zhenzhen-hunyuan-3d-v3.1-pro-fal

The plain Hunyuan3D lowprice node is the "one image, one model" path. This is the Pro path, and the difference is in the schema: it takes front, back, left and right image inputs. Feeding a subject from four angles is the classic trick for getting a much better 3D reconstruction than any single view can give, and the Pro tier is where Tencent expects you to use it.

It's part of the Comfyui-zhenzhen pack's fal family - which means it goes through Zhenzhen's proxy of fal.ai's queue API (ai.t8star.org/fal) instead of the domestic lowprice site, takes its api_key directly in the widget, and bills fal-style: pre-deducted estimate, refund the difference when the job finishes.

How it works

mode picks the task: text_to_3d (prompt → model) or image_to_3d (four views → model). The four views are front_image / back_image / left_image / right_image (or their *_image_url string twins, chosen via image_way). The node submits, polls, downloads, and returns the model.

Unlike the lowprice 3D node, the outputs are a little looser: model_url (STRING), response (STRING), asset_urls (STRING - the individual asset files), and model_3d of type FILE_3D. Note this one is not an output node (is_output_node: false), so it behaves as a mid-graph generator - the 3D output can feed downstream nodes.

Inputs that matter

• mode - text_to_3d vs image_to_3d. For image_to_3d, wire at least the front view; adding back/left/right is where Pro earns its name.
• prompt - required either way. For image mode it's a guide ("a small stylized spaceship" is the default); for text mode it's the entire source.
• generate_type - Normal vs Geometry. enable_pbr - PBR materials on/off. face_count - 40,000 to 1,500,000, default 500,000. Same tradeoffs as the lowprice sibling: density costs time and money.
• front_image_url etc. - the URL route for views if your images are already hosted.

Installing it

Manager → search Comfyui-zhenzhen → install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/Comfyui-zhenzhen

Restart, paste your key into the widget. No model downloads - this is a billed API job.

Common issues

• One-view results looking soft - you're leaving the multi-view advantage on the table. Pro is built for 4 views; give it 4 views.
• Big face_count jobs timing out - the default 600 polls × 6s is a full hour; dense meshes at 1.5M faces can run long. Raise max_poll_attempts rather than panicking.
• model_3d output needs a viewer - it's a FILE_3D-typed output; without a 3D preview node in the graph, grab model_url or asset_urls instead.
• The pack standard gotchas - empty key, 443/VPN-in-TUN-mode for the overseas server, vendor 500s needing a re-run, and the seed being ComfyUI-cache-only, not a reproducibility seed.

The naming is the trap here: "Hunyuan3D" sounds like one thing, but the v3.1 Pro tier and the lowprice tier are different endpoints with different billing. If you want cheap iteration, use the lowprice node; if you want the four-view reconstruction quality, this one - and plan to pay more per run.