Quick answer: A 3D mesh is a set of vertices, edges, and faces that together define the surface shape of a 3D object or scene, the digital equivalent of a wireframe skin stretched over an invisible skeleton. It is the standard way games, AR apps, and 3D printers describe an object’s geometry. Meshes get built by hand in modeling software, captured from the real world by scanning, or generated by software to reconstruct a shape from a set of photos.
This entry covers what a mesh is made of, how one gets created, where AR and game pipelines rely on it, how it differs from a point cloud, and where mesh-based geometry is starting to run into competition from newer techniques.
What Is a 3D Mesh?
A mesh has three basic parts. Vertices are points in 3D space with an x, y, z position, edges connect pairs of vertices, and faces are the flat polygons, usually triangles, formed by three or more connected edges that make up the visible surface.
Triangles dominate real-time graphics because they are always flat and always convex, which makes them cheap for a graphics processor to render, while modeling software often works with four-sided quads because they deform and edit more predictably, converting to triangles only at render time.
A mesh only describes a surface, not what is inside it. That is enough for almost everything a mesh gets used for: rendering, collision detection, and simulating cloth or rigid bodies all only need to know where the outer skin of an object is, not its interior.
How a mesh is organized in memory varies by use case. Some formats store only which vertices connect to which, which is compact but slow to query when software needs to know a face’s neighbors. Others store faces and vertices together so a renderer can walk the surface quickly, which is the layout most graphics hardware expects and the reason it is the default for anything meant to render in real time.
How Are 3D Meshes Created?
Artists build meshes directly in 3D modeling software, pushing and pulling vertices, edges, and faces into the shape they want, the same way a sculptor works clay from the outside in.
Meshes also get captured from the real world through photogrammetry, which reconstructs a 3D shape from a set of overlapping photographs. The software finds points that appear in multiple images, triangulates their position the way a surveyor would, and builds a mesh from the resulting cloud of points. It works well on textured, matte surfaces and struggles with anything dark, shiny, or transparent, which is why photogrammetry scans often need manual cleanup afterward.
3D scanners that use structured light or laser rangefinding skip the photograph step and measure surface points directly, generally producing denser and more accurate meshes at the cost of more expensive hardware.
AI-based mesh generation is newer and less settled. A neural radiance field, or NeRF, a technique introduced in 2020, reconstructs a scene from photos as a continuous volumetric model rather than a mesh, and can be converted to a mesh afterward with reduced precision.
Gaussian splatting is a faster, newer alternative that represents a scene as a large cloud of 3D Gaussians instead of triangles, trading the guaranteed clean geometry of a mesh for faster training and rendering. Both remain more common in research and visual effects than in shipping AR apps, where triangle meshes are still the practical default.
How Do AR and Games Use 3D Meshes?
Game engines are built around meshes end to end. Every character, prop, and piece of environment art is a mesh, paired with a texture that wraps color and detail onto its surface, moved through the engine’s rendering pipeline dozens of times per second.
Augmented reality systems build meshes of the real world as well as placing them in it. On a device with a depth sensor, an AR framework can turn distance readings into a live mesh of the room as you move through it, a rough triangulated model of the walls, floor, and furniture around you.
That mesh is what makes occlusion possible, since the app now has real geometry to hide virtual content behind, and it lets virtual objects rest realistically on a real floor or bounce off a real wall instead of floating through it. Placing that mesh correctly in the room in the first place depends on world tracking to keep everything anchored as the camera moves.
The asset pipeline in both games and AR usually runs through the same handful of file formats. OBJ and FBX are common exchange formats between modeling tools and engines, STL is the standard for 3D printing, and glTF and Apple’s USDZ have become the usual way to deliver lightweight meshes to the web and to mobile AR.
That shared pipeline is also why a mesh built for a video game can often be reused for an AR filter or a virtual try-on feature with only minor rework: the underlying vertices, edges, and faces travel between tools with relatively little translation, even when the rendering engine on the receiving end is different.
3D Mesh vs Point Cloud: What’s the Difference?
A point cloud is just a set of individual points in space, typically the raw output of a 3D scanner or a photogrammetry pass before anything gets connected together. It has no surface and no faces, nothing that tells a renderer what is solid and what is empty space between points.
A mesh adds that structure. Turning a point cloud into a mesh means deciding which nearby points should be connected into triangles, a process that has to make judgment calls about noise, gaps, and where a real edge exists versus where the scan simply missed data. That processing step is why a raw scan often looks rough and a finished mesh looks clean: a lot of decisions get made between the two.
What Are the Limits of 3D Meshes?
Meshes get heavy to work with at high detail. A game or AR asset with millions of triangles renders slowly and takes far more memory and storage than a simplified version, so both games and AR pipelines constantly trade detail for performance, using lower-resolution meshes for anything not directly in front of the viewer.
Meshes also do not capture everything about how a surface looks. Fine detail like fabric weave or skin pores would require an impossibly dense mesh, so games and AR apps fake it instead with textures and lighting tricks layered on top of a comparatively simple underlying shape.
That gap is part of why newer scene representations like NeRF and Gaussian splatting have drawn interest outside pure research settings: they can capture visual complexity a practical mesh cannot, even though they give up the clean, editable geometry that makes meshes so useful everywhere else. For now, the mesh remains the format that games, AR frameworks, and 3D printers all actually agree on.