Tripo 3D v3.1

This is the odd one out in the pack, and the most fun. Runware_tripo_v3_1 generates actual 3D models - textured, watertight meshes - from a text prompt, an image, or both, and it's the one Runware node where nothing is required. Empty prompt, no images, just hit Queue and you get a GLB or FBX file back. It's Tripo's v3.1 model (tripo:v3.1@0) doing the generating, and the whole thing costs you exactly one API call.

Reality check on what you get: this is generation-quality 3D, not scan-quality 3D - think game-asset placeholder, concept prop, or starting mesh, not a production hero model. What makes it worth a node is the loop it enables: generate a 3D asset from text, or turn a 2D image (from any other node in this pack) into a 3D version of itself. Wire a Nano Banana output into images and you've got an image-to-3D pipeline in three nodes.

How it works

3dInference task with the tripo:v3.1@0 AIR. Text and images go up; the returned model file is downloaded to your ComfyUI output folder and the node returns a file_path string - note there's no native 3D viewer tensor, so the output is a path you feed to a file node or open externally. The settings.* group is where the mesh quality lives: pbr (physically-based rendering materials, on by default), texture, geometryQuality (standard/detailed), faceLimit, quad mesh output, generateParts for segmented editable parts, and smartLowPoly for a lightweight mesh on simple inputs.

The inputs that matter

• images - the IMAGE input for image-to-3D. One clean, well-lit view of an object is the classic sweet spot.
• positivePrompt - describe the object ("a ceramic mug with a blue glaze and a handle"). Empty works; detailed works better.
• outputFormat - GLB (default) or FBX. GLB is the portable web/game format; FBX if you're feeding a DCC tool that wants it.
• settings.faceLimit - the polygon budget; lower for a lighter mesh, higher for detail.
• settings.autoSize - scales to real-world dimensions in meters, useful if the mesh has to fit a scene.
• settings.generateParts - on for individual editable parts; costs more and isn't always needed.

Installing it

Install the pack. ComfyUI Manager → search Runware → install → restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/Runware/ComfyUI-Runware
pip install -r ComfyUI-Runware/requirements.txt

Key from runware.ai/api-keys → ComfyUI Settings → Runware API key (or RUNWARE_API_KEY). No local 3D tooling needed - generation happens server-side.

Where people get burned

• The output is a file path, not a 3D view. If you wire this into a preview and see a string, that's correct behavior - the model saves to your output folder and hands back where it put it.
• Quality expectations. "Generation-grade, not production-grade" is the honest framing. Complex or multi-part subjects (the smartLowPoly tooltip flat-out warns it "may fail on complex models") come out rough.
• Text-to-3D is harder than image-to-3D. The model is generally stronger when you give it a reference image than when you describe "a dragon in armor" from nothing.
• 3D runs are heavier. A detailed, textured, part-segmented generation is one of the more expensive calls in the pack - check the title bar cost before batch-generating a model shelf.

Tripo 3D v3.1 is the "make me a thing I can orbit" node. It won't replace your modeler, but it's an outrageous amount of capability for one node and one API key.