Low Poly vs High Poly 3D Modelling: Which Fits Your Project
TL;DR
Low poly and high poly describe a model's relative geometric density. More polygons can represent finer changes in shape; fewer can reduce geometry cost when placed carefully. Neither label guarantees realism or speed. Choose the detail needed for the intended camera, animation and target device, then check the exported asset.
Low poly and high poly describe a model’s relative geometric density. More polygons can represent finer changes in shape; fewer can reduce geometry cost when placed carefully.
Neither label guarantees realism or speed. Choose the detail needed for the intended camera, animation and target device, then check the exported asset.
A project may use a detailed source model and a lighter delivery version, or model directly at the level it needs.
For a product team, that makes this a delivery question, not a style debate: what must the viewer see, on which device, and from how close? Our 3D modelling work starts from that question.
Low poly vs high poly at a glance
Both build models from polygons. The table compares them on the points that decide which one a job needs. Both can be used in many contexts.
| Lower density (low poly) | Higher density (high poly) | |
|---|---|---|
| Relative geometry | Fewer polygons, placed where the shape needs them | More polygons, able to describe finer changes in shape |
| Visible silhouette | Can hold a clean outline if the polygons sit on the edges that show | Holds smooth curves and fine profile detail at close range |
| Potential use | A constrained real-time target, such as a web viewer or AR, or a deliberately faceted style | Source sculpting, close-up offline renders and film |
| Optimisation | May reduce geometry processing, subject to the whole scene and device | Usually needs a lighter version for real-time targets |
| What to check | Silhouette at the closest zoom, shading, and behaviour on the target device | Memory and scene complexity, and whether the extra detail is ever seen |
Material maps, lighting and camera strongly affect the result in both columns. A polygon count on its own tells you little about how an image will look.
Geometry and appearance are different things
A dense model can still have the wrong dimensions or a poor material. A simplified one can look convincing when its silhouette and shading suit the view. So place geometry where it does visible work: the outline, holes, contact points, reflections, shadows and anything that deforms.
Texture resolution and shader complexity are separate budgets from polygon count. When you compare two versions of a model, compare them on the same product revision, camera, light and output size, or the comparison means nothing.
Flat, faceted shading is a style choice; it is not the same thing as a low polygon count. For how surfaces are built, see our guide to realistic materials and surface finishes.
What about mid poly?
Mid poly is a relative label between the two: enough geometry to hold a clean silhouette without the density of a sculpt. Like low and high, it has no universal polygon count. It only means something against a particular model and target.
Workflow options: retopology, decimation and baking
There is no single required route from detailed to light. The common options are:
- Retopology. A source sculpt or CAD-derived mesh is rebuilt with controlled edge flow. This suits parts that deform or need clean subdivision.
- Automatic decimation. Software reduces the polygon count. It can suit some rigid forms, followed by inspection and repair.
- Modelling at delivery density. Some assets are built directly at the level the target needs.
- Baking. Once topology and UVs are stable, shading detail from a detailed model can be transferred into maps for a lighter one.
The right method depends on the geometry, whether it deforms and what the project needs. Neither a high-to-low bake nor an all-quad mesh is universally required. Keep the editable source, and label each derivative version with its purpose and target. XO3D Labs explains what retopology is in more detail.
What a normal map can and cannot do
A normal map changes how light is calculated across a surface. It does not add a protruding edge or change the silhouette. A deep gap, a hole or the outer profile may still need real geometry, particularly when the camera is close or the part moves.
A bake must be checked in the actual target, for seams, normal map convention and tangent basis. Displacement is a separate technique that does move the surface, and support for it depends on the target. A good normal map preview in an authoring tool is not proof that the exported viewer shows the same result.
A destination checklist
Before choosing a density, record what the model is for. A simple hand-off record:
- The approved source revision.
- The intended view and the closest zoom the viewer will reach.
- The target device, and the renderer or viewer and its version.
- Any animation the model needs.
- The file format.
- Triangle and vertex counts after export.
- Material count, and texture sizes and formats.
- Download size, and observed load time and frame behaviour on the target.
Compare candidate versions under the same camera and light. If a change affects the silhouette or an important feature, ask the product owner to review it.
Use a budget measured for your project, not a borrowed benchmark used as a universal pass mark. The free XO3D Labs Polygon Budget Calculator is a useful starting estimate; it is a learning aid, not a test of how a model performs on your device.
Where each fits
- Offline stills and film. Renders made frame by frame can often carry dense geometry where it is useful, but scene complexity and memory still matter.
- Real-time games, web, VR and AR. These need a budget for the actual device and experience. Modern engines can handle high geometric detail in some contexts, so check the target rather than assume.
- Close inspection. A model that looks fine as a thumbnail may fail at inspection zoom. Test at the closest view the viewer can reach.
- 360 image sequences versus interactive 3D. A 360 sequence is a set of rendered images; an interactive viewer loads the model itself. They are different deliverables with different constraints.
For the wider picture of how a model is defined and used, see our guide to what a digital model is. For learners, the XO3D Labs guide to polygons in 3D covers the basics. More guides are in the XO3D Journal.
What this comes down to
Low poly and high poly are not two camps. They are points on a scale, and the right point depends on the camera, the movement and the device. Build the detail the job needs, keep the source, make lighter versions for specific targets, and check the exported asset where it will be used.
Start the conversation
Have a product worth showing?
We reply to every brief personally, usually within one working day.
FAQ
Common questions, answered.
What is the difference between low poly and high poly 3D models?
The labels describe relative geometric density. A high poly model has more polygons, so it can describe finer changes in shape; a low poly model has fewer. There is no universal count that separates them; the difference only means something against a purpose and a target device.
When should I use low poly models?
When the target has a tight budget for geometry, such as a web viewer, a mobile AR model or a real-time scene, or when a faceted look is the intended style. Check the exported model on the real device, because geometry is only one of several costs.
What are the advantages of using high poly models?
More polygons can hold a smooth silhouette and fine shape detail at close range, and they suit source sculpting and offline renders. More geometry does not by itself make materials, dimensions or lighting accurate.
Can a low poly model look realistic?
Yes. A lighter model can look convincing when its silhouette is right where it is seen and its materials and lighting are good. Realism depends on shape where it matters, material response and light, not on polygon count alone.
Does a normal map change the silhouette?
No. A normal map changes how light is shaded across a surface; it does not move any geometry. Deep gaps, holes and the outer profile still need real geometry, especially when the camera is close or the part moves.
Should I order both a high poly and a low poly model?
It depends on where the model will be used. A detailed source model can be kept alongside a lighter delivery version made for a specific target, such as a web viewer or AR. Agree the targets at the brief so each version is made and checked for its use.
Can I convert a high poly model to a low poly version?
Often, yes. Common routes are retopology for controlled edge flow, automatic decimation followed by inspection and repair, and baking surface detail into maps. Which route fits depends on the shape, whether it deforms and where it will be used.
How does polygon count affect rendering performance?
Polygon count is one cost among several. Textures, materials, the number of separate objects, animation and the target device all matter too. A lighter mesh may reduce geometry processing, but it does not guarantee a faster render or a higher frame rate on its own.
Start the conversation
Got a product worth showing? Let’s talk about what it needs to do.
We reply to every brief personally, usually within one working day.