A founder’s AI photo-shoot experiment exposes the “taste gap” in fashion imagery
Emily Oberg’s recent test of AI-generated fashion imagery—positioned as a potential substitute for traditional campaign production—lands at a moment when brands are under intense pressure to do more with less. A single collection shoot can run $200,000 or more, and generative tools promise a seductive alternative: instant scenes, endless variations, and a dramatically lower marginal cost per image.
Yet Oberg’s takeaway, shared via Substack, is less a rejection of technology than a precise diagnosis of its current creative ceiling. The outputs, she argued, felt “lifeless”—technically plausible but emotionally inert. That critique matters because fashion photography is not merely product depiction; it is brand meaning-making. The most effective images compress narrative, aspiration, and cultural context into a single frame—an alchemy that depends on human decisions about tension, imperfection, and attitude.
In practical terms, Oberg’s experiment highlights a widening industry reality: AI can replicate aesthetics, but it struggles to originate taste. Taste is not a style preset; it is a lived, evolving sensibility shaped by memory, subculture, and risk. Generative models learn correlations from data. They do not “want” anything, and they do not experience the social stakes that make an image feel current—or daring.
—
Why today’s generative models still struggle with emotion, motion, and intent
The limitations Oberg observed—flat affect, uniform expressions, static posing—map closely to how mainstream diffusion models and GAN-derived workflows operate in production settings. These systems are exceptionally good at pattern completion: they infer what a “fashion campaign” should look like based on statistical regularities in their training data. What they often lack is contextual intent—the why behind the image.
Several technical and creative constraints converge here:
- Training-data gravity and aesthetic averaging: When models are trained on large volumes of polished editorial imagery, they tend to reproduce a “center of mass” look—clean, symmetrical, and safe. The result can be a kind of visual consensus rather than a point of view.
- Weak understanding of embodied styling: Great fashion images communicate how garments behave in real life—weight, friction, drape, tension at seams, the micro-chaos of movement. AI can simulate these cues, but often without the physical logic that makes them feel lived-in.
- Expression without interiority: Faces can be rendered convincingly, yet still read as vacant because the image lacks a believable inner narrative. Viewers are highly sensitive to subtle signals—eye focus, asymmetry, timing—that convey intentional emotion.
- Promptability vs. directability: A photographer can direct a model toward a specific psychological beat; a stylist can introduce a disruptive element; a creative director can pivot the entire concept mid-shoot. AI systems respond to prompts, but they rarely replicate the iterative, interpersonal negotiation that produces iconic frames.
This is why the most credible near-term path is not replacement but human-in-the-loop creation: AI as an accelerator for ideation, compositing, and iteration—paired with human judgment for final selection, narrative coherence, and brand alignment.
—
The economics: cost savings are real, but brand equity is the balance sheet you can’t easily audit
From a CFO’s perspective, AI-generated imagery offers clear line-item appeal: fewer shoot days, reduced travel, smaller crews, and faster turnarounds. In an environment shaped by inflation, cautious consumer spending, and margin pressure, those efficiencies are not theoretical—they are operationally meaningful.
But Oberg’s critique points to the harder question: what is the ROI of emotional resonance? In fashion and lifestyle, the image is often the product’s first—and sometimes only—moment of persuasion. If AI imagery reduces distinctiveness, the brand may pay in less visible ways:
- Dilution of brand equity: If visuals feel generic, consumers may struggle to articulate why one label is worth a premium over another.
- Lower engagement in crowded channels: AI can increase content volume, but volume can also become noise. The risk is signal decay—more posts, fewer memorable impressions.
- Hidden costs of homogenization: As more brands use similar tools trained on similar datasets, the industry can drift toward aesthetic convergence, making offerings feel interchangeable and forcing renewed spending to reclaim differentiation.
This reframes the savings conversation. The strategic question is not simply “Can AI reduce production costs?” but “Where should the savings be reinvested to protect the brand’s distinctiveness?” For many labels, the best answer may be more human creativity, not less: stronger creative direction, better casting, more ambitious concepts, and richer storytelling across channels.
—
Competitive strategy: co-creative workflows, authenticity signaling, and the next frontier of virtual retail
Oberg’s experiment also arrives as fashion explores digital twins, virtual showrooms, AI avatars, and virtual runways. These experiences demand more than photorealism; they require believable motion, personality, and emotional timing—precisely the areas where current tools can falter.
A pragmatic competitive playbook is emerging:
- Tiered deployment: Use AI for mass-market speed—catalog variants, localization, background generation, rapid A/B testing—while reserving fully human-led production for flagship campaigns where cultural credibility and narrative depth matter most.
- Talent as a moat: The most valuable creatives will be those who can collaborate with AI—translating taste into prompts, curating outputs with rigor, and knowing when to reject “good enough” for something sharper.
- Authenticity and provenance: As AI content proliferates, transparency may become a differentiator. Regulatory and platform pressures could push brands toward clearer labeling of AI-generated content, making authenticity not just an aesthetic choice but a compliance and trust issue.
There is also a deeper, non-obvious layer: findings from neuroaesthetics suggest that humans respond to art through reward pathways shaped by novelty, meaning, and perceived intention. If audiences intuit that an image lacks intention—no lived perspective behind it—the work may fail to trigger the same engagement, even if it is visually competent.
Oberg’s “lifeless” verdict, then, reads less like nostalgia and more like market intelligence. AI will continue to reshape creative operations, but the brands that win are likely to treat generative tools as infrastructure, not authorship—using automation to scale execution while defending the scarce asset that still cannot be synthesized on demand: human taste with something at stake.
By
By
By
By
By
By
By
