Pick 8 to 12 sharp listing photos in walk-through order, turn each into a 4 to 6 second clip with an image-to-video tool or a plain pan and zoom, and add a voiceover and captions. Label any AI-generated motion on the platform and in the listing. Tool credits run about $3.50 to $17.50 a video; your time costs more.
What needs a label depends on what the AI did, not on whether you used AI at all. A pan across the real photo is ordinary editing. A model that invents new frames of a real property is not, and YouTube and TikTok both ask you to say so. Listing services add their own rules for altered images, and the rules I read cover photos but do not mention video, so treat the video like an altered image: label it and keep the originals.
Pick photos that survive motion and order them as a walk-through
Start from the full-resolution files the photographer delivered, not the compressed copies on the listing page. An image-to-video model can only work with what it sees, so a soft or dark photo gives a soft or dark clip with more room for invented detail. Before you upload anything to a third-party tool, ask the photographer whether the licence covers editing and AI processing. Some licences allow marketing use but not derived versions, and a one-line email settles it.
Judge each candidate photo on five tests:
- Sharp and level. Vertical lines should be straight and window frames readable.
- No people, pets, number plates or the photographer's reflection. A model will animate all of them, and they are hard to explain afterwards.
- Subject near the middle. A 9:16 vertical export keeps only just over a third of the width of a standard 3:2 photo, so a sofa pushed to the left edge is cropped away.
- One photo per room. Two views of the same room give you two clips that look alike.
- A plain first frame. The opening frame becomes the thumbnail, so open on a move that cannot change the property.
Order the clips the way a buyer meets the home: the front of the building, the hall, the main living space, the kitchen, dining, bedrooms, bathrooms, outdoor space, then a closing card. The table below is the shot list for the example I cost out later, an illustrative three-bedroom house where the photographer delivered 26 photos and nine make the video.
| # | Photo | Seconds | Camera move | Made with |
|---|---|---|---|---|
| 1 | Front elevation at dusk | 4 | Slow push-in | Plain zoom |
| 2 | Entry hall | 4 | Push-in towards the staircase | AI clip |
| 3 | Living room, wide | 5 | Drift right along the sofa wall | AI clip |
| 4 | Kitchen from the dining side | 5 | Push-in towards the island | AI clip |
| 5 | Dining area | 4 | Slow pan left | Plain pan |
| 6 | Primary bedroom | 5 | Push-in from the door | AI clip |
| 7 | Primary bathroom | 4 | Slow zoom on the vanity | Plain zoom (mirror and glass) |
| 8 | Rear garden and patio | 5 | Slow pan right | Plain pan (planting must stay true) |
| 9 | Closing card | 6 | Still: agent name and contact line | Text on a plain colour |
That adds up to 42 seconds with four generated clips and five plain moves. The entry, living room, kitchen and primary bedroom are the generated ones because each has depth and a clear path for the camera. The front elevation, bathroom and garden stay plain: a model asked to move through foliage, glass or a mirror tends to invent leaves, reflections and extra fittings, and the garden is exactly what a buyer expects to find unchanged.

The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowThree kinds of motion, and which ones need a label
Every tool on the market does one of three things, and the label rules follow the thing, not the brand.
| Kind of motion | What happens | Accuracy risk | Label |
|---|---|---|---|
| Plain pan and zoom | A virtual camera moves across the real photo. No new pixels are generated. | Low: it only crops | No AI label needed |
| Generated motion | A model invents the frames after the photo, including parts of the room the photo never showed. | Medium to high: lights, windows and furniture can multiply, drift or vanish | Yes: platform setting, a line in the video, a line in the listing |
| Changed contents | Furniture added or removed, sky or lawn replaced, floors or walls restyled. | Highest: it changes what is for sale | Yes, on every altered image with the original beside it; some services bar certain changes |
Platforms draw the line in similar places. YouTube asks creators to disclose realistic content that AI generated or altered, and its own examples include AI-generated extra footage of a real place and altered footage of a real place. You choose Yes under "AI use" in the Attributes section of YouTube Studio, and YouTube then adds a label in the player or the expanded description. It says cloning your own voice for a voiceover needs no disclosure, and neither do sharpening, upscaling or colour and lighting adjustments.
TikTok wants a label on content that is completely generated or significantly edited by AI, and on all AI-generated content that contains realistic images, audio or video. You can say so in the caption, with a sticker, or by switching on the AI-generated content setting before you post, and it may remove realistic AI content that carries no label. Meta adds an "AI info" label when it detects industry-standard AI indicators in a file or when people disclose AI use, so a clip from a tool that embeds those signals can be labelled whether or not you tick a box.
Listing services write their own rules, and they are the ones that can penalise you. The 2026 rules I read from one large service ask for a label such as "digitally altered" or "virtually staged" in the photo description, the unaltered original immediately before or after each altered image, and no AI-generated landscaping at all. A first breach brings a warning, and an uncorrected one a fine. Those rules name images and do not mention video. Until your own service answers in writing, apply the photo rule to video, and read AI disclosure for listing photos and copy for the still-image side. Costs and label rules for staged rooms sit in the virtual staging tutorial, so I will not repeat them here.
If you have buyers or sellers in the EU, the AI Act's transparency rule for deployers covers AI-generated or manipulated video that resembles existing places or objects and could pass as real. The content has to be disclosed as artificially generated or manipulated, clearly and no later than the first exposure, and Article 50 has applied since 2 August 2026. Check the current text and any grace periods for tool providers with a qualified adviser rather than relying on a summary like this one.
Four routes to motion: editor, Runway, Gemini and Canva
List prices below were read from the vendors' own pages in late October and early November 2026. Credits and plan names change often, so confirm them before you pay.
| Route | List price | What it gives you | Watch for |
|---|---|---|---|
| Free phone or desktop editor | $0 | Keyframed pan and zoom on the real photo, exported at 9:16 and 16:9 | Nothing is generated, so nothing to audit, but the moves can look flat |
| Runway | Standard $15 a month ($12 on yearly billing) with 625 credits; Pro $35 ($28) with 2,250; Free is a one-off 125 credits | Gen-4 Turbo, its cheaper image-to-video model, at roughly 5 to 6 credits a second; Gen-4.5 at 12 a second; Kling 4.0 and other companies' models on paid plans | Every re-roll spends credits, so a bad clip costs real money |
| Gemini app | A paid Google AI plan (Google AI Pro lists at $19.99 a month; check which tier includes video) | Turns up to five photos into a 10-second video, with an invisible SynthID watermark in the file | Clips are 10 seconds long, and limits vary by tier |
| Canva | Pro $144 a year; Free has limited AI access | Image to Video is listed as a Premium AI tool, and Pro carries 10 times the Free AI allowance | The pricing page gives no exact allowance, so test one clip before buying |
For most agents I would start with a Runway Standard month or a Google AI plan if you already hold one, because both take a photo and a one-line instruction, and keep a free editor for the plain moves and the final assembly. Presenter-style videos, where a digital avatar reads the script, are a different job with their own consent questions, covered in the HeyGen avatar tutorial.
Prompts that keep a room the same room
An image-to-video prompt needs three things: one camera move, how long and how fast, and a plain statement of what must not change. Leave out adjectives such as "stunning" or "cinematic". They invite the model to add steam, sunbeams and props, which is the opposite of what a listing needs.
[Camera move] for [number] seconds at a steady, slow speed.
Keep every [fixed features] exactly as in the photo.
Nothing else moves. No people, no pets, no new objects,
no change to light or colour.
Filled in for the kitchen in the shot list, the prompt reads:
Slow push-in towards the island for 5 seconds at a steady speed.
Keep the three pendant lamps, the two stools, the window and every
cabinet exactly as in the photo. Nothing else moves. No people, no
pets, no new objects, no change to light or colour.
Compare that with the prompt most people type first, "make this kitchen look amazing and cinematic". An illustrative result from a prompt like that: a kettle starts steaming, a bowl of lemons appears on the island and the light turns amber. Each change is small, none is true, and a buyer who visits will notice all three. The same photo with the tight prompt above gave this illustrative first render: a smooth push-in, but the three pendant lamps became four by the third second and the window blind drifted down.
Two fixes work. Trim the clip at 2.5 seconds, where the lamps still match the photo, and hold the last good frame for a second. Or run the same prompt again, which tools call a re-roll, and keep the first attempt that passes the frame check. If three re-rolls fail on the same shot, stop spending credits and switch that shot to a plain push-in. The photo is the product; the motion is decoration.
The frame check: compare every clip with its source photo
A clip that looks good at full speed can still be wrong. Scrub each generated clip at half speed with the source photo open beside it, and work through this list. It takes about two minutes a clip.
- The first frame matches the photo exactly. If it does not, the tool has re-cropped or re-lit it.
- Count anything that comes in sets: windows, doors, pendant lamps, stools, stair balusters, cabinet doors, ceiling lights. The count at the last frame must equal the count in the photo.
- Check the fixed features: ceiling height, room width, flooring pattern, fireplace, built-in shelving, the view through each window.
- Look at mirrors, windows, oven doors and television screens. Reflections are where invented detail shows up first.
- Watch the edges of the frame. As the camera moves, the model has to invent what lies beyond the photo, so look for new objects entering from the sides.
- Compare light: time of day, window brightness, shadow direction. No sunbeams or lamps that were not on in the photo.
- Read any text in shot, such as a sign, a label or a number. Generated text often warps or turns to gibberish.
- Pause on the last frame and compare it with the photo, then compare it with the first frame of the next clip. A hard cut between two rooms should not reveal a different room.
The rule is blunt: any change to a fixed feature, and any object added or removed, means the clip is rejected, however pretty it looks. Differences in how light falls across a wall are normal for generated motion, and they are the reason the video still carries a label.
A typical failure, shown here as an illustration: a dining room clip passes the first-frame check, but a second window appears on the left wall in the last second. Only the window count on the final frame, step 2 above, catches it, and the listing sheet confirms there is one. A buyer who sees two and arrives to find one has a fair complaint, and the seller has a fair complaint about a video that adds features to the house. The fix is to replace that clip with a plain pan. Staircases, patterned floors, glass and water cause the same kind of trouble, so give those shots a plain move by default.
Voiceover: your own voice, a stock voice or a clone
A 40-second video carries about 100 words at a conversational 150 words a minute. That is room for three facts and a close, not a tour. Take the facts from the listing sheet, not from the model, and read the draft against the sheet before it is voiced. A language model asked to "write a voiceover for this house" will happily add a south-facing garden or a quiet street that nobody told it about. The prompt and checking steps are the same as in the fair-housing tutorial for listing descriptions, and the same wording rules apply to speech: describe the home, never the people who should live in it or the people who live nearby.
This filled-in script runs about 80 words for the example house, which leaves a few seconds of music at each end, and every fact in it would come from the listing sheet:
Three bedrooms, two bathrooms, and a kitchen that opens onto the dining area.
The living room runs front to back, with doors straight out to the patio.
The primary bedroom has its own bathroom and a walk-in wardrobe.
The garden is fenced, with a lawn and a covered seating area.
The house is on the market now. For a viewing, message [your name] using
the number on screen, or find the full gallery and floor plan in the listing.
Before and after on wording, using illustrative lines of the kind a model produces:
- Before: "A perfect family home near the best schools." After: "Three bedrooms, with a fourth room that works as a study." The first line steers on who belongs there and makes a claim about schools you cannot verify.
- Before: "Quiet, safe street." After: "The house sits at the end of a no-through road." Say what is physically there.
- Before: "Stunning, sun-drenched kitchen." After: "The kitchen faces south and has a window over the sink." Only keep orientation if the listing sheet says it.
You have three ways to voice the script. Your own voice costs nothing and carries no label on YouTube, which says cloning your own voice for a voiceover needs no disclosure. A stock AI voice reads the script cleanly, but TikTok asks for a label on realistic AI-generated audio, so switch the setting on there. A clone of your own voice is possible on ElevenLabs, where Professional Voice Cloning needs the Creator plan or above and a verification step that confirms the voice is yours, but it earns its keep only if you narrate dozens of videos. ElevenLabs does not allow a professional clone of anyone else's voice, even with their consent, so a colleague who wants to narrate should record the script in their own voice.
ElevenLabs lists Starter at $6 a month and Creator at $22 a month, with a first-month offer shown on the page, giving about 30 and about 121 minutes of speech, and its page includes a commercial licence from Starter upwards. A 100-word script with three takes uses about two minutes, which is roughly 40 cents of a Starter month.
Label wording you can paste, and where it goes
Say what happened in plain words, near the start, in three places. TikTok asks creators to disclose directly on the post, and the EU rule says no later than the first exposure, so a line in the caption is better than a note buried in the comments.
First, an on-screen line in the opening seconds or the closing card:
Some clips use AI-generated motion. Rooms and features are shown as photographed.
Second, a line in the caption or description:
Video made from the listing photos. Four clips use AI-generated camera motion; the rooms, fittings and views are unchanged from the photographs. The original photos are in the gallery.
Third, a note in the listing's media description, which is where a listing service will look:
Video: AI-generated motion applied to photos 2, 3, 4 and 6. Originals in the photo gallery.
Then tick the platform's own control: "AI use" set to Yes on YouTube, the AI-generated content setting on TikTok, and the disclosure option on Instagram and Facebook where it is offered. If the tool wrote AI metadata into the file, the platform may add its label anyway. The step-by-step menus for each platform are in the tutorial on labelling AI images and video on social media.
Avoid the word "enhanced" for generated motion. It sounds like a brightness tweak, and listing rules separate routine corrections such as lighting and colour from changes to what the photo shows. Avoid calling the video a "virtual tour" as well: that term implies a captured walk-through, which a generated video is not. If even one clip is generated, label the whole video.
What one listing video costs, line by line
Costing the nine-clip example: four generated shots, three generations each on average gives seven generations of five seconds, which is 35 seconds of generated video. At a budgeted 6 credits a second for Gen-4 Turbo that is 210 credits, and at 12 a second for Gen-4.5 it is 420.
| Line | Cheaper model path | Higher-quality path |
|---|---|---|
| Generated video | 7 clips x 5 s x 6 credits = 210 credits | 7 clips x 5 s x 12 credits = 420 credits |
| Share of a Runway Standard month ($15, 625 credits) | 34 per cent, or $5.04 | 67 per cent, or $10.08 |
| Voiceover (stock voice, three takes) | about $0.40 | about $0.40 |
| Editor and captions | $0 with a free editor | $0 with a free editor |
| Tool cost for the video | about $5.44 | about $10.48 |
| Your time, 80 minutes at a notional $40 an hour | $53.33 | $53.33 |
| Real cost of the video | about $58.77 | about $63.81 |
The tools are only 9 to 16 per cent of the real cost. The 80 minutes break down as 10 for picking and ordering photos, 25 for prompts and renders (mostly waiting), 10 for the frame check, 10 for the script and voice, 20 for editing, captions and music, and 5 for the label, the seller's approval and posting. A plain pan-and-zoom version of the same video skips the renders and most of the frame check and takes about an hour, for no credits at all.
Those figures are the share of a plan that one video uses. What you actually pay depends on how many videos share the plan, because credits are spent per clip and the monthly price is fixed. Using the same 210 and 420 credit videos:
| Videos a month | Cheaper model path | Higher-quality path |
|---|---|---|
| 1 | Standard, $15 for the month | Standard, $15 (420 of 625 credits) |
| 2 | Standard, $7.50 a video (420 of 625 credits) | Pro, $17.50 a video (840 of 2,250 credits) |
| 5 | Pro, $7 a video (1,050 of 2,250 credits) | Pro, $7 a video (2,100 of 2,250 credits) |
| 10 | Pro, $3.50 a video (2,100 of 2,250 credits) | Max at $95, $9.50 a video (4,200 of 9,500 credits) |
Hidden costs are mostly re-rolls and revisions. If the seller asks for a different order, a clip you thought was finished is not, and a second round of generation spends credits again. Keep the final assembly in a project file in the free editor so a reorder costs minutes, not credits. For a wider comparison with paying a crew, see how much an AI marketing video costs compared with a shoot.
Get the seller's written approval first
A listing video says something about the seller's house in the seller's name, so the seller should see the cut before it posts. The message below also puts on record what the AI did and did not do, which is worth having if a buyer ever disputes a detail.
Subject: Please approve the video of [address line] before it goes live
Hi [first name],
I have made a 42-second video from the professional photos of your home.
Four of the nine clips use an AI tool to add camera movement to a still
photo; the others are plain pans and zooms. I checked every AI clip
frame by frame against the photo, and no rooms, fittings or views were
changed and no furniture was added or removed.
The video carries a line saying some motion is AI-generated, and the
original photos stay in the listing gallery.
Please reply "approved", or tell me which clip to change. Nothing goes
live until I hear from you.
[your name]
Check one more thing before sending: personal items. Family photographs, post addressed to the occupants and children's rooms are easy to miss in a still and impossible to remove from a generated clip without altering it. If a photo has them, reshoot or choose another.
Questions to put to a listing-video app before paying
Dedicated apps that promise a listing video from a photo set are a crowded category, and I cannot vouch for any of them without testing your photos. Put these questions to the vendor, and test one real set of photos before paying:
- Is the price per video, per minute or per month, and what happens when credits run out mid-video?
- Can I switch off generated motion and keep plain pan and zoom?
- Does it add people, furniture, weather, clouds or plants by default?
- Can I export without a watermark, at the sizes I need for vertical and horizontal posts?
- Whose licence covers the music and the voices, and can I use them in paid ads?
- Does it add an AI label or metadata, and can I add my own wording?
- Where do my uploaded photos go, how long are they kept, and are they used to train models?
- Can I export the project so a reorder does not mean paying again?
Then run the frame check on the output. An app that cannot pass your own audit on four photos will not pass it on twenty, and the fixed-fee plan only looks cheap until you count the clips you reject.
Check the label after posting, then review at 14 days
Posting is not the last step. Open the published video on a phone, with sound off, and confirm that the label shows where the platform puts it, that the caption line reads correctly and that the first frame is the plain one you chose. Then ask the seller to watch it once on their own phone.
After 14 days, compare the video with a similar listing that had only photos, looking at saves, shares, viewing requests and enquiries that mention the video. One listing is an anecdote, not a result, so decide after four or five listings whether the extra hour is paying back. If it is, keep the shot list and the prompts as templates for the next listing; if it is not, drop the generated clips and keep the plain version, which is the cheapest part of the process anyway.
Questions agents ask before posting an AI listing video
Does a pan and zoom over my photos need an AI label?
No. A virtual camera move over an unchanged photo is ordinary editing, and the platform rules I read target realistic content that AI generated or altered. The line is crossed when a model invents frames, adds or removes objects, or narrates in a voice that sounds like a real person. In a mixed video, label the whole video, because one generated clip is enough to need it.
Can the AI add furniture or a bluer sky to the video?
It can, but that changes what is for sale. Virtual staging needs its own label on each altered image with the original shown beside it, and some listing services bar AI-generated landscaping altogether. A sky swap on a grey day also misleads buyers about light. Use staging only with the seller's approval, and label every altered frame.
Do I need the photographer's permission to animate their photos?
Check the licence in writing before you upload. Some licences cover marketing use but not edited or AI-processed versions, and some tools keep or reuse uploads. A one-line email asking whether motion edits and third-party AI tools are covered settles it, and you keep the reply with the listing file.
What if my listing service says nothing about video?
Follow the photo rule: say the motion is AI-generated, keep every original photo in the gallery, and email the service for a written answer. Keep the reply with the listing file. If the service later rules that videos count as altered images, you have already done what it asks.
Further reads
- AI Real Estate Social Media: A Month of Posts From Your Listings — Where the finished video goes in a month of posts.
- AI Listing Mistakes That Can Mislead Buyers, and How to Check — More ways AI listing material misleads buyers, and how to check.
- How to Keep AI-Made Videos On-Brand: Captions, Voice and Intros — Keep every video on-brand with captions, voice and intros.
- Best AI for Real Estate Marketing: Ads, Email and Listing Video — The wider toolkit for ads, email and listing video.
- Free AI Video Generator for Small Business Ads: Limits to Check First — Free generator limits to test before buying credits.
- Can a Travel Agency Use AI Video? Destination Clips and Disclosure — The same disclosure question for clips of real places.
- AI Real Estate Listing Presentation: Build the Seller Deck — Win the listing before the video is ever shot.
- Free AI Home Staging: Limits, Watermarks and When to Pay — What six free AI staging tools allow, where watermarks and licence terms bite, a ten-minute quality test, and the point at which a stager should pay.
- Home Stager AI: Adding Paid Virtual Staging to Your Service Menu — Four packages, a price built from tool cost and your own minutes, a hybrid physical-plus-virtual offer, and the disclosure wording to use.
- AI Marketing for Real Estate Agents: A Four-Week Content Plan — Week by week: a listing launch, an area guide, a market update from your own figures and a proof week, each turned into posts, a script and an email.
- Staging Real Estate Photos With AI: A Step-by-Step Workflow — A seven-step workflow from raw listing photos to labelled, staged images, with a costed example and a distortion checklist.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Runway pricing page and API model guide, Gemini video generation page, Canva pricing page, ElevenLabs pricing page, YouTube Help (disclosing use of GenAI content), TikTok Support (AI-generated content), Meta newsroom (AI labelling), the EU AI Act Article 50 text, and one large listing service's 2026 rules on digitally altered images.