverted. For comparison, among the 36 cue-target pairs prelearned by 56 participants (21 men; mean 23.4 years, SD 1.84 years; Table S1 in Multimedia Appendix 1), half of the pairs were randomly assigned to the Conscious or Unconscious group and presented in the RP, while the remaining pairs were designated as the control group.

A two-way repeated-measures ANOVA was conducted with exposure condition (conscious, unconscious, and control) and time (0 hour, 24 hours, and 72 hours) as within-subject factors. A significant interaction between exposure condition and time was observed (F_3.62,199.33=2.7, P=.04, η²_ρ=0.05, 95% CI 0.00-1.00), suggesting that the effect of exposure condition varied across time points, although the main effect of exposure condition was not significant (F_1.89,103.74=0.58, P=.55, η²_ρ=0.01, 95% CI 0.00-1.00) (Figure 2A). Importantly, follow-up analyses revealed that only the immediate test (0 hour) exhibited that the METEOR scores varied significantly across the control, conscious, and unconscious conditions (F_1.7,93.59=4.874, P=.01, η²_g=0.056, 95% CI 0.00-1.00). Specifically, a ShIF effect was observed in the conscious condition, with METEOR scores significantly lower in the conscious group than in the control group (t₅₅=−2.86, P=.02, Cohen d=−0.382, 95% CI for d −0.658 to −0.110). In contrast, the unconscious group exhibited no significant difference from the control condition (t₅₅=−0.015, P≥.99, Cohen d=−0.002, 95% CI for d −0.266 to 0.262) and trending higher than the conscious group (t₅₅=−2.28, P=.08, Cohen d=−0.304, 95% CI for d −0.576 to −0.035). (Figure 2B). The analysis also revealed a strong main effect of time (F_1.98,108.93=84.73, P<.001, η²_p=0.61, 95% CI 0.51-1.00) indicating a progressive decline in METEOR scores over days.

To ensure the robustness of these findings, the same analyses were conducted while controlling for individual differences in task engagement, using participants’ game scores and total learning time as covariates. The two-way repeated-measures ANCOVA results yielded a comparable pattern, showing a significant interaction between exposure condition and time (F_3.54,169.71=2.64, P=.04, η²_ρ= 0.05, 95% CI 0.00-1.00) and main effect of time (F_1.97,94.42=83.2, P<.001, η²_ρ=0.63, 95% CI 0.54-1.00). Post hoc comparisons again revealed lower METEOR scores in the conscious condition than in the control condition (t₄₈=−2.975, P=.01, Cohen d=−0.429, 95% CI for d −0.723 to−0.132) in the immediate test (0 hour), confirming that the ShIF effect in the conscious condition remained significant after adjustment for individual differences in task engagement. The conscious condition also tended to score lower than the unconscious condition (t₄₈=−2.356, P=.07, Cohen d=−0.340, 95% CI for d −0.629 to −0.048), whereas no significant ShIF effect was observed for the unconscious condition when compared with control (t₄₈=0.037, P≥.99). Neither game score (F_1,48=0.18, P=.675, η²_ρ=0.003, 95% CI 0.00-1.00) nor total feedback duration (F_1,48=0.72, P=.98, η²_ρ<0.001, 95% CI 0.00-1.00) showed significant main effects, and none of the interactions involving these covariates were significant (Table S6 in Multimedia Appendix 1).

Applying the same analyses to gist- and BERT-based similarity measures yielded consistent and statistically significant patterns, replicating the results observed with the METEOR scores (Tables S5 and S6 and Figures S3 and S4 in Multimedia Appendix 1).

To further explore the role of emotion in the ShIF effect, we conducted a three-way repeated-measures ANOVA including emotional valence (positive and negative) as an additional within-subject factor, along with exposure condition and time (Figure 3). The analysis revealed significant main effects of emotional valence (F_1,55=42.43, P<.001, η²_ρ=0.44, 95% CI 0.27-1.00) and time (F_2,109.99=102.91, P<.001, η²_ρ=0.65, 95% CI 0.57-1.00), indicating that recall performance declined over days and was consistently lower for negative than for positive images. Post hoc comparisons for the emotion factor confirmed that negative images were recalled less accurately than positive images across all test sessions (positive valence vs negative valence: t₅₅=6.514; P<.001). Pairwise comparisons for the time factor showed a progressive decline in recall performance from 0 hour to 24 hours (t₅₅=7.70; P<.001) and from 24 hours to 72 hours (t₅₅=6.628; P<.001).

When emotional valence was incorporated into the model, the previously significant interaction between exposure condition and time was no longer statistically significant (F_3.63,199.84=1.34, P=.26, η²_ρ<0.001; see Table S7 in Multimedia Appendix 1 for details). Nevertheless, the inclusion of emotional valence provided additional insight into how the ShIF effect manifested across emotional contexts. At the immediate test (0 hour) following the game intervention, the pattern of results was generally consistent with the previous analysis, as shown by exploratory comparisons within each valence condition. For negative images, the conscious condition (mean 0.083, SD 0.574) showed lower mean METEOR scores than the control condition (mean 0.342, SD 0.57) (mean difference=−0.259, 95% CI −0.542 to 0.024), whereas for positive images, the conscious (mean 0.517, SD 0.9) and control (mean 0.602, SD 0.726) conditions were more comparable (mean difference=−0.085, 95% CI −0.452 to 0.282). Although formal post hoc comparisons were not conducted, this descriptive trend suggests that the short-term ShIF effect may have been more pronounced for negative stimuli under conscious exposure.

To account for potential variability in task engagement, we further included game score and total feedback duration as covariates in a repeated-measures ANCOVA to control for potential baseline differences. The adjusted model yielded highly comparable results with ANOVA, confirming that the main effects of emotion (F_1,48=43.94, P<.001, η²_ρ=0.44, 95% CI 0.27-1.00) and time (F_1,48=97.93, P<.001, η²_ρ=0.65, 95% CI 0.57-1.00) remained robust after controlling for these factors. Exposure conditions (F_1,84,88.5=0.1, P=.89, η²_ρ=0.004, 95% CI 0.00-1.00) and the interactions remained nonsignificant (Table S8 in Multimedia Appendix 1). Neither game score (F_1,48=0.04, P=.84, η²_ρ<0.001) nor total feedback duration (F_1,48=0.72, P=.40, η²_ρ=<0.001) exerted significant effects, supporting that the observed effects were independent of task performance or engagement.

Applying the same analyses to gist and BERT measures yielded a consistent pattern replicating the METEOR-based results (Tables S7-S11 in Multimedia Appendix 1). Full details are provided in Multimedia Appendix 1.

EEG measurements were conducted on the participants during the game, and power spectrum densities were compared. Details of the statistical methods and measurement techniques are provided in the “Methods” section. First, a comparison of EEG spectrum activation between the unconscious and conscious phases showed a relatively higher activation in the unconscious phase for EEG channels near the occipital region of the beta and theta spectra, likely due to stress from flickering stimuli [36,37]. A topographical plot of the power spectrum density difference between the conscious and unconscious phases is shown in Figure 4A. The channels with significant differences are marked with yellow circles. Results from the Wilcoxon signed rank test demonstrated higher activation patterns during the unconscious game, especially in channels near the occipital region. (O1: P=.02 and O2: P=.009 for theta spectrum, O1: P=.037 and O2: P=.02 for beta spectrum, and O1: P=.004 for gamma spectrum) (Table S12 in Multimedia Appendix 1).

Second, power spectrum analysis of EEG signals recorded in response to positive and negative images revealed that the brainwave spectrum bands exhibited significantly higher activation when viewing negative images than when viewing positive ones (Figure 4B). Notably, a significant difference was observed in the right frontal beta band power. This finding aligns with the previously observed tendency of memory attenuation [38] for negative images, further emphasizing the ShIF effect. Indeed, the results of the Wilcoxon signed rank test demonstrated highly significant activation patterns when revealed in negative images. For the beta waves, highly significant activation patterns were found in the right frontal channels (FC6: P<.001, F4: P<.001, and F8: P<.001). A significant increase in beta waves was also noted in the left frontal region (AF3: P=.03, F3: P=.002, F7: P<.001, FC5: P<.001, and T7: P<.001) as well as in the occipital and parietal regions (O1: P=.006, O2: P<.001, P8: P<.001, and P7: P=.009). For the gamma band, significantly increased activation was observed in the posterior region (O1: P<.001, O2: P<.001, and P8: P=.002) and in the centrofrontal area (F7: P<.001, C5: P<.001, F3: P=.009, F4: P<.001, FC6: P<.001, and F8: P=.03) (Table S13 in Multimedia Appendix 1).

In this study, we examined how exposure conditions and image valence influenced ShIF across 3 time points. Our work builds on the ShIF framework introduced by Zhu et al [5,18], which posits that unrelated episodic memories reactivated during the amnesic shadow, a period of reduced hippocampal activity, can be weakened. We successfully observed short-term ShIF for consciously cued images during a period of game-based amnesic shadows, but the effects of unconsciously cued images differed from those of a previous study [18], highlighting the need for further research on unconscious processing. Also, to our knowledge, this study is the first to show that ShIF occurs selectively for negative memories. Using EEG data, we provided the first evidence that inhibitory neural activity increases during the suppression of negative memories, directly supporting the mechanism underlying ShIF. Furthermore, by tracking the effects of ShIF over several days, we expanded ShIF to different temporal and emotional contexts.

Our first and second hypotheses, which state that ShIF effects will be significant in (1) both short- and long-term contexts and (2) in both conscious and unconscious exposure conditions, were not supported. In the short term (0 hour), we observed significant ShIF effects for consciously cued images, as evidenced by lower memory performance scores compared with control images, consistent with previous findings [5,18]. However, unlike a previous study [18], unconsciously cued images did not show significant ShIF effects. EEG analysis revealed that frontal beta power, a marker of inhibitory mechanisms linked to hippocampal downregulation [38-41], was not significantly different between conscious and unconscious conditions, suggesting that hippocampal activity may be similarly downregulated across conditions [42]. The lack of ShIF effects in the unconscious condition suggests that different interfering factors in the memory suppression process need further examination. Stress induced by the unconscious exposure condition in our game may explain this discrepancy. Specifically, the visual flickering used to deliver unconscious stimuli can act as a stressor [36,37], potentially impairing the effectiveness of memory suppression [43,44]. Unlike traditional paradigms, the flickering stimuli in our game were designed to be perceived unconsciously but may also function as visual distractors, elevating stress levels. Our EEG analysis shows increased occipital theta and beta band activity during the unconscious condition associated with stress responses to the visual stimuli [45-47], suggesting that the flickering stimuli, intended to deliver unconscious cues, may have acted as stress-inducing visual distractions, disrupting the ShIF mechanisms required for effective memory suppression. Future research should explore alternative unconscious exposure techniques to reduce stress while preserving the efficacy of the ShIF paradigm.

Over the long term, by 72 hours, the ShIF effect on consciously cued images diminished, with no significant differences observed at 72 hours. This finding is consistent with previous research showing that intentional forgetting effects tend to dissipate over time [48,49]. Although the ShIF effect for unconsciously cued images was not statistically significant, the trend observed at 24 hours suggests a potential delayed effect. This may tentatively align with the Negative Compatibility Effect theory [50] which proposes that inhibitory responses to subliminally activated memories can emerge over time. However, given the lack of statistical significance, a larger sample size is needed to verify this possible delayed effect.

Furthermore, we found that ShIF effects were significant only for images with negative valence. Negative memories are more likely to be intrusively reactivated and are more susceptible to suppression [11,12]. Our EEG results corroborated this finding, showing increased gamma band activity during negative cue presentation associated with processing unpleasant, high-arousal stimuli [51,52]. Significantly enhanced beta power, including in the frontal region, is linked to intentional forgetting and inhibitory processes [38-41]. This suggests that negative memories elicit stronger suppression processes, making them more susceptible to ShIFs.

In addition, memory performance scores were consistently and significantly lower for negative images at all time points (0 hour, 24 hours, and 72 hours). Higher arousal from negative stimuli may narrow attention, explaining their consistently lower recall and the lasting influence of valence on memory retention [53].

Our research has 3 main contributions to both the theoretical understanding and the practical applications of memory suppression mechanisms. First, by comparing the effects of ShIF under various conditions of awareness and emotional valence, we demonstrated that ShIF is more effective for consciously cued negative memories. Nevertheless, while our paradigm may have theoretical relevance for understanding intrusive memory processes in disorders such as posttraumatic stress disorder (PTSD), depression, and obsessive-compulsive disorder [54-59], cautious clinical application is needed. The emotional intensity and consolidation strength of real traumatic memories are far greater than those of experimentally induced ones [60], and inhibitory control may differ across clinical populations. Our findings primarily highlight a methodological and conceptual framework in a naturalistic, game-based setting, rather than providing direct clinical evidence.

Second, we introduced a novel game-based approach to induce amnesic shadows using reversed directional keys to trigger memory suppression. This method builds on the idea that suppressing motor responses engages inhibitory processes similar to those involved in memory suppression [12]. Our approach effectively induces the amnesic shadow without relying on the traditional TNT paradigm through the motor conflict in our game that imitates working memory challenges and motor sequence suppression in previous studies [15,61]. Moreover, this game-based method has several advantages. It reduces the experimenter bias inherent in the TNT paradigm [62] and may be more applicable in clinical situations where directly suppressing unwanted memories is challenging. The intuitiveness of games may be more engaging and less confrontational for individuals reluctant to participate in traditional therapeutic methods.

Finally, we used the METEOR score as a more accurate and objective evaluation metric for assessing memory performance [63,64]. Although METEOR is not a conventional memory metric, our validation analyses revealed strong agreement between METEOR and both human and embedding-based similarity scores, supporting its reliability as a semantic measure. The consistent—but sometimes more liberal—sensitivity of METEOR likely reflects its higher tolerance for semantic paraphrasing rather than inflated detection of differences.

Despite these contributions, our study has limitations. First, individual differences among participants may have influenced the results in multiple ways. Although we implemented a feedback cycle to ensure consistent memorization, variations in participants’ concentration, gaming experience, and skill levels may have affected their memory performance [65]. Likewise, no formal psychiatric or trauma-related screening (eg, PTSD, depression, or anxiety measures) was conducted beyond a basic prescreening question about current medical or psychological treatment. Unassessed differences in trauma history or emotional sensitivity could also have shaped responses to negative images and subsequent memory outcomes. Future studies should control for these individual-level factors by diversifying game tasks to sustain engagement and incorporating standardized mental health screening.

Second, methodological limitations arise from the statistical power and design of the study. The experiment involved 36 cue-target pairs distributed across a 3×3×2 within-subjects design, resulting in only 2‐3 items per cell. Although the total number of observations (n=56 participants × 36 items = 2016) appears large, the low item count per cell amplifies variability and limits sensitivity to small treatment effects, particularly for interactions involving emotional valence or unconsciously cued images. To examine this, we conducted a simulation-based post hoc power analysis using the Superpower R package (Comprehensive R Archive Network). Results indicated that the primary 3×3 design (Treatment × Time) had high power to detect medium-sized effects, particularly for day 1 contrasts where ShIF effects for consciously cued images were observed. In contrast, incorporating emotional valence in the full 3×3×2 design substantially reduced power for small treatment effects and interactions, which likely contributed to nonsignificant trends for unconsciously cued images at 24 hours. These findings highlight that while the short-term ShIF effects for consciously cued images are robust, caution is warranted in interpreting nonsignificant effects in more complex combinations, and future studies with larger samples and more items per cell are needed to fully investigate potential delayed ShIF effects under unconscious cueing.

Third, our analysis primarily focused on the weakening of cue-target associations rather than directly measuring the intrinsic strength of the memories. Although disrupting these associations is clinically relevant for reducing involuntary recall of traumatic memories [55], future research should consider the recall of the memory using independent cues to determine whether the memories themselves are weakened. In addition, recall accuracy in this study was assessed using a unified description derived from a small independent sample (n=12) rather than participant-specific references. While this approach ensured consistency across items, it likely reflected the “typical” recallable content of each image rather than each participant’s unique reconstruction. We intentionally avoided collecting participant-specific descriptions before the main task, as verbalizing image content could have influenced later recall performance. Future studies should incorporate individualized reference descriptions in separate pre- or postsessions and better isolate genuine memory strength and recall accuracy.

Furthermore, the usage of image stimuli from multiple sources and the limited norming procedure may be considered. Although this approach allowed us to select images that met specific criteria—namely, containing a context-irrelevant object suitable for use as a cue—it also introduced variability in stimulus characteristics and restricted the generalizability of the emotional ratings due to the small norming sample. Several prior studies have similarly relied on stimuli from multiple databases rather than a single standardized source [22,23], suggesting the feasibility of this approach. This approach also enabled selecting images that met specific criteria—namely, containing a context-irrelevant object suitable for use as a cue—but led to variability in stimulus characteristics and restricted the generalizability of the emotional ratings due to the small norming sample. In addition, the absence of a neutral baseline condition limits the ability to determine whether the observed effects are truly valence-specific or reflect general emotional processing. This decision was made to increase sensitivity to positive-negative contrasts and to avoid potential ambiguity associated with “neutral” stimuli, which are known to vary widely across individuals and contexts [66,67]. Future research should address these issues by recruiting a larger and more diverse group of raters, developing a unified standardized database that fulfills both emotional and contextual control requirements, and incorporating neutral stimuli to establish a clearer reference point for valence-dependent effects. Future research should recruit a larger and more diverse group of raters, developing a unified standardized database that fulfills both emotional and contextual control requirements, and incorporate neutral stimuli to establish a clearer reference point for valence-dependent effects.

Finally, our study focused on the effects of a single, 8-second game-induced amnesic shadow on recently learned memories. This limited intervention may not fully capture the potential of ShIF. Intentional forgetting may have long-term effects on consolidated memories [68], and multiple attempts at suppression can amplify the extent of forgetting [8,69-73]. Future studies should explore whether repeated game-based ShIF sessions over extended periods have greater memory suppression effects, which may enhance the clinical applicability of ShIF in long-term memory management.

Our study demonstrates that ShIF is modulated by the awareness and valence of memories and that ShIF can be efficiently induced in the short term by a game-based approach, especially for negative and consciously cued images. EEG analyses provided critical validation, revealing increased inhibitory neural activity during the suppression of negative memories and directly supporting the neural mechanisms underlying ShIF. The delayed ShIF effect observed for unconsciously cued images highlights the complex dynamics of memory suppression over time. These findings contribute to a deeper understanding of the memory suppression mechanisms of ShIF and open new avenues for its practical application in clinical settings.

Importantly, this study confirms that suppression effects can be obtained through a game-based paradigm, suggesting a more engaging and ecologically valid alternative to traditional explicit suppression tasks. Moreover, unlike previous studies that have primarily focused on transient effects limited to consciously processed negative stimuli, our investigation examined emotional valence, exposure condition, and temporal dynamics. We thereby extend the theoretical and practical scope of ShIF. Taken together, these results provide preliminary evidence supporting the potential translation of ShIF mechanisms into digital therapeutic applications, such as game-based interventions for PTSD and related conditions.