A clever trick, hidden behind a lot of setup
Modern AI vision tools can rebuild an object or a scene in 3D from a set of ordinary photos. The catch is that they need a powerful graphics computer, a fiddly setup and some patience to run. That puts them out of reach for most people.
I wanted that ability behind a simple upload page: drop in some photos, get back a 3D model. So today I built one. It's my own project, and it's in development.
How it works, in plain terms
- Upload your photos. Between 2 and 50 of them, dragged straight onto the page. They go directly into storage, so a big set of photos doesn't have to squeeze through the website itself.
- The heavy lifting happens elsewhere. A powerful graphics computer picks up the job and works out where each photo was taken from and the 3D shape they show together. Each job is capped at 30 minutes.
- Watch it progress. The page shows each stage live: queued, downloading, reconstructing, exporting, uploading and done.
- Spin it around. When it's finished, the model opens right on the page, ready to rotate and zoom. You get it as a standard 3D file that also opens in common 3D tools.
No surprise bills
Powerful graphics computers are expensive to leave running. So this one starts when a job arrives and switches off completely when it's idle. I also deliberately left out automatic retries, so a failed job can't quietly rack up computing time in the background. There's no always-on machine costing money while nobody's using it.
There's also a settings page that doubles as a health check. It tests every part of the setup in one place and shows exactly what's missing if something isn't connected.
Built to get better
The part that does the 3D reconstruction is one swappable piece. When newer, better versions of the underlying model are published, they can drop in without rebuilding the rest of the service.
Where it could fit
On its own, it's a neat trick. Plugged into another website or app, it gets useful. The flow is simple: upload photos, create a job, watch its status, show the model. Any web app can do that using the same service. Property, product or site-capture tools are the obvious places it could fit.
A tip if you try something like this: more photos from different angles, with plenty of overlap between them, give a more complete model than a few shots from the same spot.
Being honest about where it's at
It's in development. When I show it off, I use a sample object rather than a live result. I'm also not quoting running costs per job until I've measured them properly.
Heavy AI doesn't have to mean a heavy bill, or a complicated screen.
The full write-up is in the photos-to-3D case study. If you have an idea that needs serious computing behind a simple button, tell me about it, or see more about AI assistants and automation.