Skip to main content
HapHazardZA
New Participant
March 15, 2026

Model Everything, I Guess?

  • March 15, 2026
  • 1 reply
  • 94 views

Alright, Meta AI, hear me out.

A little background: I'm a small-scale web developer, videographer, and photographer with about 10 years of experience. Over the past year, I took something of a gap year to dive into AI and research the technology.

This led me to a project that's in line with my expertise: real estate and architectural photography, which is basically about capturing spaces. My idea is to train a model using a combination of Instagram & Facebook images (with their tagged locations, of course) and 3D terrain models from multi-date satellite images (In the future: add property listing images).

By combining these, we could create a virtual, living, breathing digital twin of the entire real world. This model would map both the outside world and most public indoor spaces, likely requiring a staggered rollout.

I know this would require a metric ton of servers. I'm currently tinkering with a small-scale prototype, but I don't have the resources to scale this monster. The biggest limiting factor is being from South Africa (born in '98). Our government isn't exactly tech-friendly; instead of encouraging new tech, they create roadblocks and regulations that stifle innovation, all to protect corrupt, archaic monopolies. It's infuriating.

Meta already has all the data needed for this model. So, if anyone at Meta AI sees this, please build this thing. And hey, maybe hook a man up with a tiny royalty? 😉

P.S. I would love to build this with you. Seriously, hit me up.

Kind Regards,

T South Africa

    1 reply

    Degly
    Degly
    Inspiring
    May 10, 2026

    This is actually very aligned with where spatial computing research is heading.

    The hard part is probably not reconstruction anymore, but scale, freshness, privacy, ownership rights, and synchronization of the data.

    A “living world model” built from images + geo/spatial data could have huge applications in AR navigation, real estate, robotics, digital tourism, and persistent MR experiences.

    You should look into Gaussian Splatting, NeRFs, VPS/localization systems, and large-scale scene understanding research. Meta, Google, NVIDIA, and others are already solving different parts of this problem independently.

    Terms & ConditionsAccessibility statement