The Fouquet Case Study — Trailer

Lanre Adebayo

I build video from still images.

Generative video systems, and the films they produce. Toronto.

Real estate

Listing and property video for agents at Sotheby's and Compass.

Sotheby's International Realty7,572 views
Pump Hill, Calgary — $3,950,000
A rebuilt mid-century home on a rare quarter-acre cul-de-sac lot. Built entirely from the listing's photography, with exteriors virtually staged across seasons. No shoot.
16:9 native
Pump Hill, Calgary — $3,950,000
The same listing, generated natively in widescreen rather than cropped from vertical. Both formats come out of the same source photography, each composed for its own frame.
Sotheby's International Realty2,280 views
Le Germain Residences #1801, Calgary — $1,900,000
An 18th-floor sub-penthouse with a 650 sq. ft. terrace over downtown. Panoramic interiors cut from stills.
Compass$2,649,000
2400 Crest View Drive, Laurel Canyon
A 1964 mid-century retreat on a private cul-de-sac above Los Angeles. Unobstructed views from the skyline to the canyons to the Pacific, and 1,629 sq. ft. built around the light.
Compass$1,525,000
20 Teatown Road, Croton-on-Hudson, NY
A 1929 Cape Cod estate on four and a half acres bordering Teatown Lake Reservation. Private pond, cathedral ceilings, Metro North minutes away.
Walkthrough from scanCanmore, AB — $1,110,000
601 – 102 Stewart Creek Rise, Three Sisters
A 26-second pass through the living room of a south-facing end-unit townhome in the Rockies, generated from a Matterport capture. The camera moves through the unit's real geometry — nothing invented, nothing approximated.
Stat drop / Reel
Hype Reel
A punchy, high-energy stat drop cut to three upbeat audio options. This piece skips generative models entirely: the data and motion graphics are rendered straight from code using Hyperframes.
Hybrid Workflow
Editorial
An editorial deep-dive into the property’s best features. Generative models handle the atmosphere, while Hyperframes delivers crisp, code-driven motion graphics for the hard facts.

Accuracy

Generated video invents things. A window that isn't there, a hallway that doesn't connect. That's the reason AI video hasn't been usable for real listings — a walkthrough that shows a room the buyer won't find is a misrepresentation problem, not a style problem.

So accuracy scales with the source. Listing photos give you a cinematic piece with artistic transitions. Add a floor plan and the space becomes spatially accurate. Start from a Matterport or iGuide scan and the camera moves through the property's actual geometry — every room, every angle, nothing invented.

That constraint is the product. It's what makes the output usable in a listing appointment rather than just a feed.

Systems

I come from software. Every film on this page came out of infrastructure I built.

The pipeline. Python and GCP — Cloud Run, Cloud Tasks, Cloud Storage — carrying a brief from script ingest through to a finished, graded film. Modular model integrations for Veo, Runway, Gemini, Midjourney, Flux, ElevenLabs, and Suno, so a model can be swapped without touching the rest of the system.

The agent loop. The interesting part isn't automation, it's iteration. A shot goes through dozens of prompt refinements to find the one that holds up, instead of settling for the third generation because the fourth is tedious. Volume comes from iteration being cheap, not from decisions being automated.

Determinism. Some pieces skip generative models entirely — data and motion graphics rendered straight from code. Others constrain generation against real geometry from a 3D scan. Knowing when not to generate is most of the craft.

Before this. Three years building AI systems for clients: semantic search over 23,000+ scraped and normalized profiles replacing a legacy keyword matcher; a multimodal analytics engine combining image, speech-to-text, and prosody detection for real-time assessment; video ingestion infrastructure for a recruitment platform.

Original series

Spisce — short-form history and strategy.

560,305 views
48 Laws of Power: Conceal Your Intentions
Ethiopia, 1927. Ras Tafari dismantles a 10,000-man army without firing a shot — by feeding its commander while quietly buying his soldiers away with gold.
285,703 views
The Rothschild Family, Ep. 6
1812. James Rothschild arrives in Paris at twenty, into a France where Napoleon's empire is collapsing and governments rise and fall. He makes himself necessary to all of them, then finances the railway that carries the country.
2,146 views
How Edison Monopolized Electricity
February 1879. Edison builds a working lightbulb, then builds the grid underneath it — generators, meters, power stations — and sells dependency rather than light. When Tesla's alternating current threatens the model, he wages a public campaign against it and loses.

Across the channel: 1.55M organic views and 8,000 followers in four months, no paid distribution. ~80% average view duration, with 97.5% of watch time from non-followers.

How I work

Source material

Most clients already have the assets — listing photos, architectural renders, 3D scans, archive images. I build motion out of what exists rather than scheduling a shoot.

Production

Every piece is AI-produced end to end: script, image generation and treatment, motion, voice, music, edit. Runway, Veo, Gemini, Midjourney, ElevenLabs, Suno, finished in CapCut Pro.

Volume

100+ finished pieces since January, and the pace picked up once I moved into real estate, where the source material makes the output far more deterministic. One brief in, one finished film out, same day.

What gets made next

I read retention curves, not vanity metrics. Where people drop off decides the next piece.

Contact

lanre@lanreadebayo.com

Lanre Adebayo — Toronto