Short answer
AI image tools like Nano Banana Pro make renderings nearly free, but a generated image in a listing or offering memorandum is still a factual claim about the property. The safe workflow is to treat every image as a draft: check it against the site plan and rent roll, label it as a rendering, and have a human approve it before it ships.
Why does a "free" AI image become an expensive problem?
The image is free to make. The claim it carries is not. When a generated visual sits inside an offering memorandum (OM) — the marketing and financial summary a seller gives qualified buyers — next to real square footage, rent roll figures, and cap-rate math, a reader treats the picture as part of the representation. Cheap generation raises volume; it doesn't lower liability.
Generative image use is now mainstream in marketing. According to Adobe's 2024 Future of Creativity study, a large majority of marketers already use or plan to use generative tools in their workflows — which means the review question isn't hypothetical anymore. The tools are already in your leasing team's browser.
- A rendering shows six loading bays; the site plan permits five. Now your marketing overstates the asset.
- A cleaned-up facade hides deferred maintenance a buyer will find in diligence.
- A "staged" suite implies improvements that aren't built and aren't budgeted.
Each of these is a mismatch between the image and the record. That mismatch is the actual risk — not the technology.
Is it legal to use AI-generated images in a listing?
Using a generated image isn't itself illegal, but the underlying rule is old and strict: you can't materially misrepresent a property. The NAR Code of Ethics, Article 2, obligates realtors to avoid misrepresentation and concealment of pertinent facts, and Standard of Practice 12-1 addresses truthful advertising. State real estate commissions carry parallel prohibitions on misleading advertising, and the FTC Act, Section 5, bars deceptive practices in commercial promotion.
An altered or fabricated image that changes what a reasonable buyer would understand about the asset can trip all three. The safe reading: a rendering is fine when it's clearly a rendering and accurate to the plan; it's a problem when it's presented as, or mistaken for, the property as it exists.
What should a CRE team actually check before an image ships?
Treat every generated image the way you'd treat a number in the rent roll — as a claim to verify, not a finish. Run each one against the source of record before it leaves the building.
- Match it to the site plan. Bay count, floor count, parking, setbacks, footprint. If the pixels disagree with the entitled plan, the pixels are wrong.
- Match it to the rent roll and stacking plan. Don't show occupied space as vacant, or imply a tenant mix the rent roll doesn't support.
- Label the rendering. Mark it "artist's rendering" or "conceptual — not as built." A label converts a possible misstatement into an honest illustration.
- Keep the source. Save the prompt, the model, and who approved it. If a buyer questions the image later, you can show the trail.
- Have a human approve. No generated image ships to a listing, email, or OM without a named person signing off.
This is Vantrow's governing idea in practice: propose, never commit. The system can draft the image; a human approves it; the decision lands on an audit trail.
Where does Nano Banana Pro fit — and where does it stop?
Nano Banana Pro is Google's higher-fidelity image generation model (a Gemini-based release), covered for a CRE audience in the ChatCRE newsletter's December 2025 edition. It produces sharper, more controllable renderings than earlier tools — useful for concept visuals, marketing comps, and quick "what could this look like" drafts.
What it does well:
- Fast concept renderings for pre-lease and pre-development marketing.
- Consistent style across a marketing set.
- Cheap iteration before you commit to a photographer or 3D firm.
Where it stops: the model has no idea what your site plan permits, what your rent roll says, or what's actually built. It generates plausible pixels, not verified facts. It won't store or check your property data for you — the burden of matching the image to the record stays with your team. That gap is exactly where the human review step earns its place.