LiteModeler

Overview

LiteModeler is a standalone Electron application for AI-powered 3D model generation running entirely on your NVIDIA GPU. Type a text prompt or provide a reference image — LiteModeler generates production-quality GLB meshes with controllable polygon counts, texture resolution, and quality presets.

Generation Modes

Prompt-to-3D

Type a text description. LiteModeler calls LiteImage (running locally) to generate a reference image, then feeds it into the 3D reconstruction model. End-to-end prompt-to-GLB.

Image-to-3D

Provide a reference image directly. Supports transparent backgrounds for cleaner geometry reconstruction.

AI Models

| Model | Description | |-------|-------------| | Trellis2 GGUF | Primary model, optimized for consumer GPUs via quantized GGUF format | | Hunyuan3D-2.1 | Tencent's open-source model (requires HuggingFace auth) | | Hunyuan3D-Omni | Advanced variant for higher-quality outputs |

Quality Controls

Individual control over every stage of generation:

| Parameter | Range | Description | |-----------|-------|-------------| | Sparse Steps | 6–30 | Shape generation iterations | | Shape Steps | 6–30 | Geometry refinement | | Texture Steps | 6–30 | Texture generation quality | | Face Count | 50K–500K | Target polygon count | | Texture Resolution | 1024–4096px | Output texture size |

Presets: Draft (fast iteration), Standard (balanced), High (quality), Max (production).

Seed control for reproducible outputs. Optional no-texture mode for geometry-only meshes.

Features

• Interactive 3D Viewer — Three.js viewport with orbit controls and real-time progress
• Asset Gallery — grid/list views with search, filter, sort, and metadata
• Setup Wizard — GPU detection, model downloads, HuggingFace auth
• Terminal Panel — built-in terminal for debugging and monitoring
• Multi-panel Workspace — splitter, grid, tab layouts

Architecture

| Component | Technology | |-----------|-----------| | Frontend | React 19, TypeScript, Electron 37 | | 3D Rendering | Three.js, React Three Fiber, Drei | | State | Zustand | | Inference | Python runners (subprocess) | | Model Format | GGUF (quantized), safetensors | | Output | GLB meshes |

Integration with LiteSuite

LiteModeler depends on LiteImage as a sidecar for prompt-to-3D workflows. LiteImage's API (port 7426) generates the reference image that feeds into 3D reconstruction.

Requirements

• OS: Windows 10/11 (64-bit)
• GPU: NVIDIA with 12+ GB VRAM (CUDA required)
• Disk: 100+ GB for model weights + outputs
• Python: 3.8+ for inference runners

Troubleshooting

Setup wizard shows "No NVIDIA GPU detected." LiteModeler requires a CUDA-capable NVIDIA GPU. Ensure your drivers are up to date (nvidia-smi to verify).

Model download fails. Some models (Hunyuan3D) require HuggingFace authentication. Enter your HF token in the Setup panel.

Generation is very slow. Reduce quality settings — use Draft preset or lower sparse/shape/texture steps. Ensure no other GPU-intensive apps are running.