Matching Influencer Visual Style to Brand Identity

A deep computer-vision measure of brand–influencer visual-style congruence

Qingli Zeng · Hebrew University of Jerusalem · ISMS Marketing Science 2026

The tension — and the construct

• Creators earn trust through a consistent visual style (a signature).
• Sponsored posts impose brand aesthetics → can break that signature.
• We quantify visual-style congruence: how aligned a brand's look is with a
  creator's established look.

We measure style, not content

• Same apple, five centuries — content constant, style varies. We measure how it
  looks (colour, lighting, composition), not what's shown.
• (Prior "visual congruence" classifies image content — Argyris et al. 2020; we measure style.)

The measure (1) — from image to style vector

• Each frame → a frozen ImageNet CNN (VGG19) → feature maps at 5 layers.
• At each layer, compute the gram matrix = correlations between feature channels,
  summed over all spatial positions.
• Summing over space discards "where" ⇒ keeps texture, colour, and feature
  co-occurrence — style, not layout.
• → a per-image style vector (concatenated across the 5 layers).

The measure (2) — signatures & congruence

• Average a creator's frames → their style signature; a brand's frames → the brand signature.
• Congruence = cosine between signatures (mean-centred to emphasise distinctive style).

Brands have distinct — and meaningful — styles

• Pairwise similarity of brand signatures → fast-food, athletic/outdoor, and
  beauty clusters emerge — face-valid, unsupervised.

Matching creators to brands

• Each creator's best-fit brand (cyan) — and the spread is intuitive.

On real frames — andrewiwanicki → The North Face

• Alpine/hiking creator → North Face — match driven by palette & light, not an object detector.

Faces blurred for privacy · frames © respective owners, shown for non-commercial illustration.

On real frames — nicolekauffman3 → Red Bull

• Outdoor/ski creator → Red Bull's high-energy adventure aesthetic (our strongest match, 0.78).

Faces blurred for privacy · frames © respective owners, shown for non-commercial illustration.

And it discriminates

• That ski creator's top-3 brands are all sport/outdoor; bottom-3 all beauty (negative).
• The signal is specific — not "everything matches everything."

Related work — and the gap

• Style is computable — neural style transfer (Gatys et al. 2016); stylometry (Elgammal et al. 2018).
• Marketing studies image content, rarely style (Klostermann 2018; Liu 2020; Hartmann 2021).
• Closest "visual congruence" is content-based — Argyris et al. (2020, Comp. in Human Behavior): influencer–follower theme overlap.
• Endorser–brand fit matters — match-up hypothesis (Kamins 1990).
• Sponsored content strains brand–creator authenticity — Arsel et al. (2025, J. Marketing).
• Creators respond to engagement — Mummalaneni et al. (2026, Marketing Science).
• Gap: style congruence · brand–influencer · nonlinear authenticity boundary.

What this enables

• A scalable, content-agnostic measure of brand–influencer style fit.
• Distinct from content-theme congruence (Argyris 2020) and from survey-based fit.
• What's next: the marketing questions it lets us answer, and a video extension →

Next (1) — the marketing questions

• Does congruence drive engagement & conversion on sponsored posts?
• Is there an authenticity boundary — a threshold where too much brand control backfires?
• Does it shift by brand status/price, hedonic vs. utilitarian, brand–influencer fit?
• → when to enforce brand aesthetics vs. preserve creator authenticity?

The boundary shifts by brand & product type

• Luxury / high-status: narrow tolerance — high congruence essential.
• Mass-market: wider tolerance — the constraint relaxes.
• Also: price · hedonic vs. utilitarian · brand–influencer fit.

Next (2) — from frames to video

• Today: a video = the average of its frames (the look).
• Add temporal style — editing pace (cut rate), motion energy, colour rhythm,
  and within-video style drift (consistency vs. variety).
• → congruence per video: the same unit as engagement & conversion.

Takeaway

Visual style is measurable, brand-specific, and matchable — at scale.

A working paper — feedback welcome.