Body Weight & Emotion Expressions — AI-Generated Image Datasets
Both datasets include high-quality images of individuals from the chest up, generated using ComfyUI with the Flux.1-dev model. All images were validated with a minimum 95% likelihood threshold using Py-Feat facial action unit analysis.
Version 1 — 72 Images
Body types: fit, middleweight, obese. Emotions: happiness, sadness, neutrality. Includes 4 male and 4 female identities, each expressing all emotions. Includes full Py-Feat AU analysis.
Version 2 — 96 Images
Body types: fit, middleweight, obese. Emotions: happiness, sadness, anger, neutrality. Includes 4 male and 4 female identities, each expressing all four emotions. Includes full Py-Feat AU analysis.
The Paper
The creation and validation of these datasets are described in the following article (Open access):
Hareli, S., & David, S. (2026). Creating and validating photorealistic AI-generated facial expression stimuli for emotion research. Behavior Research Methods, 58, Article 296. https://doi.org/10.3758/s13428-026-03156-0
Full supplementary materials — image datasets, study data, codebooks, analysis and validation scripts, LoRA models, and complete workflow specifications — are available on OSF.
Terms of Use
The images are available free of charge for non-commercial research purposes. Commercial use is not permitted.
Required citation: any publication, presentation, or other work that uses the images must cite:
Hareli, S., & David, S. (2026). Creating and validating photorealistic AI-generated facial expression stimuli for emotion research. Behavior Research Methods, 58, Article 296. https://doi.org/10.3758/s13428-026-03156-0
Research Applications
- How body weight influences perception of emotions and personality trait attribution
- Cross-cultural emotion perception studies
- Benchmarking automated emotion recognition algorithms
- Training machine learning models for emotion detection in naturalistic stimuli