AssetFurnace (formerly Image to 3D Lab, same repository) is a local studio and CLI that turns a text prompt or a picture into a textured 3D model for games, then optionally rigs and animates humanoid characters. When there is no starting image, Create first draws one with Qwen-Image 2.1 on the same machine, which the README times at about 4.5 minutes on a base M5 MacBook with 32 GB, and then continues to 3D. Six image-to-3D backends sit behind one interface: Pixal3D (C++/GGML, the one the README recommends starting with), Hunyuan3D-MLX in two variants, Hunyuan3D-2.1 on NVIDIA, TRELLIS.2 and Stable Fast 3D (Mac only). Finish retopologises the mesh and compresses textures, taking a typical asset from 32 MB to under 5, and Pixel Match copies real pixels from the source picture back onto the surfaces it can see so text and logos stay legible. Prop sheets split one 3x3 picture into nine upright props with three LODs of 5,000, 2,500 and 1,000 triangles. Rig uses VAST-AI SkinTokens to fit a skeleton and skin weights to a humanoid in an A-pose or T-pose, with 21 preset moves exported as a GLB for Godot, Unity, Unreal, Blender or three.js. Hardware matters: the README lists an Apple Silicon Mac (32 GB recommended) or Linux with an NVIDIA card (24 GB VRAM tested) or an AMD card with ROCm, where Pixal3D is the only 3D route; Windows has had limited testing, and the TRELLIS.2 and Hunyuan3D-2.1 CUDA setups are marked as not yet tested on real hardware. Releases v0.4.0 (rig and animate) and v0.4.1 (AMD on Linux) are tagged, up to v0.4.3; Front + back generation is described in the README and changelog but is still listed under Unreleased on the main branch. The repository code is Apache-2.0, but model weights keep their own licences: Pixal3D and TRELLIS.2 are MIT with the separate DINOv3 License for the bundled encoder, and the Hunyuan3D weights fall under the Tencent Community License and are not licensed in the EU, UK or South Korea. Each run writes a .provenance.json with the licences that apply.