Deterministic, not generative: why we edit photos instead of inventing them

There are two kinds of AI in image work right now, and the industry keeps talking about them as if they're the same thing. They are not the same thing. One of them belongs in real estate marketing. The other one belongs nowhere near a contract of sale.

Generative AI invents. You hand it a photo and a vibe, and it paints new pixels: a sky it dreamed up, grass it grew from statistical noise, a bench edge it guessed at because the real one was in shadow. The results can be genuinely spectacular. They can also be fiction. And the model cannot tell you which one you just got, because it doesn't know. There is no flag that says "I made this part up." Invention is not a failure mode of these systems. Invention is the product.

Deterministic AI decides. You hand it a photo and it makes editing decisions. Exposure, colour, tone, alignment, bracket blending: the same decisions a professional editor makes, applied to the pixels your camera actually captured. Run it twice, you get the same answer twice. Run it ten thousand times, you get ten thousand consistent answers. Nothing is invented. Everything is edited.

LUXengine is the second kind. On purpose. Permanently. This post is about why that's not a technical footnote. It's the entire ballgame.

A listing photo is not content. It's a representation.

Let's be precise about what a real estate photograph legally is, because the AI industry has never once stopped to ask.

A listing image is a representation of an asset in a transaction. In Australia, misleading conduct in real estate marketing sits under consumer law with real teeth. The UK, most of the US, most of Europe: same story, different acronyms. When a buyer walks into a home and it doesn't match the photos, that's not a vibe problem. That's the foundation of a complaint, a deal collapse, or a courtroom.

Now hold that next to what generative editing actually does. It rebuilds the kitchen bench because the reflection confused it. It straightens the fence that genuinely leans. It fills the brown January lawn with lush April turf. It invents a view through a window it couldn't resolve. Every one of those is the model doing exactly what it was built to do, which is plausible pixel invention. And every one of them is, in a listing context, a misrepresentation of an asset somebody is about to spend seven figures on.

Here's the sentence that should be printed on the box of every generative tool sold into this industry: plausible is not the same as true, and in real estate, the gap between them has a dollar value.

The industry already knows. That's why it flinches.

Talk to agents about AI imagery and watch their faces. They've seen the six-fingered hands. They've seen the mirror that reflects a room that doesn't exist. They've seen the "enhanced" twilight where the neighbour's house lost a storey.

That flinch is not technophobia. It's professional risk assessment, and it's correct. An agency principal doesn't care that the tool is impressive. They care that one hallucinated detail in one campaign puts their licence, their vendor relationship, and their reputation on the line simultaneously. The photographer takes the blame, the agency wears the exposure, and the vendor, who paid for a premium campaign, gets a dispute instead of a sale.

The generative crowd's answer to this is "the models are getting better." Sure. Better at what, exactly? Better at inventing more convincingly. That is the opposite of reassuring. A bad hallucination gets caught in review. A good one gets published.

Editing was never a creativity problem

Here's what the Silicon Valley end of this conversation has never understood about professional real estate production, because none of them have ever sat in one.

Elite property editing is not a creativity discipline. It's a consistency discipline. The craft is real and it is deep. But the job, night after night, is this: merge the brackets, balance the window pull so the harbour doesn't blow out, correct the verticals because architecture has rules, land the colour exactly where this client's brand lives, and do it identically across forty frames so the campaign reads as one coherent piece. Then do it again for the next shoot. And the next. Ten thousand times a month, to deadline, forever.

You do not need a machine that can imagine. You need a machine that can decide. Correctly, repeatably, at full resolution, at 2am. That is precisely the problem shape deterministic systems were born for, and it's precisely the shape generative systems are wrong for. Asking a generative model to do production editing is hiring a novelist to do your accounting. The prose will be lovely. The numbers will be made up.

What deterministic buys you, concretely

• Provenance on every pixel. Everything in the output traces to the input. Nothing appears that your camera didn't capture. When a vendor asks "is this what my house looks like?", the answer is yes: professionally presented, honestly represented. And you can prove it.
• Repeatability as a feature, not an accident. The same shoot processed tomorrow matches today. February matches November. Your brand look holds across an entire office's output, which is something no prompt has ever guaranteed anyone.
• Review that means something. Our platform routes output through member review before delivery. Reviewing a deterministic edit is checking craft. Reviewing a generative edit is playing spot-the-fabrication against a machine specifically optimised to fool you.
• No lottery. Generative tools hand you four candidates and a prayer, and the reroll button is a tax on your time. A deterministic pipeline hands you the edit. If the edit is wrong, it's wrong in a way you can name, fix, and never see again, because the fix becomes part of the system.

We're not anti-generative. We're anti-fiction in listings.

Generative tools have legitimate, exciting places in property: renovation concepts clearly labelled as concepts, virtual staging disclosed as staging, marketing creative that isn't representing the physical asset. We watch that space with genuine interest.

But when the image is the asset, when it's the thing a buyer relies on to spend the biggest money of their life, invention is not a feature. It's a liability with good lighting.

Real photos. Real edits. Deterministic, auditable, repeatable. That's not us being conservative. That's us being the only kind of AI you can actually build a professional media business on. The rest of the industry will arrive at this conclusion eventually.

We just got there first, because we came from production. And production doesn't get to publish fiction.