AI video for real estate agents: a per-listing workflow in credits

The AI-real-estate SERP is a tour of two things that are not what an agent needs. Half the links point at avatar-presenter talking-head tools where the agent clones themselves and reads a script to camera. The other half point at slideshow builders where you drop the MLS photos in and the software adds Ken Burns pans, upbeat music and captions. Neither one answers the question an agent is actually asking in 2026: what if the listing is not a slideshow and I do not want to film the walk-through myself.

This is the five-shot workflow we use for a per-listing AI video, which engine renders each shot, and what the credit math works out to on a 30-listings-a-year cadence.

The frame the SERP misses

A real estate buyer scrolls past the slideshow because they have seen every slideshow. The thing that stops them is a story: an agent walking a buyer through the one feature of the house that nobody else is listing on. A kitchen islanded for a family of five. A garage pre-wired for an EV. A sunroom that holds the summer sun until nine. That story needs a character — the agent — and it needs the character to hold across four or five short scenes, not sixty still images with a voiceover.

A character that holds across short scenes is a job for a reference-sheet workflow, which is the thing avatar tools don't give you (they clone one angle to camera) and slideshow tools don't give you (they don't do character at all). It is the thing the five engines in the studio were built for.

The five shots, in order

Each shot is 6–8 seconds. The whole listing is a 35-second vertical cut that drops on Reels, TikTok and the listing page. 35 seconds is long enough to tell the story, short enough that the pacing doesn't stall.

Shot 1 — the agent at the door, 6s voiced. The agent waves the viewer in, says one line: "Four bedrooms, east-facing, back on the market after a price drop." Delivered as a voiced clip because the first six seconds have to carry speech. Rendered on Veo 3.1 Fast at 720p because Veo takes the agent's reference image, holds the face clean across the eight-second cap, and bills a 6-second voiced render at 180 credits.

Shot 2 — the hero room, 8s silent. The thing the listing is really about. The kitchen island, the sunroom, the pool-of-light moment. No voiceover — the voiceover rides on top later. Rendered on Seedance 2.5 at 720p silent (motion-heavy, long take) for 560 credits.

Shot 3 — the agent inside the room, 6s voiced. The agent walks into the frame of shot 2 and delivers the one specific line about why this room matters. Rendered on Kling O3 Pro at 720p voiced because this is a character-holding shot and Kling O3 Pro is the cheapest voiced second we sell at 30 cr/s voiced — 180 credits.

Shot 4 — the feature pan, 8s silent. A pan across one more feature of the house: the primary bathroom, the yard, the view off the balcony. B-roll, silent, long enough to breathe. Seedance 2.0 at 720p silent is the economy lane for this: 45 cr/s × 8 = 360 credits.

Shot 5 — the agent closing, 6s voiced. The CTA: "Showing Saturday at two." Veo 3.1 Fast at 720p voiced again, 180 credits.

Total: five shots, 34 seconds, 1,460 credits.

The credit math per listing

shot	engine	seconds	voiced	credits
1	Veo 3.1 Fast	6	yes	180
2	Seedance 2.5	8	silent	560
3	Kling O3 Pro	6	yes	180
4	Seedance 2.0	8	silent	360
5	Veo 3.1 Fast	6	yes	180
			total	1,460

On the Creator plan (6,000 credits per month at $45) that is one of four listings a month: $11.25 per listing on the AI render alone. On Studio annual at 0.55c per credit, the same listing drops to $8.03. On Lite at 1.0c, $14.60.

A listing videographer is $300–$1,500 per property on the SERP I read for this piece. The AI render lane is 20× to 100× cheaper per listing, which is only interesting if the output is actually usable — which is the honest part of this whole workflow: the AI shot is not a videographer. It is a thirty-five-second scroll-stopper on social. The videographer is still the right call for the flagship listing where a buyer is making a seven-figure decision from a tour.

For the other twenty-nine listings a year, this is what cheap enough to default to actually looks like.

The character work the slideshow tools skip

The reason the five shots hang together is not any one engine — it is that the agent is the same person in shots 1, 3 and 5. If you render each shot cold, the face slides. The workflow:

Generate the agent as a reference sheet, once. One image, nine panels: large frontal face at the centre, front/three-quarter/side/back body views around it, four facial expressions along the bottom. The sheet is a 20-credit render on the studio's character tool and it is the only character step you do per agent. Save it. It carries across every listing you will ever shoot.

Write the identity prompt once. Build, hair, eye colour, skin tone, exact outfit (blazer, blouse, slacks or dress), accessories, posture. One paragraph. Paste it into shots 1, 3 and 5 unchanged.

Hand the reference sheet to each engine. Veo takes three images (one slot of the sheet goes in). Kling takes five elements (face, blazer, briefcase, backdrop, prop). Seedance takes up to thirty — overkill for a solo agent but useful on a two-agent listing where the second agent gets their own slot.

First and last frame on the hero shot. Shot 2 is a slow pan; shot 3 is the agent walking into the pan. The last frame of shot 2 is the first frame of shot 3. That continuity is what makes the two shots feel like one scene rather than a cut, and it is the one piece of plumbing the slideshow tools cannot do because they do not render anything — they stitch stills.

The monthly cadence

Four listings a month on the Creator plan at 6,000 credits lands exactly on the credit budget with 160 credits to spare — enough for one 4-second Seedance 2.5 draft to test a shot before the final render.

Twelve listings a month on Agency at 37,500 credits for $299 — the per-listing render cost drops to $11.65 and leaves 20,000 credits a month for everything else the agent does on social (market-update reels, feature announcements, saved-search replies).

The one place the math breaks is on a listing that needs more than five shots. Six-plus-shot listings run into Kling's shot cap: Kling O3 Pro caps a single generation at 15 seconds, so a 20-second character shot is two renders joined with the same elements in both. The joiner holds because the elements are shared, but it does spend two renders. If your listing story has six character beats, send the longer two to Seedance 2.5 (one 30-second take at 70 cr/s silent is 2,100 credits) rather than two Kling joins.

What to do before you render anything

Open a free account and burn the first 300 credits on three test renders of your own face as the agent: one on Veo at 6 seconds voiced (180 credits), one on Kling at 4 seconds voiced (120 credits — overrun the 300 by burning a 4s Kling instead), one on Seedance 2.5 at 4 seconds silent (280 credits). Compare the three faces to the reference sheet. Pick the engine that holds your face best for the voiced lines. That is the engine shots 1, 3 and 5 go on for the next thirty listings.

Everything else is subscription math.