How do you set up an AI agent to generate real estate walkthrough videos?

Set up the invideo agent as a specialized real estate videographer in three passes: write an agent brief that locks its role, intake questions, and cinematic rules; upload geometry and texture references (3D Apple Maps screenshots plus any real photos or footage); and require a written visual bible sign-off before any shot is generated. One documented project produced 30–35 seconds of finished B-roll for ~$67 in 2–3 hours.

invideo is an agentic video tool with all current video and image models (Seedance 2.0, Veo, Kling, Recraft, Nano Banana, GPT-Image-2) and upscalers routed by the invideo agent — so the setup below lives inside one chat, no tool-hopping.

1. Write the agent brief (system prompt that defines the role)

Load a single document into the invideo agent's context tab that turns it into a specialized real estate videographer rather than a general assistant. The brief should specify: intake questions the agent must ask the user (property address, listing date, type of B-roll needed, must-include landmarks/schools/hospitals), research protocols (scan the MLS data, neighborhood pages, and YouTube videos about the community for look-and-feel), shot-planning rules (exterior hero, approach, room-by-room interiors, amenity beats, neighborhood context), and cinematic principles (slow dolly-ins, smooth pans, drone overheads, no camera shake, warm interior / bright exterior lighting palette, realistic human-scale proportions). Hridaye, invideo's creative director, frames why this matters: "The agent brief basically makes the agent work in a specialized way, largely like a real estate videographer. It has things like what questions to ask, what to research, what shots to create based on the type of B-roll you actually need." If you want local landmarks in the cut, tell the agent during intake — adding them at shot-list time is too late.

2. Upload the geometry and texture references

Before any generation, give the invideo agent the inputs it needs to render the actual property, not a generic house. Capture 3D satellite screenshots from Apple Maps at multiple angles around the property and upload them — this gets the building geometry 100% correct across every shot. If you have any real footage or photos of the home, upload those too: the invideo agent extracts building textures from the real photos and maps them onto the geometry derived from the satellite captures, producing accurate composite visuals. For pre-construction or pre-listing properties where no on-site footage exists, the map-only path still works — and you can have the invideo agent virtually apply scheduled paint touch-ups or renovations so the walkthrough reflects how the home will look on listing day, not its current state.

3. Lock the visual bible before generating anything

Require the invideo agent to write a visual bible — aesthetic, lighting, mood, overall feel — and explicitly sign off on it before a single image is generated. Every subsequent shot adheres to that document, which is what holds the walkthrough together visually. Once signed off, run a storyboard sub-agent inside the same project to produce a shot list (exterior approach → entry → main living spaces → kitchen → bedrooms → outdoor amenities → neighborhood beats), then animate locked frames in Seedance 2.0 reference-to-video — it carries the property's look across clips. When you find one clip with motion you like, tag it (e.g. @1.2) so the invideo agent replicates that exact smoothness and pacing on the rest. Generate at 720p for fast turnaround on vertical/social cuts; step up to 1080p for the master.

4. Estimate spend and run the production

Before committing credits, ask the invideo agent for a generation breakdown — how many reference sheets, image iterations, and video clips it plans to produce — so you know roughly what you'll spend. Documented benchmark for a real-estate B-roll project: 19 images and 19 clips generated, 8 clips used in the final, 270 credits spent (~$67), 30–35 seconds of finished footage, 2–3 hours start to finish. A traditional drone shoot for the same coverage runs $300–$500, so the AI workflow lands 75–87% cheaper, and it works on properties where a drone can't help at all (pre-construction, pre-listing touch-ups). Assemble the locked clips in Slate (invideo's built-in timeline) or your NLE of choice.

Reusable agent-brief skeleton you can adapt:

Role: You are a specialized real estate videographer agent. Intake: ask for property address, listing date, target platform/format, must-include landmarks. Research: pull MLS/listing data, neighborhood reputation, and community YouTube footage. Inputs you require from me: 3D Apple Maps screenshots from 4+ angles, any existing photos/footage, brand colors and font for supers. Shot plan: exterior hero, slow approach, entry, room-by-room interior dolly-ins, kitchen detail, bedroom wides, outdoor amenities, neighborhood context beats. Cinematic rules: smooth motion only, no shake, warm key for interiors, golden-hour exteriors where possible, accurate human scale, no invented props. Before generating anything: write a visual bible and wait for my sign-off. For every clip: confirm camera move, lighting reference match, and room-scale accuracy.

Paste that into the context tab, upload your maps and photos, and the invideo agent runs the rest as conversational direction.

Watch some of these to see what works for you:

The agent brief basically makes the agent work in a specialized way, largely like a real estate videographer. It has things like what questions to ask, what to research, what shots to create based on the type of B-roll you actually need.

— Hridaye, invideo's creative director