Measured, not guessed.
Voxeliz turns ordinary photographs into 3D models you can measure, label and analyse, with a clear record of what was observed and what was inferred. In your browser, or through our API.
How it works
- 1
Capture
Upload overlapping photos from any camera.
- 2
Reconstruct
Real multi-view geometry builds a 3D model from what the photos actually show.
- 3
Check and measure
Every model comes back Good, Fair or Poor, with the evidence behind the grade. Measure distances to scale against a known reference.
- 4
Understand
Label parts automatically or by hand, flag defects and anomalies, and compare against a design reference or an earlier capture. Every finding shows its evidence.
- Photos
- 0
- Points
- 0
- Triangles
- 0
- Images used: How many of the uploaded images could be placed in the reconstruction.
- 29 of 30
- Coverage: How much of the subject the images actually saw.
- 97%
- Fit error: Average reprojection error (RMSE) in pixels. Lower is better.
- 0.4 px
Findings
- 3 parts labelled
- Compared with design reference
- 1 anomaly flagged
1 / 4
Capture
Upload overlapping photos from any camera.
Built for your field
One reconstruction core for every field. Pick a field and a use case to see what Voxeliz gives you.
Parts, hulls and products checked against their design, with deviations and defects flagged.
A cast gearbox housing becomes a measured model: features labelled, the bore measured, and a surface crack on the housing flagged.
Illustrative images
Manufacturing · Machined parts
Geometry, not just a picture
Photograph the part from enough angles and Voxeliz checks its geometry against the design, not just how it looks.
Where it helps
First-article inspection
Compare the first part off a new line with its CAD model and catch deviations before the batch runs.
Defects 2D cameras miss
Find cracks, warping and misaligned features from the 3D shape, each with its location and confidence.
Parts with no drawing
Capture an old spare, get its features labelled and dimensions measured, and give engineering an STL to start from.
A camera sees the surface. Voxeliz checks the shape.
- You capture
- Overlapping photos from a phone, a camera or a station on the line
- You get
- A measured model with surfaces, edges and features labelled
- It finds
- Deviations from CAD and defects, each with location and confidence
- You receive
- A report, plus GLB, OBJ, STL or PLY for your QA system
An inspector photographs the housing on the bench from many angles. Voxeliz reconstructs it, identifies its surfaces, edges and features (the bearing bore, the mounting flange, the cooling fins) and aligns the model with the CAD reference. Dimensions such as the bore diameter are measured to scale, deviations are measured across every surface, and the surface crack on the housing is flagged with its position and confidence.
Regions the photos covered poorly are marked low-confidence, so nobody mistakes a gap in the data for a defect, or a defect for a gap. The result is a report with the model, a deviation map and every finding, in the browser or through the API, ready for the system that already tracks quality.
Know what's real
Every model shows how much of the subject your photos actually covered, with a quality rating and the evidence behind it. Anything the system had to estimate is marked as inferred, never presented as fact.
- Images used: How many of the uploaded images could be placed in the reconstruction.
- 11 of 12
- Coverage: How much of the subject the images actually saw.
- 61%
- Fit error: Average reprojection error (RMSE) in pixels. Lower is better.
- 0.9 px
Illustrative values for this demo.
The front is solid. The back was never captured, so it is marked inferred, never shown as measured.
Can you measure it?
Photos show the shape, not the size. Give Voxeliz one real length and the numbers become real.
1.42 units
Shape only. This number is not in cm or inches yet.
Most 3D tools make a shape. They don't tell you what to trust.
Generative tools guess the parts of an object they can't see, and they don't tell you which parts those are. That's fine for a game asset. It isn't fine when someone is making a decision from the model.
| Compared on | Generative 3D | Desktop photogrammetry | Voxeliz |
|---|---|---|---|
| Input | One photo or a text prompt | Many overlapping photos | Many overlapping photos of the real thing |
| Hidden side | Invented so it looks complete | Left open | Marked as inferred, with a confidence level |
| Numbers | Look like dimensions | Measured, once scaled | Calibrated only from a real scale, and labelled |
| Understanding | None | Done by hand, by an expert | Labelled parts and flagged defects, with evidence |
| Quality | Not stated | Reports for an expert to interpret | Good, Fair, or Poor, with the evidence |
| Where it runs | A web app | Installed software, run by a specialist | Your browser, or your software through an API |
| Speed | Seconds | Minutes to hours | Minutes, because it measures |
Input
- Generative 3D
- One photo or a text prompt
- Desktop photogrammetry
- Many overlapping photos
- Voxeliz
- Many overlapping photos of the real thing
Hidden side
- Generative 3D
- Invented so it looks complete
- Desktop photogrammetry
- Left open
- Voxeliz
- Marked as inferred, with a confidence level
Numbers
- Generative 3D
- Look like dimensions
- Desktop photogrammetry
- Measured, once scaled
- Voxeliz
- Calibrated only from a real scale, and labelled
Understanding
- Generative 3D
- None
- Desktop photogrammetry
- Done by hand, by an expert
- Voxeliz
- Labelled parts and flagged defects, with evidence
Quality
- Generative 3D
- Not stated
- Desktop photogrammetry
- Reports for an expert to interpret
- Voxeliz
- Good, Fair, or Poor, with the evidence
Where it runs
- Generative 3D
- A web app
- Desktop photogrammetry
- Installed software, run by a specialist
- Voxeliz
- Your browser, or your software through an API
Speed
- Generative 3D
- Seconds
- Desktop photogrammetry
- Minutes to hours
- Voxeliz
- Minutes, because it measures
Sovereign by design
Voxeliz runs on open tools and open models on our own GPUs, not on anyone else's AI service. The output is yours, and every result shows the evidence behind it.
Built on
- COLMAP
- Open3D
- PyTorch
- NVIDIA GPUs
- GLB · OBJ · PLY · STL
No third-party AI services
Reconstruction runs on open-source tools (COLMAP, Open3D and PyTorch) on our own GPU servers. Your photos are never sent to an outside AI provider.
Observed and inferred, kept apart
Every region is marked as measured from your photos or inferred by the system, with a confidence level, so a reviewer can check the model instead of trusting it.
No lock-in
Models export as GLB, OBJ, PLY and STL: open formats you can keep, move and open in any tool, with or without Voxeliz.
Standard hardware
It runs on ordinary NVIDIA GPUs, with no proprietary chips and no closed AI service behind it.
Use it your way
In the browser
Upload, view and measure, review labels and findings, and download models and reports, with nothing to install.
Through the API
Everything the web app can do is available to your own systems, with webhooks and full documentation.
GLB
Web and AR viewers
OBJ
Any 3D tool
STL
CAD and printing
PLY
Point clouds and survey
POST /v1/projectsPOST /v1/projects/{id}/imagesPOST /v1/projects/{id}/reconstruct{ "webhook_url": "https://you.example.com/hook" } # webhook when the job finishes (example){ "status": "COMPLETED", "quality": "Good", "coverage": 0.97, "rmse": 0.41, "labels": 3, "findings": [ { "type": "surface_crack", "confidence": 0.91 } ], "inferred_regions": 1} GET /v1/projects/{id}/export# GLB + OBJ + PLY + STL + report.pdfThe team building Voxeliz
Six people at GYTWorkz, from the first photo to the finished model.
AI Product & Technology Lead
Product direction and technical leadership
AI Solutions Engineer
Bringing models and product components together
AI Research Engineer
Exploring models and improving results
AI Platform Engineer
Infrastructure and model deployment
AI Application Engineer
Building product features powered by models
AI Pipeline Engineer
Data processing and model workflows
See it on your own subject
Voxeliz is in early access. Tell us what you would capture and we will be in touch.