AI motion vs a camera move: what your listing video is actually made of

This is the guide we would want to read before buying an AI video tool, and it is mostly about a distinction the category does not like to draw.

Two completely different things are sold as "AI video"

A camera move — often called a Ken Burns push-in — takes your photograph and slowly scales and pans across it. Every pixel on screen came out of your camera. It cannot show a room you do not have, because there is no generation step. It is a hundred-year-old technique and it is entirely honest.

Generated motion hands your photo to an image-to-video model, which produces new frames. The model is not tracking a camera through a three-dimensional room; it is predicting what the next frame plausibly looks like. When it works, you get motion that feels like a real dolly shot — parallax, things passing behind other things, the sense of depth a still cannot give you. When it does not, you get a doorway that is not in your house.

Both are legitimate. They are not the same product, and a tool that ships the second while describing it as the first is describing your listing inaccurately on your behalf.

What goes wrong, specifically

These are the failure modes worth being able to name, because once you can name them you can see them:

• Invented geometry. A doorway, an archway, a window, a hallway that continues where a wall was.
• Fixture drift. Cabinet runs that gain a section, an island that changes shape, taps that migrate.
• Outlook changes. A view through a window that becomes a different view — this one is subtle and it is the one buyers are most likely to notice on site.
• Material substitution. Countertop stone, flooring, cabinet colour shifting mid-clip.
• People and pets. Figures appearing in an empty house.
• Non-rigid warp. The whole room subtly breathing — straight lines that stop being straight. Often the tell that a clip is generated rather than moved.

The uncomfortable part: these are not rare edge cases in the current generation of image-to-video models handed real interior photographs. They are common enough that shipping generated motion unchecked, on an advertisement for a specific address, is a real exposure rather than a theoretical one.

What we do about it, said plainly

Every generated segment is sent to a separate model before it ships, asked one question against the photo it came from: does this show anything that is not in the source photograph? A segment that passes ships as generated motion. A segment that fails is regenerated, and if it fails again it ships as a slow camera move over your real photo — and a push-in over your own photograph cannot invent a room, because there is nothing in the frame you did not photograph.

Two limits on that, both of which we would rather you heard from us. When the checker itself cannot be reached, the segment ships labelled not confirmed in your summary rather than as a pass — an unread verdict is not a clean one. And a segment that does pass can still get a detail wrong: the check is a model, not a proof, and we have not pinned down how often it lets one through. The only outcome that is safe by construction rather than by inspection is the camera move.

Your video summary lists which segments are which. That is the part that makes the claim checkable rather than a promise: you can look.

The number we will not round up

At the operating point our checker currently runs, roughly one segment of a 6-segment video is generated motion. The rest are camera moves over your own photographs.

We are not going to publish that as a percentage, a pass rate or a multiplier, and it is worth saying why rather than just declining. The measurement behind it comes from a small number of photos and a few dozen judged clips. Its confidence interval is wide enough that quoting a specific figure would be reporting noise to three significant figures. When we have production numbers from real jobs that survive an interval, we will publish those with the sample size attached. Until then the honest form of the number is "roughly one", and the honest form of the residual error rate is that we have not pinned it down.

Is a mostly-camera-move video worth having?

Only you can answer that, but here is the case, stated without spin.

What you get is a 6-segment, roughly 28-second video in 3 aspect ratios — 16:9, 9:16, 1:1, one file each — cut, crossfaded, scored, captioned and hosted, from photographs you already paid for, without opening an editor. Most of the motion in it is a push over your own photograph.

What you are not getting is six seconds-long hallucinated dolly shots of your listing. Nobody can sell you that honestly at this price: the models that produce grounded motion reliably cost several times more per second, and the ones that are cheap enough are the ones that invent rooms.

If a competitor is offering six fully generated segments for a comparable price, the arithmetic says one of three things is true — they are using a model that fails often, they are not checking, or they are absorbing a cost they will not absorb forever. Ask which.

How to check any tool's output yourself

You do not need our checker to do this. Take the finished video, pause it on the last frame of each segment — drift accumulates, so the end of a clip is where it shows — and compare against the source photo, looking specifically for:

1. Count the windows. Then count them in the photo.
2. Follow every straight line: door frames, counter edges, ceiling junctions. Straight lines are where warp is visible.
3. Look at what is through the glass.
4. Look at the far end of any hallway.
5. Count cabinet doors.

Do this once on a listing you know well and you will never watch an AI listing video the same way again. That is a reasonable thing for a vendor to teach you, because the alternative — you finding out on site, with the buyer — is worse for everybody.

The free tier is one watermarked teaser of 3 segments from your own listing photos, no card, so you can run that check on your own house before you decide anything.

General information, not legal advice. Advertising law, MLS rules and broker policy vary by market and change often. Confirm anything specific with your broker, your MLS, and — where money or rights are at stake — your own attorney.

ReeltyPro turns the listing photos you already have into a 6-segment video in 16:9, 9:16 and 1:1, and tells you segment by segment which parts are generated motion and which are a camera move over your own photograph.