3D
Add Node → Generate → 3D
Feed it pictures of a subject and it builds a textured 3D object out of them. Feed it nothing but a description and it builds one from words alone.
What makes this card different from the rest of the Generate family is what happens after the build. The result isn’t a fixed picture: it’s the object itself, live in the card, and you can turn it with the mouse. Whatever angle you leave it at is the angle that leaves the card — a transparent PNG, drawn again a moment after you stop turning.
So the card makes a thing, and you choose the view of it. Everything downstream just sees a picture.
What’s on the card
| Area | What it does |
|---|---|
| Above the card | The card’s name, then its buttons: help, favourite, Save As, Preview and delete. Save As… saves the 3D model itself. |
| The prompt | Generate new 3D object, and the description if you write one. It’s used only when no picture is wired in. |
| The foot | The model, the Type and Generate. |
| The object | Once there’s a result, the card is the object: drag to turn it, and it stays where you leave it. |
| P | A description, for text to 3D. One only. |
| I | The pictures, one per view of the subject, numbered as you wire them. The first must be the front view — see below. |
| Output | The ring on the right: a 1920 × 1080 PNG of the object as you left it, transparent around it. |
A ring is hollow until something is connected and fills with colour once it is. There’s more on rings in How nodes connect.
The views
Each numbered ring stands for one view of the subject, and the model is told which view each picture is — so ring 2 is a particular view, whether or not you fill it, and a gap left between two rings stays a gap. Plugging a left view into the back ring doesn’t fail: it quietly builds the object wrong.
| Ring | Hunyuan3D V3 | Hunyuan 3D 3.1 Pro |
|---|---|---|
| 1 | Front | Front |
| 2 | Back | Left |
| 3 | Left | Right |
| 4 | Right | Back |
| 5 | — | Top |
| 6 | — | Bottom |
| 7 | — | Left front |
| 8 | — | Right front |
The front view is the one that matters, and it has to be there before any other. The rest sharpen the shape rather than adding to it.
Writing the description
Select the card and the Prompt bar opens on it, with the model and the Type in one row and Generate at the end, with the price of the run under it.
The description is optional once a picture is wired in, and the bar says so.
Which model should you use?
There’s one 3D model per provider, both Tencent’s Hunyuan3D in the cloud, so the choice is made by the provider you’ve connected.
| Model | Provider | Views | Type |
|---|---|---|---|
| Hunyuan3D V3 | WaveSpeed | 4 | Normal · LowPoly · Geometry |
| Hunyuan 3D 3.1 Pro | Runware | 8 | Normal · Geometry |
Prices are indicative and set by the providers, who revise them: see wavespeed.ai or runware.ai for current rates.
Type is what the object is made of. Normal gives it its own surface, painted from your pictures. Geometry returns the bare shape with no materials — the right choice when you’re going to texture it yourself elsewhere. LowPoly, on Hunyuan3D V3, returns a deliberately faceted mesh.
Turning the object
Drag anywhere on the object to turn it. Hover the card and a bar of tools appears along its foot:
- Texture Yes / No — show the object with its own surface, or bare.
- The colour swatches — what a bare object is painted. They appear only with the texture off, because a colour on a textured object would tint the picture printed on it, not colour the thing. Switch the texture back on and the paint is parked, not lost.
- Smaller and Bigger — zoom.
- Reset — back to the view the object opened at.
None of this touches the mesh on disk. What changes is which picture of it leaves the card.
Getting the object out
- Save As… above the card saves the 3D model itself.
- Save View… in the Preview saves the still you’re looking at.
You can also put the object on the canvas on its own: hold the result for half a second, then drag it out, and you get a 3D Object card — no prompt and no settings, just the object in a frame you can turn. It behaves like any other picture on the graph: turn it, and the still on its output ring is drawn again at the new angle.
Dropping a .glb, .gltf or .obj file from the Finder onto the canvas makes one of these too, so you can bring in objects you didn’t build here.
A 3D Object card and a Layer Composer are independent, both ways. Wire an object into a composition and the layer takes its own view of it, seeded from however the object is facing right then. Turning it on the layer doesn’t move the original, and turning the original doesn’t move the layer.