Bingeljell/image-to-3dlab

AssetFurnace: the most complete open-source image-to-3D pipeline (Oct 2026). Prompt or picture → textured, rigged, animated game character, locally on Apple Silicon or NVIDIA. Pixal3D, TRELLIS.2, Hunyuan3D, SF3D, auto-rig and 21 moves. For indie game devs and 3D artists. Free, Apache-2.0.

AssetFurnace (formerly Image‑to‑3D Lab)

What it does – AssetFurnace is a locally‑run tool that turns a single text prompt or a 2‑D image into a game‑ready 3‑D asset. It builds a mesh, textures it, rigs a skeleton, and applies one of 21 animation presets, then outputs a GLB file that can be dropped into Unity, Unreal, Godot, Blender, or three‑js.

Key features

  • One‑click pipeline – The web‑based “studio” runs all stages (image generation, shape creation, retopology, texture repaint, Pixel‑Match, compression, rigging, animation) automatically, with the ability to stop after any step.
  • Pixel‑Match – After the model is generated, the original image’s exact pixels are projected back onto visible surfaces, preserving logos, text and facial details even on a low‑poly (~5 k faces) mesh.
  • Prop sheets – A single image containing a grid of props can be turned into nine separate, LOD‑packed assets in one run.
  • Multiple back‑ends – Supports several 3‑D generation engines (Pixal3D, Hunyuan‑3D, TRELLIS.2, Stable Fast 3D) and runs on Apple Silicon, NVIDIA, or AMD (ROCm) GPUs. The back‑ends are selected in the UI or via CLI.
  • Local‑only, licence‑aware – No cloud upload; all processing stays on the user’s machine. Every output includes a provenance JSON that records the model weights and licences used.

Installation & usage

  • One‑line installer for macOS/Linux (curl … | bash) or PowerShell for Windows. The script installs Python 3.11, pulls the code, and launches the studio in a browser.
  • The UI lets you type a prompt (uses Qwen‑Image 2.1 locally) or upload a masked PNG, then hit Create.
  • A CLI mirrors the UI: scripts/pixal3d_generate.py, scripts/hunyuan_mlx_xiong_generate.py, etc., with options for seeds, steps, and resume modes.

Hardware requirements

  • Apple Silicon (M‑series) – recommended 32 GB RAM.
  • Linux with NVIDIA (≥24 GB VRAM) or AMD ROCm (tested on RX 7900 XTX, Ryzen AI Max+ 395).
  • Required toolchains: Xcode on macOS, CUDA toolkit for NVIDIA, ROCm SDK for AMD, and the uv Python package manager.

Why it matters AssetFurnace bundles several state‑of‑the‑art image‑to‑3‑D models into a single, user‑friendly package that runs entirely offline. It lowers the barrier for indie developers and hobbyists to generate rigged, animated characters and props without paying for cloud services or dealing with complex model‑weight licensing.

Related

  • Project
    pwilkin/trellis.cpp

    trellis.cpp is a C++ GGML implementation of Microsoft’s TRELLIS 2‑4B image‑to‑3D pipeline. It converts a single image (or a Stable Diffusion‑generated image) into a fully textured GLB model, running entirely in native C++ with optional GPU back‑ends (CUDA, Vulkan, ROCm, Metal, Qualcomm HTP). It provides a CLI, a Tauri‑based desktop UI (Trellis Studio), and an HTTP server for programmatic use. The repo includes the full pipeline – background removal, three flow‑transformer stages, VAE decoders, mesh extraction, QEM decimation, UV unwrapping and WebP‑compressed PBR textures – and ships pre‑built binaries and GGUF model files.

  • Project
    dreamers-laboratory/image-to-3d-pipeline

    A pipeline for turning images into 3D meshes that allows users to run and compare multiple open-source reconstruction models to find the best output.

  • Project
    KennethJAllen/proper-pixel-art

    A tool that converts noisy, high-resolution pixel-art-style images and animations into clean, true-resolution pixel art assets by automatically recovering the underlying pixel grid.

  • Project
    lukasHoel/video_to_world

    A 3D reconstruction pipeline that resolves geometric inconsistencies in videos generated by video diffusion models to create stable 3D worlds.

  • Project
    visualbruno/ComfyUI-Trellis2

    ComfyUI‑Trellis2 is a custom‑node pack that integrates Microsoft’s TRELLIS.2 3‑D diffusion pipeline into the visual ComfyUI interface. It adds multi‑view generation, voxel‑to‑mesh conversion, texture mapping, and a suite of geometry‑processing nodes (smoothing, hole‑filling, simplification, etc.). The repo supplies Windows wheels for required C++ extensions, detailed installation steps, and example workflows, enabling users to turn images into textured 3‑D meshes directly within ComfyUI.