Open-source text-to-3D models can now generate usable 3D scenes and GLB files on a single consumer GPU. This site covers Qwen 3.5 4B 3D Coder and its variants: what it does, how to run it in ComfyUI, what hardware you need, and how to get results into Blender or Unity.
The 4B model family generating Three.js scenes and GLB files from text prompts — variants explained.
Get the model loaded in ComfyUI step by step, from weights download to first generation.
What the community实测 reports say about running the 4B and 1.7B variants, and what to do with 8GB cards.
Take generated scenes into Blender, Unity and Unreal via the GLB variant — pipeline notes.
Free local generation vs Tripo, Meshy and friends: cost, control, quality and when to use which.
OOM errors, missing nodes, broken scenes — the common failure modes and fixes.
For years, "text to 3D" meant cloud services with per-generation credits. Open-weight models changed that: a 4-billion-parameter model can draft interactive Three.js scenes from a prompt, and a GLB-focused variant outputs files that drop straight into game engines. If you build game prototypes, 3D-print previews or web experiences, running this locally means no per-click cost and no upload of your prompts.
Start with the model overview, then follow the ComfyUI setup for your GPU tier.