Research
Vela is an independent behavioral aesthetics research platform studying how viewers form aesthetic judgments about figurative art — and how artistic representation of the human body has varied across cultures and centuries.
The Vela platform presents museum collection works — drawn from open-access institutional APIs — to participants in editorially curated sequences. Behavioral signals (dwell time, resonance ratings, preference patterns) are recorded and used both to personalize the experience and to generate research data on aesthetic response. All works are public domain (CC0 or equivalent). Full attribution — institution name, accession number, source URL — is displayed alongside every work.
Behavioral aesthetics and viewer response
How does extended engagement with figurative art shape aesthetic preference?
The central study of the Vela platform. We present museum collection works to participants in editorially sequenced sessions and record behavioral signals — dwell time, resonance ratings (1–5), save behavior, and boundary signals — to study how viewers form aesthetic judgments over time. Unlike survey-based aesthetic research, our signals emerge from natural viewing behavior: what a participant chooses to linger on, return to, or pass over reveals preference structures that self-report cannot access.
Do aesthetic preferences stabilize, drift, or evolve across repeated sessions?
Which visual features predict high resonance ratings across heterogeneous viewer populations?
How does the sequential context of an image affect its perceived quality?
Does exposure to unfamiliar cultural traditions shift subsequent preference patterns?
A computational census of depicted human representation
How has the human body been represented across five centuries of institutionalized art?
A systematic census of figurative artworks drawn from ten or more major open-access museum collections, spanning the period from antiquity through the twentieth century. Using computer vision analysis, we catalog the characteristics of depicted figures — skin tone, gender presentation, body type, gaze direction, and pose orientation — and examine how these distributions vary across historical periods, cultural traditions, and collecting institutions. This study addresses a gap in the existing literature: prior empirical work (Topaz et al., 2019, PLOS One) quantified the demographics of artists in museum collections. We study the depicted subjects — who got painted, in what register, with what gaze.
How does the distribution of depicted skin tone change across historical periods from antiquity to the twentieth century?
Does gender presentation correlate with passive pose and averted gaze in ways that shift over time? (Empirical test of the Berger/Mulvey framework)
Do museums with similar collection missions show different patterns of depicted representation?
When Western artists depicted non-Western subjects, were those subjects more likely to be feminized, passive, and objectified than in non-Western artistic traditions? (Quantitative test of Said's Orientalism thesis)
Adaptive authorship and reader response
Does the register and framing of an argument affect how readers receive and retain it?
An experimental study of written communication using Vela's magazine platform as the experimental environment. We present editorially identical arguments — the same claim, evidence, and counterarguments — in systematically varied registers (accessible, literary, academic), entry points (which evidence leads), and reference sets (which cultural examples are drawn on). Reader response is measured through behavioral signals: scroll depth, time on page, whether readers mark themselves as having recognized, learned from, or wondered about the piece. We examine whether these signals vary across reader profiles established through prior visual engagement.
Does register variation (accessible vs. literary vs. academic) produce measurable differences in engagement depth and reader-reported response?
Do readers whose visual aesthetic profile favors contemplative imagery engage differently with contemplative literary registers?
Is there a cross-modal coherence between visual preference and textual register preference?
Does the cultural tradition of examples drawn on (Western vs. non-Western reference sets) affect engagement among readers with different visual engagement histories?
Desire modeling and aesthetic preference formation
How do aesthetic preferences form, stabilize, and change over time?
A longitudinal study of aesthetic preference formation using behavioral signals collected across extended platform engagement. Distinct from one-time preference elicitation studies, this work tracks preference across time — examining whether preferences stabilize after sufficient exposure, drift in response to new content, or exhibit systematic patterns (such as convergence toward specific dimensional profiles or periodic exploration of unfamiliar territory). The Vela platform's adaptive sequencing engine generates testable hypotheses about individual preference that are then confirmed or disconfirmed by subsequent ratings.
Do aesthetic preferences stabilize, or do they exhibit ongoing drift even in experienced viewers?
Is there a 'cold start' period for aesthetic preference formation, and what exposure volume is required to produce stable profile estimates?
Do preference profiles generalize across media (visual art to written art), or are they medium-specific?
Can behavioral signals predict which unfamiliar works a viewer will find resonant before that viewer has seen them?
We do not collect self-reported demographic data from participants. Behavioral inference is made from action, not identity. Cultural coordinates are inferred from engagement patterns, not from who we think viewers are.
The computational census analyzes what artists depicted — a visual content analysis of historical artworks — not claims about the identity of depicted subjects. This follows standard art historical cataloging practice.
Every work displayed on the platform carries full attribution: institution name, accession number, artist, date, and a direct link to the institution's collection page. We do not modify, redistribute commercially, or claim ownership of any institutional images.
Research datasets from the computational census will be deposited under CC0 on Zenodo upon publication, enabling replication and extension by other researchers.
Institutional partnerships and API access inquiries
We welcome collaboration with museum collection teams, academic researchers, and digital humanities programs.