GPT-6 Astra generates a full real-estate video tour from a single Zillow listing

venturetwins · x · 2026-10-25

A user fed GPT-6 Astra a real Zillow listing and asked for a 30-second vertical video walkthrough hosted by a real estate agent. The model autonomously picked reference images (8 listing photos), wrote a timestamped prompt, and synced the audio narration to the right moments of the tour.

The result was strikingly accurate — even reproducing odd details like a weird dining-room lighting fixture. The demo highlights how far multimodal models have come at combining real-world data with coherent video generation.

Original post →

More from Multimodal

Multimodal channel →