What Brands Look Like: How Computer Vision Is Transforming Visual Branding Research

Five million images are uploaded to Instagram every single day. A significant share of them feature a brand — a logo, a product shot, a face associated with a company. Humans process these visually, intuitively, and fast. Algorithms can now do the same at scale, and with more consistency than manual methods ever allowed.

This is not a peripheral development. By 2022, 34% of firms had embedded computer vision as a core AI capability — surpassing natural language processing and virtual agents. Yet its systematic application in branding research has lagged behind its technical development. Most studies borrow computer vision as a tool without grounding it in branding theory. The result is an empirically rich but conceptually fragmented field.

A new integrative review, co-authored with my wonderful colleagues Yaqiu Li and Hsin Hsuan Meg Lee, and published in the Journal of Business Research, maps exactly what we know, what we have consistently overlooked, and what the next decade of visual branding research should look like.

A framework that connects process to outcome

The paper proposes the CTV-CBBE framework, which connects the Computational Theory of Vision (Marr, 1982) with Keller's Customer-Based Brand Equity model. The logic is straightforward: CTV explains how visual systems — biological or computational — process images. CBBE explains why that processing matters for brand strategy. Linking them creates a structured vocabulary that the field has been missing.

The framework operates at three levels. At the computational level, it identifies which visual features are relevant to brand equity — from low-level attributes such as color, texture, shape, and spatial composition to high-level semantic content such as human presence, facial expressions, objects, places, and actions. At the algorithmic level, it maps how computer vision tasks (i.e., object detection, facial expression recognition, image generation, multimodal analysis) process those features. At the implementational level, it connects outputs to CBBE dimensions: brand identity, brand meaning, brand response, and brand relationships.

What 106 articles tell us

The review covers a decade of research across marketing, information systems, computer science, and management. The growth rate alone is notable: a 66% average annual increase in computer vision branding studies since 2018. This is not a niche methodology.

Engagement dominates as the most-studied branding outcome (27 articles), driven by the prevalence of visual platforms with measurable interaction metrics. Brand identity, paradoxically, is the least studied dimension (just 7 articles) despite being foundational to everything else.

The practical applications documented in the literature are concrete. L'Oréal deploys computer vision to analyze skin needs, age, gender, and ethnicity in real time, personalizing product recommendations at the individual level. Research on Airbnb demonstrates that cover image attributes — room type, brightness, spatial clarity — directly predict booking demand. Studies in physical retail show that facial expression recognition can inform merchandising strategy before customers articulate a single preference.

Three structural shifts define the field's trajectory: from single visual features to integrative constructs like aesthetics, complexity, and similarity; from static images to dynamic content, including video and servicescape design; and from single-modality analysis to multimodal approaches combining image, text, and audio. Each shift enables a qualitatively different class of research question.

What the field keeps missing

The review is as useful for what it finds absent as for what it documents. Texture and shape receive far less research attention than color and human presence, despite evidence that they affect brand attitude and purchase intention. Motion — body language, gestures, pose, logo dynamics — is almost entirely absent from branding research, despite the volume of video content brands now produce and the technical tools already available to analyze it.

Place, or environmental context, is studied primarily in tourism research and rarely applied to service brand credibility, despite clear relevance. Consumer well-being as a branding outcome is nearly invisible in the computer vision literature. Personal brand credibility, directly relevant to virtual influencers and human brand equity, is explicitly raised in the research agenda as an open question.

Longitudinal studies on brand identity evolution are scarce. Most work provides a cross-sectional snapshot rather than tracking how visual brand associations shift over time and across platforms. The tools to do this exist. The studies largely do not.

Where generative AI enters

Image generation is rapidly becoming a practical tool for branding. Models can already assist in logo redesign, packaging iteration, and design consistency evaluation. Text-to-image models are less studied in branding contexts but have obvious implications for scalable visual communication. The challenge is not technical capability; it's knowing what to generate and why, which requires exactly the kind of theoretical grounding the CTV-CBBE framework provides.

Multimodal analysis opens another frontier. Combining image, text, and audio signals allows researchers to study discrete emotions — envy, awe, interest — that are difficult to capture through any single channel. This has direct implications for theories of consumer emotion that have been constrained by the limitations of single-modality data.

What this means in practice

For researchers: the white spaces in this review are not marginal gaps. They are core branding constructs, identity, credibility, well-being, loyalty, that computer vision is now technically capable of addressing. The infrastructure for rigorous visual analysis exists. The theoretical framing to interpret it is what the CTV-CBBE framework provides.

For practitioners: if your brand produces visual content at scale (e.g., social media posts, product imagery, video, servicescapes…), you likely already have a visual analytics problem you haven't framed as one. The question is not whether computer vision applies to your strategy. It's whether you have a framework to interpret what it finds, and whether your research partners do.

Visual data is not just a content output. It is evidence about how your brand lives in the world. It should be treated as such.

The full paper is available open access in the Journal of Business Research: doi.org/10.1016/j.jbusres.2025.115329. If you're working on visual analytics in marketing or exploring computer vision for brand research, reach out or follow for more on the intersection of AI, visual data, and consumer behavior.

Source:

Li, Y., Lee, H.H.M., & Blasco-Arcas, L. (2025). Computer vision in branding: A conceptual framework and future research agenda. Journal of Business Research, 193, 115329. https://doi.org/10.1016/j.jbusres.2025.115329