Marble 3D AI on Martini
Marble 3D AI generates explorable 3D scenes from a text prompt or reference image, producing pre-visualization environments you can use as backgrounds, camera reference, and world-building drafts inside Martini. It is best treated as a spatial sketchpad for narrative work, not as a production asset exporter, since outputs are scene previews rather than clean meshes.
What it creates
Marble 3D AI creates navigable 3D scenes — interior rooms, outdoor landscapes, set pieces, and atmospheric environments — from a short text prompt or a reference image. Inside Martini, the output appears as a preview you can rotate, frame, and screenshot for downstream nodes. Think of it as a spatial mood board: a way to lock in mood, lighting direction, scale, and camera language before you commit to a final render or shoot.
Beyond static screenshots, Marble scenes give you a consistent world to plan from. Re-shoot the same environment from multiple angles, sketch shot lists, plan blocking, or pull single frames to feed an image model for hero stills. Most projects use Marble as the planning layer that sits above their video and image generation — the spine that keeps the look of a sequence coherent across many separate AI calls.
The output is best understood as a fast pre-viz scene rather than a polished, exportable asset. You will not ship the raw Marble render to a final timeline, and you should not expect a clean topology mesh — but you will get spatial reference that is dramatically faster than building anything comparable by hand in a traditional 3D tool.
Inputs and outputs
Marble accepts a text prompt describing the scene, an optional reference image to anchor look and composition, and parameters that control style and scale. On Martini, you can wire reference images straight from an Image node — a Nano Banana 2 still, a Flux frame, or a photograph you uploaded — and connect the prompt input to a Text node so you can iterate on language without rebuilding the workflow. The output is a viewable 3D scene preview rendered in the canvas; from there you can capture frames, feed those frames into image or video models, or use the scene as continuous reference while planning a shot list. Treat it as a spatial preview layer, not a downloadable mesh pipeline.
Best workflows
- •Pre-visualization for short films and ads — block out a set, decide camera angles, and lock in lighting mood before you commit to expensive final renders.
- •World planning for episodic or branching narratives — keep one shared 3D scene as a canonical reference so every shot in the sequence stays spatially coherent.
- •Storyboard generation — pull multiple frames from one Marble scene and pass them through Sora 2 or Kling 3 to animate sequential shots that share geography.
- •Concept art and pitch decks — generate a reference scene, screenshot the strongest angle, and feed it into Nano Banana 2 to upgrade the look for a presentation still.
- •Set design exploration — iterate on prompt variations to compare interior layouts, props, and atmospheres side-by-side on the canvas before committing.
- •Shot reference for VFX and compositing — drop the scene into a Martini board to plan parallax, foreground/background separation, and depth cues for a hybrid live-action + AI pipeline.
How to use it in Martini
- 1
Open a Martini canvas and add a 3D node configured to use Marble. Optionally drop in a Text node for the scene description and an Image node for a visual reference — this lets you swap prompts and references without rebuilding the graph each time.
- 2
Wire the Text node into the prompt input on the 3D node. If you are working from an image reference, connect a Nano Banana 2 or Flux Image node into the reference input so Marble anchors mood, palette, and rough composition to that frame.
- 3
Run the node and wait for the scene preview. Once it loads, frame the angle you want and capture a still — you can drop this still into another Image node for retouching, or feed it directly into a Video node like Sora 2 or Kling 3 as the first-frame reference.
- 4
Build out your shot list by duplicating the screenshot pattern: capture five or six angles from the same scene, label each on the canvas, and connect each to its own video generation node. The whole sequence will share one consistent world even though every clip is generated independently.
- 5
When you have the layout you want, save the project. The Marble scene stays in place as your spatial spine, and you can keep generating new shots, variations, and re-frames against it for as long as the project lives.
Pair with image / video models
Limitations
- !Output is a viewable scene preview — not a clean .obj, .fbx, .glb, or USD mesh. Do not plan to drop the result into Blender, Maya, Unreal, or a game engine pipeline.
- !Geometric fidelity is pre-viz quality. Expect approximations of objects and materials, not production-grade topology, UVs, or PBR materials.
- !Re-running the same prompt will not produce a pixel-identical scene; treat each generation as a draft, and screenshot the angles you like before iterating.
- !Complex prompts with many specific props, characters, or interactions tend to lose detail. Keep prompts focused on environment, mood, and a few hero elements.
- !The 3D node may not yet appear in every Martini documentation page — if you do not see it in the node picker, check the latest release notes or contact support before relying on it for a deadline.
Related features
Related docs
Frequently asked questions
Can I export a Marble scene as a .obj or .fbx mesh?
No. Marble outputs a viewable 3D scene preview inside Martini — not a clean exportable mesh. Use it for pre-viz, reference, and world planning, then capture frames to feed into image and video models for final delivery.
How does Marble compare to traditional 3D tools like Blender or Unreal?
Marble is a fast pre-viz layer for ideating worlds in minutes from text or images. It does not replace a production 3D pipeline — it sits in front of one. For mesh work, lighting bakes, or game engine integration you still need a traditional tool.
What kinds of prompts work best for Marble inside Martini?
Environment-focused prompts with a clear mood and one or two hero elements work best. Describe the location, time of day, lighting, and atmosphere, then add any signature props. Avoid asking for many distinct characters or fine-grained interactions.
Do I have to use a reference image, or can I generate from text only?
Both work. A text-only prompt is the fastest way to start. Adding an image reference — for example a Nano Banana 2 still — gives you tighter control over palette and composition, which matters when you need scene continuity across many shots.
Why pair Marble with Sora 2 or Kling 3 instead of using video alone?
Video models generate one clip at a time and rarely keep a world coherent across many shots on their own. A shared Marble scene gives every clip a consistent geography to anchor to, so a five-shot sequence reads like one set instead of five different sets.
Can I keep iterating on the same Marble scene later?
Yes. Save the Martini project and the scene preview stays in your canvas. You can capture more angles, swap reference images, or branch off variants without losing the original scene as your reference.
Ready to build with Marble 3D AI?
Open Martini, drop a 3D node, and chain it into your image and video pipeline. No GPU required.
Get started in dashboard