# 3D models in Spacelab Turn a photo or a description into a 3D model, then shoot it from any angle as a still or a camera move. Rendering is free. Last updated: 2026-10-27 **Key facts** - Spacelab's 3D Generate node makes a textured 3D model, a GLB file, from a photo or a description, at a flat price per run set by the model and its settings. - Spacelab's 3D Render node is free: it frames the model with a camera and renders a still, or a camera-move video such as a 360° turntable, 2 to 30 seconds at 24 fps. - A 3D model reaches an image, video or Compositor node only through 3D Render, so a 3D canvas runs photo → 3D Generate → 3D Render → an image or video model. - The Image to 3D, 3D Control: Rabbits and 3D to Scene: Futuristic Car templates in Spacelab are ready-made 3D canvases. Web page: https://sloosh.ai/docs/spacelab-3d ## What the 3D nodes do Two nodes work together. 3D Generate makes the model and 3D Render puts a camera on it. Both are under 3D in Add node, or search for them by name. ### 3D Generate Node: 3D Generate (`model3d-gen`) - Inputs: `prompt` (Prompt, text), `image` (Image, image), `back_image` (Back view, image), `left_image` (Left view, image), `right_image` (Right view, image), `top_image` (Top, image), `texture_image` (Texture image, image) - Outputs: `model` (3D Model, model3d) Makes a textured 3D model, a GLB file, from a photo of an object, from a few photos of it from different sides, or from a description. It costs a flat price per run, set by the model and its settings (Texture, Quality, and whether extra views are wired); the node's Run button shows the price before you run. A run takes one to several minutes. Full entry: [3D Generate](https://sloosh.ai/docs/nodes-generate.md#3d-generate). ### 3D Render Node: 3D Render (`model3d-render`) - Inputs: `model` (3D Model, model3d) - Outputs: `image` (Image, image), `video` (Video, video), `model` (3D Model, model3d) Frames the model with a camera and renders it, for free. On the node's Image tab you drag the model to the angle you want and get that shot as a PNG. On its Video tab you pick a move (360° turntable, Half orbit, Dolly in, Crane up or Push past) or set your own keyframes, and get an MP4 at 24 fps, 2 to 30 seconds long. Full entry: [3D Render](https://sloosh.ai/docs/nodes-edit.md#3d-render). ## Which model Pick the model on 3D Generate. Its ports change with the model: a port the model doesn't use is hidden. | Model | Takes | Best for | | --- | --- | --- | | Trellis 2 (default) | One photo | Fast, detailed models. The best value. | | Hunyuan3D 3.1 Pro | A prompt, a photo, or a photo with back, left and right views | Products that must look like the photo. | | Tripo H3.1 | A prompt, a photo, or four views (front, left, back, right) | Clean models with tidy surfaces. | | Rodin Gen-2.5 | A prompt, or up to five photos (front, back, left, right and one more) | Detailed models from several photos. | | Meshy 7.1 | A prompt or up to four photos, plus a texture prompt or texture image | Restyling the model's surface. | | Hi3D 3.0 | A photo, or a photo with back, left and right views | The most faithful product models. The slowest. | | SAM 3D Objects | A real photo and the object's name | Lifting one object out of a busy photo. | ## Your first 3D model 1. Open a canvas and add an Image Input. Upload a clear photo of one object, such as a sneaker, a bottle or a lamp. 2. Add 3D Generate, keep Trellis 2, and wire the photo into Image. 3. Press Run on 3D Generate. After a few minutes the node shows the model: drag to turn it, and Download saves the GLB. 4. Add 3D Render and wire 3D Generate's 3D Model into it. On the Image tab, drag the model to the angle you want, and set Background to White. 5. Press Run on 3D Render. Its Image is your shot, ready for an Image Model reference or a Video node's First frame. 6. For a video, open the Video tab, pick 360° turntable under Moves, wire Video into a Download and run 3D Render again. It's free, so re-shoot as often as you like. ## Use case: a turntable video from one product photo Why: a 360° turntable shows every side of a product, from one photo. Image models draw the missing sides first, so the 3D model has more to go on, and the camera move is exact, which a video model can't promise. **Wiring** ```text Product photo → Image Model (back view) → 3D Generate: Back view Product photo → Image Model (side view) → 3D Generate: Left view Product photo ─────────────────────────→ 3D Generate: Image (Rodin Gen-2.5) 3D Generate → 3D Render (360° turntable, White) → Download ``` Canvas diagram: A Product photo feeds two Image Models that draw its back and side; the photo and both views feed 3D Generate, whose model goes to 3D Render, whose turntable Video goes to a Download. Nodes: - `photo`: Product photo (Image Input). Inputs: none. Outputs: `image` (Image, image). - `backp`: Back view (Prompt). Inputs: none. Outputs: `text` (Prompt, text). - `sidep`: Side view (Prompt). Inputs: none. Outputs: `text` (Prompt, text). - `back`: Back view (Image Model). Inputs: `prompt` (Prompt, text), `aspect_ratio` (Aspect ratio, setting), `reference_images__0` (Image 1, image). Outputs: `image` (Image, image), `images` (Images, image). - `side`: Side view (Image Model). Inputs: `prompt` (Prompt, text), `aspect_ratio` (Aspect ratio, setting), `reference_images__0` (Image 1, image). Outputs: `image` (Image, image), `images` (Images, image). - `model`: 3D model (3D Generate). Inputs: `prompt` (Prompt, text), `image` (Image, image), `back_image` (Back view, image), `left_image` (Left view, image), `right_image` (Right view, image), `top_image` (Top, image), `texture_image` (Texture image, image). Outputs: `model` (3D Model, model3d). - `render`: Turntable (3D Render). Inputs: `model` (3D Model, model3d). Outputs: `image` (Image, image), `video` (Video, video), `model` (3D Model, model3d). - `download`: Download. Inputs: `images__0` (Image 1, image), `videos__0` (Video 1, video). Outputs: none. Wires (node.output -> node.input): - photo.image -> back.reference_images__0 - backp.text -> back.prompt - photo.image -> side.reference_images__0 - sidep.text -> side.prompt - photo.image -> model.image - back.image -> model.back_image - side.image -> model.left_image - model.model -> render.model - render.video -> download.videos__0 1. Add an Image Input named Product photo, with a clear photo of the product on a plain background. 2. Add two Prompts and two Image Models with Nano Banana Pro, and wire the photo into each one's first reference image. One prompt: `Show the product in the attached photo from behind, a back view, on the same plain background. Keep its shape, materials, colours and logo exactly the same.` The other says `from its left side, turned 90 degrees` instead. 3. Add 3D Generate and pick Rodin Gen-2.5. Wire the photo into Image, the back view into Back view and the side view into Left view. 4. Add 3D Render. On its Video tab pick 360° turntable under Moves, set Background to White, and wire Video into a Download. 5. Run the canvas. The edits and the model cost credits; the render is free, so try other moves without paying again. The Image to 3D template does the same with three drawn views (top, side and back), then stages a still of the model in an ad scene with Nano Banana Pro. ## Use case: choose the camera angle, then restyle Why: an image model guesses an angle you describe. Turn the model in 3D Render and you get exactly the angle you want; an Image Model then paints the scene and the style around that shot and keeps the angle. **Wiring** ```text Character → 3D Generate (Hunyuan3D 3.1 Pro) → 3D Render (Image tab: your angle) 3D Render: Image → Image Model: Image 1 Character ──────────────────────→ Image Model: Image 2 Prompt (Scene) ─────────────────→ Image Model: prompt ``` Canvas diagram: A Character photo feeds 3D Generate; 3D Render shoots the model at a chosen angle; an Image Model restyles that shot, with the photo as a second reference and a Scene prompt. Nodes: - `photo`: Character (Image Input). Inputs: none. Outputs: `image` (Image, image). - `model`: 3D model (3D Generate). Inputs: `prompt` (Prompt, text), `image` (Image, image), `back_image` (Back view, image), `left_image` (Left view, image), `right_image` (Right view, image), `top_image` (Top, image), `texture_image` (Texture image, image). Outputs: `model` (3D Model, model3d). - `view`: Camera angle (3D Render). Inputs: `model` (3D Model, model3d). Outputs: `image` (Image, image), `video` (Video, video), `model` (3D Model, model3d). - `scene`: Scene (Prompt). Inputs: none. Outputs: `text` (Prompt, text). - `edit`: In the scene (Image Model). Inputs: `prompt` (Prompt, text), `aspect_ratio` (Aspect ratio, setting), `reference_images__0` (Image 1, image), `reference_images__1` (Image 2, image). Outputs: `image` (Image, image), `images` (Images, image). Wires (node.output -> node.input): - photo.image -> model.image - model.model -> view.model - view.image -> edit.reference_images__0 - photo.image -> edit.reference_images__1 - scene.text -> edit.prompt 1. Wire an Image Input with your character, product or sketch into 3D Generate, pick Hunyuan3D 3.1 Pro and run it. 2. Add 3D Render after it. On the Image tab drag the model to the angle you want: low and from the side, from above, from behind. Lock keeps the camera put while you move the node. Set Background to White. 3. Add an Image Model with Nano Banana Pro. Wire the render's Image into Image 1 and the original photo into Image 2. 4. Prompt: `Place this character on a grassy hill. Keep its look from the second image and its angle from the first.` 5. Run it. For another angle, turn the view, then press Run on 3D Render and on the Image Model. The model isn't made again. The 3D Control: Rabbits template starts from a sketch: GPT Image turns it into a render, Nano Banana Pro draws its right side, back and left side, Rodin builds the model from all four, and three Nano Banana Pro edits place the chosen shot on a hill, change the sky and add a second rabbit. ## Use case: compose 3D objects into a scene, then animate Why: each object keeps its exact shape and angle, you place them yourself, and the video model starts from a frame you have already approved. **Wiring** ```text Object 1 → 3D Generate → 3D Render (Transparent) → Compositor: Layer 1 Object 2 → 3D Generate → 3D Render (Transparent) → Compositor: Layer 2 Prompt (Backdrop) → Image Model (16:9) ──────────→ Compositor: Background Compositor: Image → Video: First frame ← Prompt (Motion) ``` Canvas diagram: Two objects each go through 3D Generate and 3D Render on a transparent background into a Compositor's layers, over a backdrop from an Image Model; the Compositor's image is a Video node's first frame. Nodes: - `obj1`: Object 1 (Image Input). Inputs: none. Outputs: `image` (Image, image). - `gen1`: 3D model 1 (3D Generate). Inputs: `prompt` (Prompt, text), `image` (Image, image), `back_image` (Back view, image), `left_image` (Left view, image), `right_image` (Right view, image), `top_image` (Top, image), `texture_image` (Texture image, image). Outputs: `model` (3D Model, model3d). - `r1`: Render 1 (3D Render). Inputs: `model` (3D Model, model3d). Outputs: `image` (Image, image), `video` (Video, video), `model` (3D Model, model3d). - `obj2`: Object 2 (Image Input). Inputs: none. Outputs: `image` (Image, image). - `gen2`: 3D model 2 (3D Generate). Inputs: `prompt` (Prompt, text), `image` (Image, image), `back_image` (Back view, image), `left_image` (Left view, image), `right_image` (Right view, image), `top_image` (Top, image), `texture_image` (Texture image, image). Outputs: `model` (3D Model, model3d). - `r2`: Render 2 (3D Render). Inputs: `model` (3D Model, model3d). Outputs: `image` (Image, image), `video` (Video, video), `model` (3D Model, model3d). - `bgp`: Backdrop (Prompt). Inputs: none. Outputs: `text` (Prompt, text). - `bg`: Backdrop image (Image Model). Inputs: `prompt` (Prompt, text), `aspect_ratio` (Aspect ratio, setting), `reference_images__0` (Image 1, image). Outputs: `image` (Image, image), `images` (Images, image). - `comp`: Scene (Compositor). Inputs: `background` (Background, media), `layers__0` (Layer 1, media), `layers__1` (Layer 2, media). Outputs: `image` (Image, image), `video` (Video, video). - `motion`: Motion (Prompt). Inputs: none. Outputs: `text` (Prompt, text). - `video`: Video. Inputs: `prompt` (Prompt, text), `first_frame` (First frame, image), `last_frame` (Last frame, image), `aspect_ratio` (Aspect ratio, setting), `duration_seconds` (Duration (s), setting). Outputs: `video` (Result, video). Wires (node.output -> node.input): - obj1.image -> gen1.image - gen1.model -> r1.model - obj2.image -> gen2.image - gen2.model -> r2.model - bgp.text -> bg.prompt - bg.image -> comp.background - r1.image -> comp.layers__0 - r2.image -> comp.layers__1 - comp.image -> video.first_frame - motion.text -> video.prompt 1. Make a 3D model of each object as in your first model, and give each its own 3D Render with Background set to Transparent. Frame them at angles that suit one camera. 2. Make a backdrop with an Image Model at 16:9 from a Prompt, or upload one with an Image Input. 3. Add a Compositor. Wire the backdrop into Background and each render into a Layer slot. Press Edit to place, size and order the objects, and add a headline if you like. 4. Add a Video node with a Prompt for the motion, such as `slow push in, the car's headlights turn on`, and wire the Compositor's Image into First frame. Run it. 5. For a title over the clip, wire the clip into a second Compositor, tick Timeline mode in its editor and add the text on top. The 3D to Scene: Futuristic Car template builds one model from four photos of a car, shoots it from four angles, turns each shot into a car photo, sets it in a generated scene and animates it, then titles each clip with a Compositor on a timeline. ## Tips - Source photos: one object, whole and in focus, on a plain background, in even light without hard shadows. Cut a busy photo out with BG Remove first, or use SAM 3D Objects and name the object. - More sides, a better back: a model only knows what it sees. Wire back and side views when the model takes them, drawn by an Image Model from your one photo if that is all you have, in the same light and on the same background. - GLB limits: a model is a GLB file and travels only along 3D Model wires, into 3D Render. Image, video and Compositor nodes can't take it. 3D Render takes models up to 150 MB. - Small print and logos can come out soft on a model. Give the image model after 3D Render the original photo as a second reference to bring them back. - 3D Render or an image model? Use 3D Render when the angle or the camera move must be exact, or the same object must match across shots: it's free and does the same thing every time. Use an image model when the look matters more than the shape: light, materials, a whole scene. Most 3D canvases use both, 3D Render for the angle and an Image Model for the look. - A transparent background applies to 3D Render's image only. Its video is always on white. ## Templates Three ready-made 3D canvases are in Templates, in the top-left toolbar of a canvas: Image to 3D (one product photo to a 3D model and a staged ad), 3D Control: Rabbits (a sketch to a render to a 3D model, then pick the camera angle and edit the shot) and 3D to Scene: Futuristic Car (four photos of a car to one 3D model, four scenes and four titled clips). ## Frequently asked questions ### How much does a 3D model cost in Spacelab? Spacelab's 3D Generate costs a flat price per run, set by the model and its settings, such as Texture, Quality and extra views; its Run button shows the price before you run. 3D Render, which shoots the model as an image or a video, is free, so you can re-frame a model as often as you like. ### Can I make a 360° product video in Spacelab? Yes. Make a model with Spacelab's 3D Generate, then on 3D Render's Video tab pick 360° turntable under Moves and wire its Video output onward. The video is an MP4 at 24 fps, 2 to 30 seconds long, on a white background, and rendering it is free. ### Can I use a 3D model in an image or video model in Spacelab? Yes, through 3D Render. A 3D model is a GLB file, which image and video nodes can't read, so wire 3D Generate into 3D Render and use its Image (one exact angle) or its Video (a camera move), for example as an Image Model's reference image or a Video node's First frame. ### Which 3D model should I pick in Spacelab? Start with Trellis 2, the default: one photo, fast and the best value. Use Hunyuan3D 3.1 Pro or Hi3D 3.0 for products that must match the photo, Rodin Gen-2.5 for detailed models from up to five photos, Tripo H3.1 or Meshy 7.1 to make a model from a description, and SAM 3D Objects to lift one named object out of a real photo.