Skip to main content
skill CX Codex

3d-high-poly-models — Claude Code skill: detailed 3D models

3d-high-poly-models is a Claude Code skill that creates or integrates high-poly 3D models with smooth silhouettes, bevels, and close-up geometry, then prepares runtime LODs and loading.

MIT 873 installs ★ 7,995 updated 2026-10-25

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add MengTo/Skills --skill 3d-high-poly-models -a claude-code
Open on GitHub

What 3d-high-poly-models does

Skill works with detail at three scales: silhouette (outer contours, roof curvature, fabric folds) uses geometry; surface form (bevels, seams, deep cracks) uses geometry or displacement when the view reveals parallax and shadow shape; microdetail (pores, scratches, fibers) uses material maps unless extreme close-up actually resolves geometry. For procedural branches, skill places rings along a smooth curve, transports the local frame without twist, tapers continuously, and buries child starts inside the parent surface; adds denser rings at a flared root or strong curvature.

After building the master version, skill exports a validated runtime asset (usually glTF/GLB), checks units, transforms, materials, UV seams, normals, and bounds in the actual viewer. Creates LODs for distant use while preserving silhouette and baked shading; derives thresholds from projected size and compares transitions from the real camera. Instances repeated geometry/material pairs (roof tiles, stones, foliage), partitions large repeated fields spatially so one visible object does not force the whole site to draw. Without FFmpeg or glTF support in the stack, export may break.

Skill loads visible hero geometry first and defers optional interiors or distant detail. Generates procedural meshes offline or in bounded work, uses workers where useful, reuses buffers instead of regenerating identical parts. For selection and collision, uses suitable proxies or a spatial acceleration structure.

What it helps with

  • create a 3D model with close-up detail and smooth silhouette
  • prepare LOD versions preserving shape and baked shading
  • optimize loading and runtime without losing visual quality
  • instance repeated geometry like roof tiles, stones, foliage
  • build procedural branches with smooth curves and adaptive ring density

Example prompts for the agent

  • Build a high-poly building model with believable bevels on edges and roof detail for close-up
  • Create LOD versions for a stone model, preserve silhouette and set transition thresholds based on camera projection
  • Generate a procedural tree with smooth branches, flared roots, and buried child starts

How to install 3d-high-poly-models

1Skills CLI (Vercel)

npx skills add MengTo/Skills --skill 3d-high-poly-models -a claude-code

2Manually: put the files in place

~/.claude/skills/3d-high-poly-models/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/3d-high-poly-models/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about 3d-high-poly-models

How to install 3d-high-poly-models in Claude Code?

Run npx skills add MengTo/Skills --skill 3d-high-poly-models -a claude-code, the skill will appear in ~/.claude/skills/3d-high-poly-models/SKILL.md.

What does 3d-high-poly-models skill do?

Creates or integrates high-poly 3D models with smooth silhouettes, bevels, and close-up geometry, then prepares LOD versions and optimizes loading for runtime. Works with detail at three scales: silhouette, surface form, microdetail.

What export formats does the skill support?

Skill exports runtime assets usually in glTF/GLB when the stack supports it, and checks units, transforms, materials, UV seams, normals, and bounds in the actual viewer.

Source text

SKILL.md · MIT
---
name: 3d-high-poly-models
description: Create or integrate highly detailed 3D models with smooth silhouettes, shaped surfaces, believable bevels, and close-up geometry, then prepare suitable runtime LODs and loading. Use when a user requests high polygon counts, ultra-realistic models, detailed architectural assets, or removal of visibly faceted geometry.
---

# 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](https://threejs.org/docs/pages/LOD.html).
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.

## Avoid a startup stall

Load visible hero geometry first and defer optional interiors or distant detail. Generate procedural meshes offline or in bounded work, use workers where useful, and reuse buffers instead of regenerating identical parts. Keep a useful coarse representation until the detailed asset is ready. Prewarm shader variants with supported asynchronous compilation when it materially reduces first-use hitches.

For selection and collision, use suitable proxies or a spatial acceleration structure. Repeated full-resolution triangle scans during pointer movement can make a visually smooth scene feel unresponsive.

## Verify

- Compare the master and runtime mesh at the closest camera distance under neutral and final lighting.
- Inspect silhouette, bevel highlights, seams, contacts, shadow shape, and wireframe density.
- Walk across LOD thresholds and confirm there are no visible holes or distracting jumps.
- Measure startup, first interaction, full-frame draw calls/triangles, and frame time on desktop and mobile.
- Claim ultra-realistic appearance only when the rendered result supports it; record the measured geometry count separately.

Read [REFERENCES.md](REFERENCES.md) for source and APIs. Seijaku combines detailed swept trunks and scanned rock geometry with inexpensive instanced leaf cards. Do not interpret its appearance as evidence that every object needs a dense mesh.

© MengTo · MIT · MengTo/Skills

Similar skills

All skills for Game dev and 3D: Unity, Unreal, Godot, Blender →

From the same set178

The whole set →
threejs-weather threejs-weather adds controllable rain, storms and snow to Three.js scenes. Use it when weather needs to affect particles, lighting and ground appearance. CX Codex +1 1,500installs implement-fog-of-war implement-fog-of-war — a skill that walks an AI agent through building and debugging fog of war in a Three.js game: wall-aware visibility, player and enemy vision ranges. CX Codex +1 1,500installs build-vesperfall-review-assets build-vesperfall-review-assets builds review pairs for the Vesperfall catalog by pairing transparent PNGs with live Three.js or FBX models. Useful when adding characters or equipment and setting up isolated moveset routes. CX Codex +1 1,500installs 3d-virtual-tour 3d-virtual-tour — a skill for building virtual tours through a 3D scene: authored camera paths, room navigation, orbit inspection. Useful for architectural walkthroughs and showrooms in Three.js. CX Codex +1 913installs 3d-high-resolution-textures 3d-high-resolution-textures picks texture resolution and PBR maps so close-up surfaces stay sharp without blowing up GPU memory. CX Codex +1 870installs 3d-retina-resolution 3d-retina-resolution keeps 3D canvases sharp on Retina and HiDPI displays, including fixed 200% rendering. Useful for blurry WebGL previews and display-density changes. CX Codex +1 869installs 3d-sky-background 3d-sky-background builds camera-correct 3D skies with procedural atmosphere, horizon haze, aligned sun lighting, and coherent environment reflections. CX Codex +1 860installs 3d-sky-rays 3d-sky-rays — a skill that adds crepuscular rays to a Three.js scene, shaped by roofs and foliage. Useful for golden-hour architectural shots. CX Codex +1 859installs

Project sponsors

Companies that keep this analytics open