3D High Poly Models

3D High-Poly Models

Build the requested close-up detail into the model, then make it practical to render. Judge realism through silhouette, proportions, material response, and lighting as well as polygon count.

Establish what the camera needs

Inspect the reference, intended hero angle, closest approach, orbit range, model scale, current mesh, and target hardware. Record the current triangle count and draw calls rather than labeling an asset high-poly from appearance alone. Keep a high-detail source/master when building new geometry so runtime optimization remains reversible.

Separate detail into three scales:

• Silhouette: outer contours, root flares, carved edges, folded fabric, roof curvature. Use geometry.
• Surface form: bevels, dents, joints, raised seams, deep cracks. Use geometry or justified displacement when the view reveals parallax and shadow shape.
• Microdetail: grain, pores, shallow scratches, fiber structure. Use material maps unless the extreme close-up actually resolves geometry.

Use the project's modeling tools or asset pipeline. Match real dimensions and proportions first, then add adaptive subdivisions, support topology, sculpting, or appropriately licensed scans. A dense mesh with incorrect proportions remains incorrect.

Model for believable close-ups

Give manufactured edges a plausible bevel so they catch highlights. Preserve designed hard edges while smoothing curved surfaces; do not smooth every normal or indiscriminately weld UV seams. Recompute or preserve normals/tangents according to the modeling/export pipeline.

Spend curve segments where projected curvature is visible. Increase them around hero rims, roof silhouettes, and tight bends; use fewer on tiny distant parts. Check subdivision boundaries for pinching and displacement for cracks or detached surfaces.

For procedural branches, sweep rings along a smooth curve, transport the local frame to avoid sudden twist, taper continuously, and bury child starts inside the parent surface. Add denser rings at a flared root or strong curvature. Seijaku's treeMesh is a useful example of adaptive radial/longitudinal detail rather than uniform subdivision.

For scanned stones, preserve the scan's characteristic proportions. Excessive nonuniform scaling stretches both shape and surface features. Reuse the mesh with limited variation in rotation, placement, and scale; keep the most recognizable hero forms distinct.

Prepare runtime versions

1. Keep the authored master and export a validated runtime asset, usually glTF/GLB when the stack supports it. Check units, transforms, materials, UV seams, normals, and bounds in the actual viewer.
2. Create lower-detail versions for distant use, preserving silhouette and baked shading. Derive thresholds from projected size and compare transitions from the real camera. Use hysteresis or an appropriate transition to avoid rapid LOD switching. See Three.js LOD.
3. Instance repeated geometry/material pairs such as roof tiles, stones, or foliage where suitable. Partition large repeated fields spatially so one visible object does not force the whole site to draw.
4. Keep shadow geometry and shadow distance proportional to visible benefit. A detailed object rendered in several shadow maps can cost far more than its beauty pass suggests.
5. Compress geometry transport where supported, then validate decoded geometry and appearance. Compression reduces transfer size; it does not inherently reduce runtime triangle count. Treat quantization and simplification as distinct operations with separate visual checks.

There is no universal correct triangle budget. Profile the representative scene with its actual materials, lights, shadows, transparency, and target resolution. Report unique asset triangles separately from the triangles drawn across instances and passes. Geometry compression, instancing, and LOD solve different costs.