Tripo: Text to Model

Tripo: Text to Model is the closest thing to "type a sentence, get a riggable 3D asset" that ComfyUI has, and it's the node most people start with when they wander into the 3D side of the graph. Prompt in, textured GLB out. That's the whole pitch, and for props, weapons, furniture, and "I need a placeholder that doesn't look like a placeholder" it delivers.

It's a Partner Node, which means it's built right into ComfyUI core - nothing to install. But it's not running on your GPU. Your prompt goes up to Tripo's API (through Comfy's servers), a job spins up on their side, and ComfyUI polls until it's done. You don't need a Tripo account or API key of your own; you log in with a ComfyUI Account and pay out of prepaid credits. The node shows a price badge estimating the run before you execute, and the estimate moves when you flip texture or quality settings. "Free" is not a word that applies here.

The inputs that actually matter

Most of the settings are optional, and the defaults are sane, so this is a short list:

• prompt - required, multiline. This is the whole job. Be specific about shape and style; "steampunk mechanical dolphin" beats "dolphin."
• model_version - defaults to v2.5, which is the well-rounded choice. The older v1.4 matters for one reason: it's the only version that supports the Tripo: Refine Draft model polish step later. v3.x adds detailed geometry quality (at extra credits) if you want the fancy pass.
• texture and pbr - both default on. Off = gray mesh and fewer credits. PBR gives you the material maps; keep it on unless you're just blocking out shapes.
• style - a dropdown of preset art styles, None by default. Set it if you want a specific look, leave it alone otherwise.
• Everything else - texture_quality, face_limit, quad, the three seeds, geometry_quality - lives in the advanced section and can wait until you know you need it. quad (quadrilateral output) and face_limit (cap on triangles) are the two you'll reach for when a mesh is destined for a game engine.

What comes out

Three outputs: a model_file string (backwards-compat), a model task_id (the handle every Tripo post-processing node eats), and the GLB itself. Wire the GLB into the built-in Preview 3D & Animation node to eyeball it, or Save 3D Model to write it out - files also land in ComfyUI/output/ automatically.

The task_id is where the workflow gets interesting. Instead of saving, chain it: Tripo: Rig model on the task_id, then Tripo: Retarget rigged model to animate it, or Tripo: Convert model to get an FBX for Unreal. One prompt becomes a whole asset pipeline.

Watch out for

Two recurring traps. First, don't expect strict reproducibility: the seeds exist, but the generation is stochastic enough that you'll get a new sibling on each run. Second, credits - the advanced quality settings quietly double or triple a run's cost, and the price badge is the only place you'll see that coming. Read it before you click run.

The honest verdict: it's the one I'd reach for first in this whole family. Quality is genuinely good for a text prompt, and because it's cloud-based, your 4GB GPU is irrelevant. The catch is the meter running.