Innovation Living has used 3D modelling for years, and began combining 3D product models with AI-generated environments and human figures to replace traditional photoshoots. The shift promised speed and flexibility — but quickly revealed a critical constraint: generative AI cannot reliably reproduce the physical product. Outputs altered dimensions, stitching details, seam lines, and structural proportions in ways that were often invisible to untrained eyes but immediately apparent to product experts. Because Innovation Living manufactures the furniture itself, these errors are not cosmetic — they misrepresent a product that must match what arrives in a customer’s living room.

This challenge sharpened as AI outputs became more photorealistic and therefore harder to evaluate. Meanwhile, retailers began generating their own AI imagery using downloaded product assets — sometimes producing representations that were simply wrong. One incident involved a published image with non-existent screws and a leg in the wrong position. The core problem is therefore not generation but judgment: ensuring that what is produced is accurate, trustworthy, and representative of the real product at scale.

The workflow Innovation Living has developed is a clear instantiation of what the HI Manifesto defines as the prediction–judgment frontier (Sherson et al., forthcoming). AI handles what it does well — generating environments, populating scenes with diverse human figures, and producing visual variations at speed and scale. But it cannot preserve product fidelity. That task requires human judgment: deep, embodied knowledge of how the furniture is constructed, how textiles behave, how dimensions relate, and how the product should feel when seen.

In practice, this means the workflow is never one-shot. The in-house specialists — combining deep 3D modelling and visual production expertise — work through multiple prompt iterations per image, evaluating each output against detailed knowledge of the physical product. They use 3D product models (packshots) as anchoring references, control for camera angle, lighting, and lens characteristics with domain-specific precision, and assemble final images from a combination of AI-generated and 3D elements. This reflects the Manifesto’s Human-in-the-loop 1 structure: high sub-task automation with structured human judgment at each iteration.

Critically, this is not general AI skill — it is firm-specific human capital. The ability to spot a wrong seam or an off-dimension cushion in an AI-generated image draws on years of accumulated experience in image production and deep product knowledge. This is precisely what the HI Manifesto identifies as the human-premium condition: as AI makes image generation abundant, value concentrates around the contextual knowledge required to ensure accuracy, authenticity, and trust.

The co-creative dimension is also visible in how the workflow shapes output diversity. AI enables Innovation Living to generate models representing a wide range of ages, ethnicities, and body types — a flexibility that would have been economically impossible with traditional photoshoots. Human judgment governs which outputs are used, how scenes are composed, and whether the final image is consistent with the brand’s Nordic aesthetic identity and seasonal trend direction. The result is not AI replacing craft, but AI expanding the palette within which expert craft operates.

Innovation Living has already deployed this workflow across major catalogue productions and continues to roll it out across markets. The next step is organizational extension: scaling the capability beyond the current specialists to additional team members, broadening the workflow’s reach within the organization. The ambition is to document the process — prompt examples, flow descriptions, quality criteria — as transferable organizational knowledge. This maps onto what the HI Manifesto calls HI capacity scaling: converting individual tacit expertise into reusable governance assets that can be transferred without losing the judgment layer.

The external dimension is equally significant. Innovation Living has developed a set of AI image guidelines — shared with partners and distributors across the supply chain — that explicitly address the risks of uncontrolled AI image generation. The guidelines help partners understand what responsible use looks like, how to spot errors, and when to involve Innovation Living directly. Partners who wish to generate images themselves are supported through a structured guidance process. This ecosystem-level effort reflects the HI Manifesto’s human-premium signaling dimension: making the human judgment embedded in the workflow externally visible, substantiated, and valuable — not just internally, but across the supply chain.