Producing listing videos at the scale of realestate.com.au was commercially impossible through traditional production methods.
REA needed to validate whether agentic AI could generate professional, multi-scene property videos from existing listing photographs before committing to a full production rollout. Could this new technology deliver acceptable quality, manageable cost, and with a workable automated moderation approach?
Kablamo ran a focused 6-week time-boxed POC using the Lightning AI model: a small, senior team of 3 engineers co-building directly in REA's sandbox GCP environment. Built on Kablamo's AMG (Agentic Media Generator) accelerator, the POC delivered a headless REST API to trigger agentic video generation and a moderation API, with agentic quality loops automating parts of the moderation process.
The team co-built alongside REA engineers from day one, with deliberate knowledge transfer built in so REA could take ownership and continue development independently after the 6 weeks. A structured quality evaluation framework was used throughout, giving REA clear metrics to assess the output themselves. MCP tools were incorporated to allow flexible model and tool swapping as the underlying technology evolves.
REA came into this engagement with a clear question: can agentic AI generate professional property listing videos from photographs, at quality and cost that makes production rollout viable? Six weeks later, they had their answer — and they owned it.
The POC landed in REA's hands as a fully working system in their own GCP sandbox, with source code in their own Git repository and engineers who'd co-built it throughout. Video quality and accuracy validated. Per-video generation costs quantified. The moderation approach — agentic automation via a Judge LLM, with a final human publish approval — tested and proven. What Kablamo delivered in six weeks was the foundation REA needed to move forward with confidence.
