Skip to content

Layers, inputs, and bindings

Supported authored schema: 110. The JSON Schema is the exhaustive field reference; the catalog is the named-kind and port reference.

Every layer graph requires exactly one kind: "output" node, including module/group layer records. Connect the final compatible colour output to its input port. Disconnected intermediate nodes are allowed, but Output’s ID must be unique. <layerId>-output is only a convention; any unique ID works. Use the exact ID as toNodeId on its wire. The Stripes example shows a complete generator → Output graph. See Authoring node parameters for complete operator nodes.

layers is the ordered composition. Each layer has a stable id, name, opacity, blendMode, and graph. Legacy layers may retain inputId for compatibility; new cards and shaders reference Sources directly. Its graph has nodes and wires. Every wire names existing node IDs and valid source and destination port IDs. The reachable Output node determines the layer image. Use catalog.layerGraphNodes for card kinds and texture ports; Blend adds in:N ports for its authored inputs. Image operator nodes store a patch; use catalog.imageOperators for the exact operator ID, input/output ports, control IDs, defaults, and structural options. A control’s min/max describes its slider viewport; only hardMin/hardMax expresses an engine limit.

A layer can instead hold a group or embedded project composition. A module retains its nested project snapshot and public parameter mapping; the outer project owns placement and exposure. Preserve stable IDs and namespace dependent resources when copying a layer or module from another project. Studio’s Add layers workflow can perform that remapping.

Texture inputs can reference a layer by identity and view:

{ "kind": "layer-output", "layerId": "layer-a", "output": "composite" }

composite is the composition through that layer, after its opacity and blending, within its composition scope. result is its stack output before opacity and blending; effects can still sample the image beneath the layer. Composite outputs are color images. Result outputs retain the stack’s field type, so a typed field may need Field to Image before a color texture input.

An already-rendered layer supplies the current frame. Sampling itself or a later-rendered layer reads the preceding transport step. Embedded module references remain private to their snapshot; the parent references the module’s public layer output. When copying references, remap layerId while retaining output. Existing Layer Tap nodes retain their historical behavior.

Importing an original producer retains its authored graph and resource dependencies. Its composite is evaluated in the destination composition; copying a layer does not freeze its former Background or implicitly copy every layer below it. Import the desired enclosing composition when those relationships should travel together.

The safest route is Add layers in Studio, which presents source mapping and keeps required dependencies together. For direct file editing, treat the copied layer and everything it references as one unit:

  1. Give copied layers, graph nodes, Sources, Rack parameters, binding graphs, and typed resources new IDs whenever they collide with the destination. Keep an explicit old-ID → new-ID table.
  2. Rewrite references to those IDs in graph wires, inputId fields, binding graph sources and targets, shader/particle/Scene3D card fields, and shared texture ownership. Map exact IDs, not substrings of names.
  3. Copy required resource records and project-owned assets as well as the layer. A simulation view needs its instance, definition and transitive reusable simulation modules. Remap exact simulation output, state, attribute, kernel, helper and operation IDs; keep definition parameter/input IDs, local WGSL names and source text intact. Give asset descriptors unique IDs and importKey paths; update every project-asset: reference to the new asset ID.
  4. Validate the merged workspace, package it, then open it in Studio. A missing dependency is an error to resolve, not a reason to drop a layer’s binding or source silently.

projectInputs defines reusable Sources for image, video, sequence, audio, device, and network content. Cards and shader graph nodes refer to a Source by inputId, not by copying its media URL. backgroundInputId points to the project’s one Background visual Source and must name an existing compatible Project Input. A bundled file uses a project-asset: reference. Device selection and live consent remain runtime concerns. See Sources and Project Inputs.

New video and sequence placements create independent Sources while reusing asset data. Choosing an existing Source shares its playback. Replacing Source content keeps its identity and updates every linked consumer. Older direct media texture references retain their playback behavior when schemas 101–109 are opened in the current Studio.

globalParams are project Rack parameters. bindingGraphs drive target controls through typed node connections and source references. A binding graph’s target identifies an authored destination; its nodes and connections must use valid kinds, ports, and shapes. catalog.bindingGraphNodes and catalog.parameterReferences define the names and fields. macroExposures exposes original authored controls by ParamRef; it does not duplicate their values. See Parameter bindings.

The validator catches missing IDs, malformed references, graph field incompatibilities, and invalid structural targets. It does not invent missing artistic connections. For an executable example, extract Audio-Reactive Glow, inspect its projectInputs, bindingGraphs, and globalParams, then repackage it unchanged before making edits.