> ## Documentation Index
> Fetch the complete documentation index at: https://comfytv.org/llms.txt
> Use this file to discover all available pages before exploring further.

# 3D Model Stage

> Generate a 3D model (GLB) from a prompt, reference images, and/or existing models—your text/image-to-3D generation stage.

<Frame caption="3D Model Stage · screenshot coming soon">
  <img src="https://mintcdn.com/comfytv/dMkMxJDZdIo4YzAU/images/nodes/_placeholder.svg?fit=max&auto=format&n=dMkMxJDZdIo4YzAU&q=85&s=b54171099e5b02d4313fe581171ebbcc" alt="3D Model Stage" width="800" height="450" data-path="images/nodes/_placeholder.svg" />
</Frame>

## What this node does

**3D Model Stage** is a **Generate** stage (`ComfyTV/Generate`). You give it a prompt (and optionally reference text, images, or models), pick a backend **workflow**, and click **▶ Run**. The wrapped workflow—which must end in a **SaveGLB** (or compatible) node—produces a `COMFYTV_MODEL`, and the node's in-body 3D preview lets you orbit the result.

This node **has a ▶ Run** (it invokes a real generation backend / GPU). Alongside the model it also emits a `COMFYTV_IMAGE`: a snapshot captured from the preview viewport, useful downstream as a thumbnail or for image-based follow-up stages.

Media flows as ComfyTV project snapshots, not native ComfyUI tensors. To interoperate with native ComfyUI 3D nodes, insert a **Bridge** — see [bridges.md](https://github.com/jtydhr88/ComfyTV/blob/main/docs/bridges.md).

## When to use it

* Turn a text prompt into a first-pass 3D asset (text-to-3D).
* Turn one or more reference images into a mesh (image-to-3D).
* Re-generate or vary from existing models by feeding them back in as references.

## Parameters

### workflow

Combo selecting which backend 3D-generation workflow to invoke on Run. Options come from the installed `model`-kind workflows; the workflow must end in a **SaveGLB** (or compatible) node. Defaults to the configured default for the `model` kind.

### main\_prompt

The primary prompt describing the model you want. Upstream text inputs are treated as additional context and combined with this. Hidden text field edited in the node body (multiline).

### texts (autogrow)

Optional `COMFYTV_TEXT` inputs (grows up to 4 slots) that add extra textual context to the prompt.

### images (autogrow)

Optional `COMFYTV_IMAGE` inputs (grows up to 4 slots)—reference images for image-to-3D generation.

### models (autogrow)

Optional `COMFYTV_MODEL` inputs (grows up to 4 slots)—existing models fed back in as references.

### captured\_image (internal)

Hidden. The `/view?` URL of the latest preview-viewport snapshot, written by the 3D preview in the node body; it becomes the **image** output.

### custom\_params / project\_id / parent\_output\_id / force\_run\_token (internal)

Hidden plumbing—user-defined parameter attachments, project scoping, lineage, and the Run trigger. You rarely touch these.

## Outputs

| Output    | Type            | Meaning                                     |
| --------- | --------------- | ------------------------------------------- |
| **model** | `COMFYTV_MODEL` | The generated 3D model (GLB) snapshot       |
| **image** | `COMFYTV_IMAGE` | Snapshot captured from the preview viewport |

## Tips

* The selected **workflow** must terminate in a SaveGLB-compatible node, or the stage has nothing to emit.
* For image-to-3D, wire clean reference images into the **images** slots; for text-to-3D, lean on **main\_prompt**.
* Orbit the in-body preview to frame a good angle before relying on the captured **image** output.

## Related nodes

* **Load 3D Model** — bring an existing GLB/GLTF/FBX/OBJ file into the pipeline instead of generating.
* **Load 3D Model from Asset** — load a model from the project asset library.
* **Material Stage** — author PBR materials to bind onto a loaded model.