Edit and organize a graph

Add or insert a node

Search the node library by title or operation ID, then add a node to the canvas. Connect an output port to a compatible input. When inserting into an existing branch, use the connection insertion affordance if the candidate node accepts the upstream type and can produce the downstream type.

If no node is compatible, inspect both port types. Converting a mask to labels usually needs Label Connected Components; applying an image as if it were a mask is not an acceptable type conversion.

Undo, duplicate, and delete

Use undo/redo for graph edits and parameter changes. Duplicate a configured node when comparing two parameter choices, then branch both from the same upstream output. Delete a branch only after confirming that no output, tunnel, or table merge still depends on it.

Keep long graphs readable

• Arrange flow from left to right.
• Put alternative methods on parallel branches, not one after another.
• Keep source, QC, and output nodes visually distinct.
• Use Auto structure graph as a starting point, then preserve meaningful parallel alignment.
• Add graph notes at decisions: why a channel was selected, how a threshold was chosen, or what an exclusion means.
• Use graph search to find titles, operation IDs, tunnel names, and batch tags.

Choose a port-label mode

The Port labels setting has three display modes:

Mode	Behavior
Ambiguous only	Default. Show persistent names on nodes with more than one input or output, where choosing the correct port matters.
Show all	Label every input and output port. Useful while learning or reviewing a workflow.
Hide all	Hide persistent labels for the most compact graph; hover and connection behavior are unchanged.

Long names are elided on the card and show their full text in a tooltip. Labels reserve card space but deliberately do not reposition the graph when the mode changes. If existing cards now overlap, the status bar reports how many pairs overlap. Run Auto structure graph to rebuild a size-aware source-to-sink layout, or move those cards manually. Auto structure remains a one-shot, undoable layout edit; port-label visibility does not change the workflow's connections or scientific calculation.

Use tunnels for repeated sources

Named tunnels let one output feed distant nodes without long crossing wires. Create a tunnel from an output, give it a semantic name such as nuclei_labels or red_channel, and attach compatible inputs as subscribers.

Tunnels change graph presentation, not data. Rename a tunnel when its meaning changes; never leave a subscriber attached to a misleading name. Use Tunnels… to focus the source, reveal subscribers, rename, or remove it.

Show alternatives honestly

Two methods being compared should share the same input and run in parallel:

flowchart LR
  A["Prepared input"] --> B["Method A"]
  A --> C["Method B"]
  B --> D["Compare outputs"]
  C --> D

Serial placement (Method A → Method B) means that B operates on A's output; it is not a side-by-side comparison.

Save before structural changes

Save a named workflow checkpoint before replacing a segmentation method, changing axes, or rewiring measurement branches. Version control is preferable for important workflow JSON files because it makes parameter and connection changes reviewable.

If a node is currently being tuned in isolation, a history-backed graph edit (including layout, notes, undo/redo, loading another workflow, or structural changes) first applies the current tuning result. This prevents Cancel tuning from restoring state across two different graph revisions. The temporary isolation boundary itself is session-only and is never saved in the workflow.