Scene Builder

PatchWork's canvas (the graph editor view) is what we've called the Scene Builder — the place where workflow nodes are arranged and wired. This page covers building workflows visually inside PatchWork, instead of having them generated by our pipeline.

When you'd build manually

Our pipeline (Manager + Script Writer + Visual Planner + PatchWork Importer) auto-generates .nbflow files for you. The Scene Builder is for the cases when you don't want that automation:

• Experimenting with a new node-type combination the pipeline doesn't yet support
• Building a one-off workflow that doesn't fit the standard template (a single-scene experiment, an unusual format)
• Debugging a workflow — sometimes easier to rebuild a problematic section by hand than to find what's wrong in the existing graph
• Learning PatchWork's mechanics — manually wiring 10 nodes teaches you more than reading 10 docs about them

The canvas basics

When you open a project, the canvas shows:

• The toolbar at the top — pan / zoom / select / connect / delete
• The node palette on the right — drag node types onto the canvas
• The properties panel on the right when a node is selected — edit that node's settings
• The node graph itself — boxes (nodes) connected by lines (links)

Building a workflow from scratch

The typical sequence:

1. Pick a template OR start from blank (see Chapter 4 — Templates)
2. Add prompt nodes for the dialogue, scene composition, motion specs
3. Add Media nodes for the reference images (avatar, product)
4. Add NanobananaAPI nodes for image gens (one per scene that needs an image)
5. Add Veo3 nodes for video gens (one per scene that needs a clip)
6. Add Approve nodes between each gen and its downstream consumer
7. Wire the connections — drag from one node's output port to another's input port

The architecture patterns from Chapter 11 still apply when building manually:

• Dynamic Prompt feeds Template feeds Gen node for per-scene variable content
• One generation node per (template + input media) — let the dynamic fan out, don't multiply gen nodes
• Approve nodes between each gen and its consumer — gives you a candidate-picking step

Properties panel — per-node settings

Click any node to edit its properties:

Node type	Key properties
Prompt	text, mode (plain / template / dynamic), dynamicRows, variableName
Media	mode (reference / output), imageData (R2 URL)
NanobananaAPI	model (always nano_banana_2), outputCount, aspectRatio, negativePrompt
Veo3	model, outputCount, negativePrompt, resolution, seed, timeout
Approve	mostly UI state — what's been picked, gallery position

If you're hand-building, the pre-generation sanity check is critical. Hand-built workflows often have subtle schema issues (missing inputs, mismatched variableNames, stale link refs).

Save vs. Export

Two operations to know:

Save

Persists your current canvas state to PatchWork's internal storage (in-browser, plus any cloud sync if you've configured it). The workflow is recoverable across sessions.

Export

Writes the workflow out as a .nbflow file you can download. This is the file you'd hand to the Generation Runner, share via Drive, or commit to git.

If you're hand-building a workflow you intend to use repeatedly, export it — that gives you a portable file. PatchWork's internal storage is browser-local and not shareable.

When NOT to use the Scene Builder

For our normal pipeline workflow:

• The Manager + agents produce a .nbflow for you — don't rebuild it by hand
• The pipeline ensures architecture patterns are followed; hand-built workflows often violate them
• Hand-built workflows are harder to variant later (variants assume the auto-built structure)

Use the Scene Builder for exploration and experimentation. Use the pipeline for production.

When you're ready

→ Next: Compliance Audit — the final check before going live.

(For PatchWork's built-in starting templates, see Chapter 4 — Templates.)