The two jobs of an AI image
"Generating visuals with AI" is not one activity; two very different jobs share the label, with opposite requirements.
The first job is exploration: seeing an atmosphere, a material character or a massing idea quickly. Accuracy is not the point — speed and variety are. AI is strong here; fifty façade characters can be tested in an afternoon.
The second job is representation: telling a buyer, investor or committee "this is what the project will be". Here fidelity is the point: ceiling heights, window layouts, views and materials in the image must match the project. AI cannot do this job alone, because generative models know general building patterns from training data — not your CAD file.
- Exploration: speed and variety → AI is strong
- Representation: geometric and material fidelity → must be produced from project data
- Mixing the two ends with sales visuals that contradict the delivered building
The legal and commercial line on sales visuals
In off-plan sales, a visual shown to a buyer functions as a promise. The gap between image and delivered building is dispute material, and consumer law in many markets regulates misleading promotion explicitly.
An AI-generated image can be striking, but because it is not derived from the approved geometry it cannot carry the claim "your apartment will be like this". Window proportions, ceiling heights, balcony depths — the details buyers decide on — are incidental in an AI image.
Professional practice therefore labels the split: concept imagery presented as concept, sales-facing imagery produced from project data. MPANDO applies the same rule on its own sites: its concept films are explicitly marked as fictional demonstration material, never presented as delivered project evidence.
Where AI genuinely fits a digital twin pipeline
AI and the digital twin are not rivals; they are stations on the same line. In an efficient pipeline AI contributes in four places:
- Concept and direction: rapid scanning of material, atmosphere and landscape alternatives
- Environment fill: textures, vegetation and background outside the twin's focus areas
- Variant production: format and mood derivatives of approved hero visuals for marketing channels
- Text side: listing copy, multilingual drafts and social media adaptations
What buyers should be told
Buyer trust follows source transparency. The practical rule: if a visual was not produced from the project's data, it should not be implied to represent the project.
In the sales office this distinction resolves itself: every frame shown during a digital twin walkthrough comes from the project's real geometry — the buyer looks out of their own window toward the real view direction. That produces a kind of trust an AI image cannot.
Search and answer engines are learning the same distinction: content with unclear visual provenance earns less trust than sourced, consistent content. Transparent labelling is not only legally sound but also the right visibility strategy.
MPANDO's approach
MPANDO produces the MTWIN twin from approved project data (DWG, RVT, IFC); every sales-facing frame comes from that geometry. AI tools are used on the line for exploration, environment fill and content derivation — not for representation.
This is the production-side counterpart of the site's evidence-boundary principle: the source of anything shown to a buyer must always be stateable.