3D Modeling

3D Modeling is the creation and editing of digital geometry that represents a shape in three dimensions. A model can represent a real object or an imagined one. Modeling establishes the shape; materials, lighting and rendering determine how that shape appears in an image. Modeling software can provide several editable geometry types, including polygon meshes, curves and surfaces.

In XR, modeling supplies objects and environments for Virtual Reality and Augmented Reality applications. Models made in authoring tools must be converted into data supported by the application's graphics engine, with suitable scale and complexity for the target device. Geometry may also originate in object capture, including Photogrammetry, before editing and integration into an application. Immersive modeling is a separate use of XR: the author creates or edits geometry while working inside a virtual environment. Original studies have examined spatial drawing, surface construction and architectural modeling with immersive interfaces.

The technique is distinct from the applications described in 3D Modeling Software. It also overlaps with Computer-Aided Design, which includes 2D drafting, dimensions, design rules and technical documentation as well as 3D geometry creation.

Reviewed 12 October 2026. Geometry representations and modeling methods, documented XR asset preparation and interchange limits, and the task-specific findings and limitations of the original immersive-authoring studies cited below. About review dates.

Geometric representations

Different representations give authors different ways to control a shape. A polygon mesh is edited through its vertices, edges and faces; a NURBS surface is controlled through a mathematical curve or surface representation; a signed distance field describes a surface through distances sampled in space.

Representation	Shape description and editing
Polygon mesh	Connected faces bounded by vertices and edges. Faces may be triangles, quadrilaterals or polygons with more sides. Mesh topology describes how those components connect, rather than only where the vertices are positioned.
NURBS curves and surfaces	Non-uniform rational B-splines represent both standard geometric forms and freeform geometry. Degree, control points, weights, knots and evaluation rules determine a curve; moving control points changes its shape.
Subdivision surface	A control mesh defines a smooth limiting surface through subdivision rules. Catmull-Clark subdivision updates and introduces vertices at successive steps; the mathematical surface can also be evaluated directly without explicitly generating all those subdivisions.
Volumetric or implicit representation	Geometry can be represented through values in a volume. For example, Adobe Substance 3D Modeler stores its digital clay in a voxel grid containing signed distances to the surface. The sign identifies inside and outside; polygons are generated to display the surface.

The editable source representation and the exported representation can differ. Autodesk Maya, for example, provides conversions from NURBS and subdivision surfaces to polygons, while Substance 3D Modeler exports polygon meshes from its clay representation. This conversion is not necessarily reversible without loss. Adobe documents that converting an imported mesh into its clay representation loses UV coordinates, material color information and some surface precision.

Modeling methods

Direct mesh editing changes a model's vertices, edges or faces. Procedural modeling instead records operations and their relationships. In Houdini, a network can create a box, extrude it, subdivide it and edit points; changing an earlier node's parameters propagates through the network to alter the result. Procedural construction is therefore an authoring method, rather than a separate requirement that the resulting model use a particular geometry representation.

Sculpting changes geometry with brush strokes. A surface sculpting tool can move mesh vertices or NURBS control points through operations such as pushing, pulling and smoothing. A digital clay workflow can add and remove material, warp a shape or smooth its surface. The clay metaphor does not itself specify the underlying data structure. Boolean operations provide another construction method: union combines shapes, subtraction cuts one shape from another, and intersection retains their overlap.

Captured data can be a starting point for modeling rather than a finished surface. A Point Cloud contains sampled points; surface reconstruction creates a surface from such samples. Kazhdan, Bolitho and Hoppe's original Poisson surface reconstruction paper formulates reconstruction from oriented points as a spatial Poisson problem, then extracts an isosurface. The authors discuss fitting scanned data, filling holes and remeshing as applications. Microsoft's mixed reality asset guidance includes capture followed by editing in digital content creation software as an asset creation route.

Preparing models for XR

Topology and surface detail

Remeshing regenerates a mesh with approximately the same shape but different face count or connectivity. Retopology can create a replacement mesh over an existing surface, with connections chosen for the intended use. Blender's documentation distinguishes automated uniform remeshing from manual topology designed for deformation, and treats decimation separately. A mesh with fewer faces and a mesh with suitable deformation topology are therefore different production goals.

Export tools may automate part of that work. Substance 3D Modeler offers raw triangles, automatic retopology toward a target triangle or quad count, and retopology with generated UVs. Its target polygon count is an estimate, and adaptive topology distributes more polygons where detail requires them. For Unreal Engine static meshes, the FBX pipeline requires triangulated geometry. Epic recommends controlling triangulation in the authoring application so the placement of edges can be inspected before import.

UV mapping associates a surface with two-dimensional texture coordinates. Editing the model can change that relationship, so UV layout is part of preparing geometry for texturing. Baking can transfer surface characteristics from a detailed source mesh into texture maps for a simpler target mesh. A normal map gives the lower density surface the appearance of finer detail; it is a texture output, rather than the source geometry itself.

Scale, animation and interaction

The asset's unit scale, axis orientation and origin must be interpreted consistently across applications. Adobe's exporter provides unit-scale and up-axis settings, while the glTF specification defines linear distances in meters and a right-handed coordinate system with Y pointing up. Correct scale is especially relevant when virtual objects appear alongside the physical world. A pivot is also part of the asset setup: it determines the point around which an object's translation, rotation and scaling are applied.

Animated geometry needs information beyond a static shape. In glTF, skinning connects vertex joint indices and weights with a joint hierarchy; morph targets supply displacements relative to a base mesh. Those target attributes must contain the same number of elements as the corresponding base attributes. Interaction can use a separate collision representation. Epic's static mesh pipeline supports simplified collision geometry and custom collision meshes, including shapes that preserve openings where objects must pass through.

Complexity and interchange

Level of Detail models provide versions of varying complexity for runtime use. Unreal Engine can import authored LOD meshes or generate them after import; its static mesh documentation describes reducing triangles and sometimes simplifying materials as objects become farther from the camera. Geometry count alone does not define an XR performance budget. Microsoft's asset guidance treats active assets, shaders, textures, physics and device capability as combined tradeoffs. Epic's VR guidance likewise identifies gameplay, draw-call and GPU shading costs, recommends limiting material complexity and translucency, and calls for profiling throughout development.

Interchange requires checking the data that survives the chosen exporter and importer. FBX can carry static meshes, UV sets, vertex colors and LODs through Unreal Engine's documented pipeline, but its material import has specific texture and connection limits. glTF defines rendering geometry with vertex attributes, materials, skins and morph targets. These capabilities do not establish that an arbitrary authoring application's construction history or procedural network will remain editable after export: SideFX documents that network as part of the authoring process, while glTF specifies the resulting asset data.

Immersive authoring

Immersive interfaces can let an author draw spatial curves, manipulate surfaces or sculpt digital clay within a virtual scene. Drawing on Air investigated haptic assistance for controlled 3D curves, while Vlah and colleagues examined surface creation with Gravity Sketch. Substance 3D Modeler documents digital clay creation on desktop and in VR, with export settings for handing the resulting geometry to other applications.

Creating geometry and reviewing an existing design are different tasks. A 2026 study tested desktop and immersive modeling in one experiment, then desktop, immersive and hybrid design review in another. Its hybrid condition used VR to identify issues and desktop tools for correction, rather than assuming the same interface was best for every operation.

Research findings

Original study	Tested task and finding	Boundary of the evidence
Keefe, Zeleznik and Laidlaw (2007)	Drawing on Air combined one-handed drag drawing and two-handed tape drawing with haptic assistance. In a quantitative study with illustrators, both techniques performed better than the tested freehand and haptic-friction alternatives.	The study concerns controlled 3D line illustration. It does not establish performance for an entire mesh production workflow.
Vlah, Čok and Urbas (2021)	Seven mechanical engineering students modeled a computer mouse from reference images in Gravity Sketch. They produced enclosed surface models that researchers subsequently converted to solid bodies.	Shape matching was uneven. Conversion required desktop work after the participant test, and that conversion time was not measured. The result is bounded to the tested task and software of 2021.
Tastan, Heyik and Er (2026)	Experiments involved 34 novice architecture students in modeling and 48 in design review, using SketchUp Pro 2024 and VR Sketch v16. Immersive modeling took longer and required greater physical effort; precision differences depended on task complexity. Hybrid review had shorter completion time than desktop review.	The modeling and review experiments had different designs. Findings from these students, tools and tasks do not establish a general advantage or disadvantage for professional immersive modeling.