Practical reconstruction
This realitycapture tutorial gives you a repeatable route from a folder of photographs to an inspectable 3D reconstruction. Follow the sequence, check the quality gates, and avoid the common shortcuts that create unusable geometry.
Continue exploring
Use these related guides when your project starts with a different question, such as choosing a tool, planning a drone survey, or understanding the underlying process.
Core workflow
A dependable reconstruction is less about pressing one button and more about preserving useful evidence at every stage.
Choose sharp photographs with consistent exposure and strong overlap. Remove accidental duplicates, blurred frames, and views where the subject is hidden. Keep the original files together so the project can be audited later.
Import the images, run alignment, and review the registered cameras before building dense data or a mesh. Check for gaps, drifting cameras, and weak regions while corrections are still inexpensive.
Trim unwanted surroundings, inspect the model from several angles, and select a format that matches the next tool. Export a test asset first, then make the full-resolution version once scale, texture, and topology look right.
Quality boundaries
This workflow can produce impressive results, but it cannot recover information that the photographs never captured.
A model built from front-facing images will usually contain holes or guesses on the back, underside, or inside of an object.
Workaround
Capture a complete orbit and add elevated, low, or underside viewpoints wherever the subject allows.
Blur, glare, repeated patterns, low texture, and changing exposure can prevent reliable feature matching even when the subject looks clear to a person.
Workaround
Use controlled light, lock exposure when possible, keep the camera steady, and add texture-rich viewpoints.
A visually convincing reconstruction may still be the wrong physical size when the image set contains no reliable scale reference.
Workaround
Include a measured reference, known distance, or surveyed control information when dimensions matter.
Automatic reconstruction can leave floating fragments, thin walls, noisy edges, or texture defects around the main subject.
Workaround
Inspect the result section by section, remove irrelevant geometry, and validate the export in the software that will receive it.
Working reference
The same reconstruction principles apply across projects, but the best setup changes with capture conditions, review needs, and the final deliverable.
Typical subject
Controlled object capture
A statue, artifact, room detail, product, or compact structure
Large-area field capture
Terrain, roads, roofs, stockpiles, or a broad site
Camera movement
Controlled object capture
A close orbit with several elevations and deliberate viewpoints
Large-area field capture
Planned passes that cover the area with consistent spacing
Main quality risk
Controlled object capture
Hidden faces, reflective surfaces, and fine geometric detail
Large-area field capture
Weak overlap, changing altitude, motion blur, and uneven coverage
Best first check
Controlled object capture
Confirm that cameras surround the subject and every important face is visible
Large-area field capture
Confirm that the image footprint covers the full area with usable overlap
Useful output
Controlled object capture
A textured mesh for inspection, presentation, archival work, or downstream modeling
Large-area field capture
A map-oriented reconstruction, terrain surface, orthographic view, or measured site reference
Scale strategy
Controlled object capture
Add a known dimension or reference object if physical size is important
Large-area field capture
Use control points, checkpoints, or a measured ground reference when accuracy is required
Review habit
Controlled object capture
Rotate around the object and inspect holes, floating parts, and texture seams
Large-area field capture
Inspect edges, elevation transitions, coverage gaps, and alignment across the full site
Remember the sequence
These simple checkpoints keep a long processing job from becoming a blind one-way experiment.
Common questions
These answers address the practical decisions people usually face when searching for a realitycapture tutorial.
Start with image preparation and camera alignment before exploring advanced reconstruction settings. If the source photographs are incomplete or inconsistent, changing processing options rarely fixes the underlying problem.
There is no universal number because the subject size, surface detail, camera distance, and desired output all matter. Aim for consistent overlap and complete coverage rather than chasing a fixed count, then use a small test run to find weak areas.
Sharp images with stable exposure, visible surface texture, and substantial overlap are the strongest starting point. Avoid relying on images dominated by reflections, transparent areas, motion blur, or changing viewpoints that leave large gaps.
A camera may lack shared visual features with neighboring images, or the scene may contain blur, glare, repeated detail, or too little overlap. Add intermediate viewpoints, remove poor frames, and make sure the subject remains visible across the sequence.
Usually, no. First run a quicker reconstruction to confirm coverage, alignment, scale, and the approximate shape; only then use more detailed settings and export the final asset.
Inspect the model for holes, floating fragments, distorted edges, incorrect scale, missing texture, and unwanted background geometry. Open a test export in the destination application before generating a large final file.