Hunyuan Text to 3D

What it is

Hunyuan Text to 3D is the prompt-to-mesh version of the ComfyUI-Hunyuan3D-v3 pack: type a description, get back a textured GLB model. No reference image needed - if Hunyuan Image to 3D is for "I have a picture, make it 3D," this one is for "I have an idea, make it 3D." Same pack, same underlying Tencent Cloud API, just a different entry point into it.

Same disclaimer as the rest of the pack applies here too: this calls Tencent Cloud's Hunyuan 3D Global API v3, a paid cloud endpoint, not a locally-run model. It's a different thing from the open-weight Hunyuan3D checkpoints people self-host - this one bills your Tencent account per generation instead of using your own GPU.

How it works

You write a prompt, the node sends it to Tencent's API along with your generation settings, and then polls for up to max_wait_time seconds while the model gets built server-side. Text-to-3D is inherently more open-ended than image-to-3D - there's no reference to anchor the geometry to, so the model has to infer shape from description alone - so expect more variance run to run than you'd get feeding it a clean image.

Inputs and outputs

• prompt - a multiline text field, defaulting to "A cute robot toy" as a placeholder so you know what it expects: short, concrete, object-level descriptions work better than abstract or scene-level ones. Think "a rusted mechanical owl" rather than "a fantasy world."
• enable_pbr - turns on physically-based rendering materials on the mesh (default off). Worth enabling if the model needs to look correct under real lighting downstream; skip it if you just want the geometry.
• face_count - polygon budget, 40,000 to 1,500,000, default 500,000. Same tradeoff as everywhere else in this pack: higher means more detail and a heavier file, and there's rarely a reason to push it to the max.
• generate_type - Normal, LowPoly, Geometry, or Sketch. Normal is the full-detail default; the others trade fidelity for speed or a specific look (LowPoly for game-ready meshes, Sketch/Geometry for rougher blockouts).
• polygon_type - triangle or quad, depending on what your downstream tooling expects.
• max_wait_time - 60–3600 seconds, default 600. The README doesn't split out text-to-3D timing separately from image-to-3D, but treat it the same way: a few minutes is typical, so the default has room built in.
• config - the HUNYUAN_CONFIG from a Hunyuan Config node. Required; this node can't authenticate without it.

Output is a single model_path string pointing at the saved GLB. It's text, not a native mesh type, so feed it into a Hunyuan 3D Model Preview node to actually view it in ComfyUI, or use the path directly outside ComfyUI.

Install

Via ComfyUI Manager: search "Hunyuan 3D", install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/exedesign/Hunyuan-3D-v3.git
cd Hunyuan-3D-v3
pip install -r requirements.txt

You'll need a Hunyuan Config node upstream with your Tencent Cloud API keys - see that node's page for the setup walkthrough, which is the part that actually takes effort here.

Troubleshooting

A ResourceInsufficient error is an account problem, not a node problem - your Tencent Cloud balance is empty or the Hunyuan 3D service was never activated for that account. Fix it in the console, not in the workflow. If the prompt just isn't producing what you pictured, that's a text-to-3D limitation rather than a bug: vague or scene-scale prompts tend to produce vague or malformed geometry, so tightening the prompt to a single concrete object is the first thing to try before assuming something's misconfigured. And if you're running this somewhere that enforces its own outer timeout on a job - comfy.icu defaults to 5 minutes per run - make sure that's raised above your max_wait_time, since the outer timeout will kill the run first otherwise.