- Site photos fail because they lose location, context, and history.
- Engineers need to track condition and remediation over time, not rediscover issues.
- This workflow anchors photos spatially inside Revit using EXIF GPS and constrained AI descriptions.
- The AI describes only what’s visible — no inference, diagnosis, or assessment.
- Accuracy isn’t survey‑grade; consistency is what enables comparison over time.
- The real shift isn’t technology. It’s taking photos with intent.
(click to expand the tl;dr)
Every project generates thousands of photos. They’re taken in good faith: someone walks the site, notices something worth recording, pulls out a phone, and clicks. The assumption is that the job is done.
It rarely is.
Six months later, those images are buried in folders, email threads, or a CDE. The person who took them remembers what they were looking at. Everyone else sees a file without context. The photo still exists, but the memory around it is gone.
On the project that triggered this work, engineers weren’t asking for AI or live dashboards. They were asking three straightforward questions:
- Is this defect new?
- Is it progressing?
- Has it already been remediated?
The photos existed. The answers didn’t.
Below is a recording of my pre-presentation test run for my BrisBIM presentation “A Practical Workflow for Turning Site Photos Into Usable Model Data” in which I describe the high level overview of the process.
The Code
As promised during the session, I’ve shared a cut‑down version of the proof‑of‑concept code below.
- Python extracts EXIF metadata and translates coordinates into the project system.
- Dynamo places the Revit markers and links the images and descriptions.
A production implementation would use the Revit API add‑in or an either Revit’s extensible storage, or an external database with blob storage, but the underlying logic is unchanged.
No Comments