A Real Estate Agency Used AI to "Furnish" Empty Apartments — Viewing Conversion Doubled
Lao Zhou runs a real estate agency in Chengdu with about 20 people, focused on second-hand sales and rentals. By late 2025, he hit a very practical problem — empty apartments were getting harder to rent out.
The reason was simple: tenants browse listing photos first. An empty apartment or bare-bones room photographs as four white walls. Nobody clicks.
The traditional solution was physical staging — spend 2,000-3,000 RMB to rent furniture, set it up, shoot photos, then return everything. From taking on a listing to going live: at least one week.
How AI Got Involved
Zhou tried a three-step approach:
Step one: AI virtual staging. Take a wide-angle photo of an empty room, use Midjourney's inpaint feature to "paint in" furniture. Sofa, bed, dining table, curtains — a complete Nordic-style setup, 10 minutes per image.
Step two: AI-generated listing descriptions. Feed floor plan data, neighborhood info, and location data to ChatGPT, which produces emotionally engaging listing copy. No more "3-bed, 2-bath, south-facing, convenient transport" boilerplate.
Step three: AI-generated short video scripts. Select the best AI-staged images, pair them with AI-written voiceover scripts, edit into 15-second clips for Douyin and Xiaohongshu.
Total cost: Midjourney $30/month + ChatGPT $20/month. Under 400 RMB. No designer needed, no furniture rental.
First Pitfall: AI Furniture Proportions Were Wrong
The very first batch of images was a disaster.
The AI-staged rooms looked beautiful at first glance, but on closer inspection — the sofa was half the height of the window, the dining table was squeezed into a corner like a coffee table. Proportions completely off.
The problem was the prompt. The initial prompt just said "Nordic-style furniture" with no proportion reference. AI doesn't know if the room has 3-meter or 2.6-meter ceilings — it has to guess.
The fix: place a standard reference object in the photo — a regular 40cm-high dining chair. Then specify in the prompt: "The reference object is a 40cm-high chair, please proportion furniture accordingly." Proportion issues mostly resolved, though occasional manual tweaks are still needed.
Second Pitfall: Tenants Spotted the AI Copy
After a month, a tenant commented: "This listing description is obviously AI-written, it feels so fake."
Zhou went back and reviewed recent listings — sure enough, they all read like "sunlight fills the living room" and "a warm, cozy haven." Every listing sounded the same.
The issue wasn't AI capability — it was lazy prompting. The revised strategy: have AI first analyze each apartment's pros and cons (near subway → "commute-friendly," poor natural light → "quiet and private" — not lying, just reframing), then write descriptions based on those points. The copy suddenly had personality. No one called it AI-generated again.
Results
After three months:
- Listing go-live time dropped from 5-7 days to half a day
- Viewing booking conversion rate went from 11% to 23% — doubled
- Douyin listing videos averaged 3,000+ views, far exceeding previous real photos
- Per-listing staging cost dropped from 2,000-3,000 RMB to roughly 50 RMB (API + labor)
Zhou did the math: 20 people handle about 60 listings per month. Old staging costs would run roughly 1.4 million RMB per year (60 listings × 2,500 RMB × 12 months × 80% vacancy rate). Now, under 40,000 RMB annually. The savings alone could fund two more hires.
Is This Right for You?
If your business requires showcasing products visually when physical costs are too high — real estate, furniture, home renovation, auto, jewelry — AI virtual solutions are a viable path.
But remember: AI images "look like" not "are like" the real thing. If you're in the high-end market where clients are detail-sensitive, AI images may backfire. Best ROI is in the mid-to-lower market.
Also, some platforms are already checking for AI-generated listing photos and require "render for reference only" disclaimers. Don't try to sneak past — just label it properly.