How to make a real estate video from listing photos, without faking the house

Most listings have twenty good photographs and no video. Portals and social feeds both reward video, and nobody with forty listings a year is booking a videographer for each one. You do not need one. The photos are the footage.

The catch is specific to AI. Ask a video model to "animate this living room" and it will happily give the room a second window, turn the fireplace into a doorway and open a view onto a garden that is really a car park. Sell from that and you have advertised a house that does not exist.

1. Choose the photos and the order

Pick twelve to twenty photos, one per room unless the room is the reason to buy. Drop the near-duplicates and the close-ups of taps.

Put them in the order a buyer walks: the approach, the front door, the living space, the kitchen, the bedrooms with the main one first, the bathrooms, then the garden or the view. Ahead of all of it, put three seconds of the single best photo, whichever room sells the house, so the video opens on its strongest moment rather than on a front path. End outside, ideally at dusk.

Import the photos (Assets → Import files) and drag them onto the timeline in that order. A dropped photo gets four seconds, which happens to be about right: fifteen photos make a one-minute video, a good length for a portal or a website. When a room deserves longer, change it under Timing → Duration.

2. Move the camera, not the house

Give every photo a slow camera move. Select it, open Animation → Motion and set a Start window and an End window, and the editor glides between them. Nothing is generated, so nothing can be invented: the room on screen is exactly the room in the photo. That is why this is the default for every shot.

What works for property:

• Interiors: a slow push toward the room's focal point (the window, the island, the fireplace) or a gentle drift across a wide shot. About 10 to 15% over four seconds.
• Exteriors: a slow pull back from the front door to the whole house, or a drift along the frontage.
• Keep the verticals vertical. Property photography lives on straight lines, and a door frame that leans looks like a mistake. Move straight in, straight out or straight across, and never rotate.
• Alternate directions, so the sequence does not pulse in and out like a template. The Ken Burns effect, done properly has more on that.

3. Animate two or three photos, and only what moves anyway

Generated motion is worth spending on the two or three rooms that sell the house, and only for one kind of movement: things that move by themselves. Water in a pool, leaves on a tree, a fire in the grate, curtains in a breeze, clouds over the roof, sunlight sliding across a floor. With the camera held still, that kind of motion turns a photo into a moment, and the model has no reason to invent anything to produce it.

In Generate → Video → Animate, attach the photo and describe only the movement: "the water in the pool ripples gently, palm leaves stir in a light breeze, static camera". Four seconds is plenty.

Do not ask for the camera to walk through a door, turn a corner or circle the room. The moment a move reveals something the photo did not show, the model makes it up, and what it makes up is a room that is not there.

Then check each generated clip against its photo as if you were the buyer's surveyor. Count the windows and doors. Check the view through the glass, the fittings, the line of the ceiling. If anything has changed, the clip does not go in, however good it looks. Why AI video looks fake has the habits that catch this kind of change.

4. What you may change, and what you must say

There are three tiers, and the lines between them matter.

Fine without comment: the corrections a photographer makes anyway. Exposure, white balance, a little colour. Effects → Color does all three, per clip.

Allowed, but labelled: anything that shows the property other than as it was photographed. That includes the animated photos from step 3, virtual staging (furniture added to an empty room), a dusk exterior made from a daytime photo and a grey sky replaced with a blue one. A dusk version of the front is an image edit: attach the photo in Edit with references and ask for "the same house at dusk, warm light in every window, deep blue sky", then check it against the original, because an edit can quietly move a chimney. Label every such shot on screen while it is showing ("Virtually staged", "Digitally enhanced", "AI-animated"), and tick the platform's AI disclosure when you upload. In California this has been the law for licensees since January 2026 (AB 723): an altered listing image has to say so, and buyers must be able to see the original. Many listing services already require the same.

Never: removing permanent defects (the power line, the damp patch, the neighbour's wall), adding features (a fireplace, a view, a pool), or changing the size or layout of anything. That is misrepresentation, and in most places it is a legal problem as well as an ethical one. Your local rules and your portal's photo policy have the final word, but here is a test that rarely fails: if a buyer would be surprised at the viewing, it should not be in the video.

5. Words on screen

Price and bedrooms on the opening shot; location and floor area on the next. One short line each, on a text layer (the + on the overlay track, then Add text), with a background box if the photo behind it is busy.

After that, one label per room at most, and only when it says something the picture cannot: "Kitchen, renovated 2024", "South-facing garden". Small, low in the frame, two seconds each.

Finish on an end card with the agent's name, number and logo. The Graphics tab has end-card templates that take your brand kit's colours and fonts, which is the difference between a card that looks designed and one that looks typed. Titles that do not look like a template has the rest.

6. Sound

Instrumental music, calm, with no lyrics to compete with the words on screen. Generate a track at exactly the video's length (CassetteAI makes instrumentals from 30 seconds to three minutes, so set the duration to the cut) or use one you have the rights to.

A short voiceover in the agent's own voice is worth more than any music: three or four sentences, recorded with Record voiceover in the Assets panel. "Three bedrooms, ten minutes from the station, and wait until you see the kitchen." Flag the music to duck in its Audio tab and it dips under every word. How to mix voice over music has the levels.

7. Export twice

Export the full version at 1080p, 16:9, for the portal, the website and YouTube.

The vertical version for Reels, TikTok and Shorts is a re-edit, not a resize. Duplicate the project from the ⋯ menu and switch the copy's aspect chip to 9:16. A tall frame shows only about a third of a landscape photo, so the pushes become slow sideways drifts that travel across each room, and the edit keeps only the six to eight strongest shots, 20 to 30 seconds in all. Keep the text in the middle of the frame, where the apps' buttons will not cover it (this is where they sit), and upload the file to each platform directly rather than posting a link.

What the video is for

A listing video has one job: to get someone to book a viewing. The viewing then has to confirm the video. Show the house on its best day and never better than it is. A buyer who walks in and finds the room you showed them is halfway to an offer. One who finds a different room is gone, and so is their trust in your next listing.