Use the 3D editor’s scoped AI to add, remove, transform, group, and reparent scene nodes; configure materials, lights, environment, cameras, particles, and post effects; and expose selected properties for timeline animation.
Where: Select a Scene3D element → full-screen 3D editor → AI
Before you start
- A Scene3D element must be selected; AI scene edits target the active selected scene.
Key ideas
- Scoped scene operations: The assistant edits the selected Scene3D graph rather than generating an unrelated page or opaque replacement.
- Ordered changes: A request can translate into ordered scene operations—create a node, transform it, apply material/light/post settings, then place it in the hierarchy.
- Animation bridge: Exposing a numeric or color property promotes it to a normal f0 field that can be keyframed, scroll-scrubbed, or driven by interactions.
Steps
- Select the intended Scene3D element and open its full-screen editor before opening the 3D AI surface.
- Describe the exact nodes to add or modify, their hierarchy, visual result, and what existing scene content must remain unchanged.
- Ask for the scene change in a bounded stage, then inspect the outliner and viewport before requesting the next stage.
- Name any material, lighting, camera, particle/replicator, effector, environment, or post-FX requirements explicitly.
- Ask to expose only the values you intend to animate, then open the f0 timeline/interactions surface and test those fields.
- Choose the camera framing, inspect existing keys, apply a named move with duration and amount, and scrub before rendering. Use the detailed cinematic guide for preset names, rail behavior, and backdrop boundaries.
Useful request boundaries
- Objects and hierarchy.
- Transforms and imported-model placement.
- PBR, layered, procedural, or shader materials.
- Lights, HDRI/environment, exposure, cameras, and post effects.
- Particles, replicators, and effectors.
- Properties to expose to the timeline.
Review after each stage
Check object names, parent/child relationships, active camera, material assignment, lighting, viewport performance, and exposed field names before asking the assistant to continue.
Cinematic camera and motion workflow
Camera Lens presets and Focal length express the existing field of view in millimetres. The motion library writes real editable cinematic keyframes relative to the camera or object’s current pose. Guide paths support camera rails without adding a visible swept tube. The image backdrop is a modeling aid that stays out of cinematic playback and delivered output.
- Inspect animation before changing existing motion. keyframe_scene and apply_motion_preset replace the channels they touch by default; choose merge only deliberately.
- Camera presets require cameraId; object presets require nodeId. These are internal scene IDs, not the Flowgen node ID or HTML element ID.
- For MCP, open the node editor, select the Scene3D host element, and enumerate document tools. The selected or sole scene is pinned by the bridge; a caller cannot redirect the operation with an id argument.
- inspect_scene reads current scene values and animation without writing; animate_scene_field controls exposed fields via scroll, mouse, hover, or load instead of cinematic time.
Tips
- Request named groups/objects so later edits have stable targets.
- Expose the smallest useful set of fields; a focused timeline is easier to animate than dozens of low-level controls.
Troubleshooting
The assistant reports that no scene is selected
Close unrelated panels, select the Scene3D element on the canvas, reopen its full-screen editor, and retry.
The right object was created but is not visible
Inspect hierarchy, transform/scale, active camera framing, material opacity, light/environment, and whether the node was parented under a hidden or transformed group.