The research team, led by Katie Hoemann of the University of Kansas and including Jeffrey Girard, Marissa Ogren, Lauren Leotti, Lisa Feldman Barrett, Lisa Oakes, and Vanessa LoBue, took advantage of an existing longitudinal corpus of home videos known as the SEEDLingS data set, shared by Elika Bergelson’s research group. Rather than staging laboratory interactions, the study examined naturalistic footage of infants going about ordinary life with the people closest to them. This ecological approach matters because laboratory studies of emotion perception often present infants with posed, prototypical faces, yet those faces may bear little resemblance to what babies actually encounter in daily life.

The technical core of the study involved coding the facial behavior visible in the recordings using facial movement analysis, a method that quantifies the visible contractions of facial muscles rather than relying on subjective labels of what an expression means. The team asked two distinct questions: how much of the face were infants actually exposed to, and how often did the visible configurations match the hypothesized prototypes for emotion categories such as happiness, anger, sadness, and fear. This dual approach allowed the researchers to characterize both the quantity and the emotional quality of infants’ facial input.

The first major finding concerned visibility. Infants were exposed to far more partial faces than full faces across the entire age range studied. In the flow of everyday interaction, caregivers turn away, lean in close, move out of frame, or orient their faces at oblique angles, meaning that a baby’s view of another person’s face is typically fragmentary rather than complete. Exposure to full faces, while low overall, did increase across the observed time periods, hinting that infants’ visual access to complete facial configurations gradually improves as they and their caregivers develop more structured patterns of face-to-face engagement.

The second finding was more striking. Facial configurations hypothesized to serve as expressive prototypes for emotion categories were extremely rare in the input. When babies did glimpse a full face, it almost never displayed the canonical smile, scowl, or grimace that emotion recognition tasks assume they should be learning. This scarcity strikes at a central assumption of classical theories of emotional development, which posit that infants acquire emotional meanings by repeatedly pairing specific facial configurations with specific internal states. If the configurations themselves are seldom visible, that learning mechanism cannot operate as traditionally imagined.

Equally revealing was what happened on the rare occasions when hypothesized prototypes did appear. They often occurred in unexpected contexts, suggesting that the expressions did not necessarily function as cues to their presumed emotional meanings. A smile might accompany frustration rather than joy; a furrowed brow might signal concentration rather than anger. This contextual mismatch reinforces a growing body of evidence that emotions are expressed heterogeneously in daily life, and that the same facial movement can serve many different functions depending on the situation, the person, and the ongoing activity.

Descriptive analyses of the facial movement data added a quantitative layer to the story. Infants’ facial input was dominated by a small number of mostly simple facial configurations, and these frequent configurations did not correspond fully or uniquely to the hypothesized category prototypes. In other words, the statistical structure of the visual environment does not carve itself neatly along the lines of happiness, sadness, anger, and fear. Instead, babies see a recurring set of modest facial movements, many of them partial and many of them ambiguous with respect to any single emotional category.

The implications ripple outward across developmental science. If infants must learn much more than one-to-one mappings between facial configurations and context or behavior in order to acquire expressive vocabularies for emotion, then the learning problem they face is far more complex than previously assumed. Rather than memorizing a small dictionary of expression-to-meaning pairs, babies may need to track how facial movements unfold dynamically, how they combine with voices, gestures, and physical settings, and how particular individuals in their environment use faces in idiosyncratic ways. Emotion understanding, on this view, is constructed from rich, variable, and context-dependent evidence rather than absorbed from a fixed set of signals.

The study also carries methodological weight for the field. Decades of infant emotion perception research have relied on posed, prototypical stimuli precisely because they are easy to control and categorize. The new findings raise the question of whether performance in such tasks reflects learning about real emotional communication or merely sensitivity to artificial faces that infants rarely, if ever, see outside the laboratory. Naturalistic corpora like the one used here, combined with automated facial movement coding, offer a path toward reconciling laboratory findings with the realities of the developing child’s environment.

For parents, the takeaway is quietly reassuring rather than alarming. Babies appear to thrive without being fed a steady diet of textbook expressions, and the messy, partial, and context-bound faces of everyday life seem to be exactly the input from which human emotional understanding is built. The research, funded in part by the National Science Foundation and the Eunice Kennedy Shriver National Institute of Child Health and Human Development, marks a significant step toward grounding theories of emotional development in the world as infants actually experience it, one fleeting and fragmentary glimpse of a caregiver’s face at a time.

Subject of Research: Naturalistic analysis of infants' everyday exposure to facial expressions of emotion from 6 to 17 months

Article Title: Infants’ facial input from 6 to 17 months rarely includes hypothesized expressive emotion prototypes

Article References: Hoemann, K., Girard, J. M., Ogren, M., Leotti, L., Barrett, L. F., Oakes, L. M., & LoBue, V. (2026). Infants’ facial input from 6 to 17 months rarely includes hypothesized expressive emotion prototypes. Communications Psychology. https://doi.org/10.1038/s44271-026-00539-z

Image Credits: AI Generated

DOI: 10.1038/s44271-026-00539-z

Keywords: infant development, emotion perception, facial expressions, facial action coding, longitudinal study, naturalistic observation, caregiver interaction, Communications Psychology, emotional development, SEEDLingS corpus, developmental psychology, social cognition

The research team set out to close a persistent gap between laboratory science and everyday experience. In the real world, faces never float in a void; they appear against beaches, streets, living rooms, and hospital corridors, all of which carry their own emotional charge. Previous work had hinted that context matters for judging a single face, and studies of individual emotion recognition have long shown that surrounding information can disambiguate ambiguous expressions. But ensemble perception, the summarizing of many faces at once, had rarely been tested under naturalistic conditions. The Korean team therefore built crowd scenes that preserved the richness of real social environments while still allowing precise experimental control. Their central question was deceptively simple: when you judge the overall emotion of a group, how much of that judgment comes from the faces, how much from the scenery behind them, and how much from the state of mind you bring to the task?

Experiment 1 tackled the external context. The researchers constructed crowd scenes by superimposing images of human figures onto emotionally positive, negative, or neutral backgrounds. Crucially, they did not treat facial emotion as a crude categorical variable such as happy versus angry. Instead, the facial emotional value of each figure was quantified on a continuous scale derived from naturalistic photographs, meaning that every crowd had a computable average emotional tone that could be compared against participants’ judgments. This continuous quantification is what gives the study its technical teeth: it allowed the team to model, trial by trial, how strongly the actual facial statistics of a scene predicted what observers reported, and whether the background valence shifted those reports above and beyond the faces themselves.

The analysis relied on Bayesian model comparison, a statistical framework that weighs competing explanations by how well each predicts the observed data. Rather than asking whether a single model is significant, the researchers compared models that included facial emotion only, background valence only, or both together, using the evidence in the data to decide which structure best described participants’ judgments. The verdict was clear: both sources of information contributed. Facial emotion strongly anchored the mean ratings, as expected, but background valence exerted its own independent pull. Positive backgrounds biased crowd judgments in a more positive direction, while negative backgrounds dragged the same crowds toward negativity. In other words, an identical set of faces could be read as a happier or gloomier collective depending entirely on the scenery framing it.

This contextual bias is more than a laboratory curiosity; it speaks to how the brain constructs social reality. Ensemble perception is often celebrated as a form of statistical summary, a mechanism that averages over many items to escape the limits of attention and working memory. The new results show that this averaging process is not sealed off from the rest of perception. Scene gist, the rapid global impression of an environment’s meaning and mood, appears to be folded into the summary computation itself. From an adaptive standpoint this makes sense: a crowd’s threat value depends on where the crowd is gathered, and a celebration in a park signals something different from the same expressions in a crowded subway car. The visual system seems to exploit background valence as a prior, nudging ambiguous collective signals toward the interpretation most consistent with the setting.

Experiment 2 turned the lens inward, asking whether the observer’s own emotional state joins this integration. The team induced positive or negative affect in participants using an autobiographical recall task, a standard mood manipulation in which people vividly remember personal events that made them feel happy or sad. Participants then judged the overall emotion of crowd scenes, and the researchers extended their Bayesian modeling framework to include observer mood as an additional factor. The results revealed a subtler pattern than the background effect. Mood did not produce a reliable shift in the average ratings: being in a good mood did not simply make every crowd look happier, and a bad mood did not uniformly darken judgments.

Instead, mood left its fingerprint on a different statistical property: response variability. Participants in a positive mood showed reduced variability across their judgments, indicating more consistent ensemble perception, whereas those in a negative mood produced more scattered ratings. This dissociation between mean and variance is one of the study’s most intriguing contributions. It suggests that internal affect does not act as a simple additive bias on the summary signal but instead modulates the precision or stability of the perceptual process. One plausible interpretation, consistent with broader theories of positive emotion, is that positive affect broadens the scope of attention and stabilizes the extraction of global statistical summaries, while negative affect narrows or disrupts the sampling of the crowd, injecting noise into the final judgment.

The study also carries methodological lessons for the field. The authors note that none of the experiments were preregistered, and they have made trial-level data and analysis code openly available on the Open Science Framework, allowing other researchers to scrutinize and extend the modeling approach. The use of continuous facial emotion values, naturalistic photographic backgrounds, and Bayesian model comparison collectively offers a template for studying ensemble perception under conditions that better approximate the visual world. The work was supported by a National Research Foundation of Korea grant, and the experiments were approved by the Institutional Review Board of Yonsei University, with all participants providing written informed consent.

The broader implications stretch into clinical and applied territory. Prior research has linked ensemble emotion coding to social anxiety, with anxious individuals showing distinctive patterns when summarizing emotional crowds, and cross-cultural work has shown that observers from different societies weight attended faces differently when judging overall mood. If both scene context and observer mood systematically shape these judgments, then measures of ensemble perception may need to account for the testing environment and the participant’s affective state, factors that most current paradigms ignore. For everyday social life, the message is equally striking: the mood we attribute to a room is a construction, assembled from faces, scenery, and our own inner weather. The next time you sense that a crowd is hostile or jubilant, it may be worth asking how much of that impression belongs to the people, and how much to everything, and everyone, else in the frame.

Subject of Research: How scene context and observer mood shape ensemble emotion perception from facial crowds

Article Title: Beyond faces: The effect of context on ensemble emotion perception

Article References: Lee, W., Yoh, H., & Chong, S. C. (2026). Beyond faces: The effect of context on ensemble emotion perception. Attention, Perception, & Psychophysics, 88(7), Article 194. https://doi.org/10.3758/s13414-026-03335-3

Image Credits: AI Generated

DOI: 10.3758/s13414-026-03335-3

Keywords: ensemble perception, emotion perception, face perception, scene context, observer mood, Bayesian modeling, social perception, visual perception, affect, crowd scenes, psychophysics, response variability

The research, published in Behavior Research Methods, describes both a method for producing photorealistic AI-generated emotional expressions and a comprehensive validation framework for testing whether such images can stand in for photographs in behavioral experiments. The work, led by Shlomo Hareli and Shlomo David, addresses a long-standing tension in emotion research between experimental control and ecological validity. Traditional photographic datasets are constrained by the actors available, the expressions they can convincingly produce, and the ethical complications of using images of real people. Computational alternatives such as 3D morphable models and FACS-based animation tools offer more control but require expensive specialized software, months of technical training, and often produce faces that fall squarely into the uncanny valley.

Generative AI, the researchers argue, relaxes this trade-off. The method is built on the open-source platform ComfyUI and the Flux.1-dev diffusion model, enhanced with Low-Rank Adaptation, or LoRA, fine-tuning. The authors trained custom LoRA models for each combination of emotion and gender using 15 images per combination drawn, with permission, from the validated FACES database of real facial expressions. Each model was trained for 16 epochs on a single consumer-grade laptop GPU, requiring roughly six hours per emotion-gender pair, with a fixed random seed to guarantee that other laboratories can reproduce the results exactly. Prompt engineering then controlled body weight, clothing, lighting, background, and gaze direction, with weighted attention values emphasizing specific features during generation.

Crucially, the pipeline does not stop at generation. Every image was screened by Py-Feat, a Python library for facial expression analysis, using a strict criterion that the intended emotion had to reach at least a 95 percent classification likelihood before the image was accepted. Images that failed were iteratively refined with a node-based expression editor until they passed. The final Study 1 dataset contained 72 images, evenly balanced across gender, three body weight categories, and three emotional expressions, with a mean intended-emotion confidence of 98.38 percent. All workflows, models, datasets, and analysis scripts were released openly on the Open Science Framework, a move the authors say is central to making the method genuinely reproducible.

The first human validation study recruited 606 US participants through Prolific, each viewing a single image. The results confirmed the manipulations across the board: happy faces were rated significantly happier than neutral ones, which were rated happier than sad ones, and perceived weight rose in clear steps from the thin through the average to the higher-weight categories. Perhaps most strikingly, when asked whether the person in the photo was real or AI-generated, on a scale where 3 meant they could not decide, no condition was clearly identified as artificial. Even the least convincing cell, higher-weight male faces, scored only around 3.25, barely past the uncertainty midpoint.

The age judgments revealed an unexpected form of validation. Happy faces were perceived as younger than sad or neutral ones, and perceived age increased systematically with body weight, patterns that replicate well-documented findings from research on real human photographs. Rather than revealing artificial confounds in the stimuli, these effects suggest the AI generation process captured realistic correlations between facial morphology, body weight, and perceived age, enhancing rather than undermining the ecological validity of the images.

Study 2 extended the framework in two important ways. First, it added anger as a fourth emotion. Second, it adopted a same-character design, in which a single AI-generated identity appears across all four emotional expressions, eliminating potential confounds from differing facial features between conditions. This posed a technical challenge: maintaining identity while altering expression. The team solved it with a cyclic workflow combining an expression editor, LoRA-based inpainting with the identity-preserving PuLID technique, and a face-swapping node, validated computationally with FaceShape software that confirmed 100 percent within-character similarity, benchmarked against the ADFES database. A total of 96 images depicting 24 distinct characters passed both identity and emotion checks, and 736 new participants confirmed that all four emotions and all weight categories were perceived as intended.

A third study, Study 2b, pushed validation beyond simple emotion recognition. With 731 additional participants rating the same 96 images, the researchers examined valence, intensity, naturalness, authenticity, and non-focal emotions. The stimuli behaved like real-expression photographs on every dimension: happiness was rated positive, sadness and anger negative, and neutral faces affectively neutral yet not expressively empty, with measurable intensity. The ordering of naturalness and authenticity ratings, happiness and neutrality highest, anger lowest, mirrored the genuineness hierarchy previously established for posed expressions in real databases. Male faces were rated as angrier than female ones, consistent with known gender stereotypes in emotion perception, and body weight influenced sadness and fear ratings, showing that the stimuli carry meaningful social-category information rather than functioning as abstract emotion displays.

The authors are careful about the limits of their achievement. Photorealism as judged by human observers does not mean AI-generated and real faces are equivalent in every respect; computational algorithms can still detect subtle artifacts invisible to people, and the current framework addresses only static images, whereas dynamic expressions are known to aid emotion recognition. The pipeline also still demands considerable technical skill, and the authors call for future template interfaces to make it accessible without workflow-level expertise. Nevertheless, they conclude that AI-generated stimuli can now serve as a scientifically validated, ethically attractive alternative to photographs, particularly for research on stigmatized characteristics such as body weight, and that the dual computational and human validation framework they established provides a template any laboratory can adapt as generative technology continues to evolve.

Subject of Research: Creation and validation of photorealistic AI-generated facial expression stimuli for emotion perception research

Article Title: Creating and validating photorealistic AI-generated facial expression stimuli for emotion research

Article References: Hareli, S., & David, S. (2026). Creating and validating photorealistic AI-generated facial expression stimuli for emotion research. Behavior Research Methods, 58(11), Article 296. https://doi.org/10.3758/s13428-026-03156-0

Image Credits: AI Generated

DOI: 10.3758/s13428-026-03156-0

Keywords: artificial intelligence, facial expressions, emotion perception, stimulus validation, body weight, generative AI, LoRA, ComfyUI, Py-Feat, behavioral research, photorealism, psychology methods