Build and inspect connected Flowgen node workflows: add nodes, configure the card and sidebar, route typed source noodles to exact fields, switch between card/graph views, and run the smallest useful branch.
Where: Dashboard → open a Flowgen project → canvas.
Key ideas
- A node represents an operation or a source: Nodes provide typed fields and outputs for generation, editing, transformations, data, or editor surfaces. A useful flow makes each source and dependency visible rather than hiding content in one large prompt.
- A noodle carries a typed connection: A connection runs from a source output socket to a target input field. Drop it on the specific field when a card has several compatible input rows.
- Card mode and graph mode share one canvas: Card mode hides the connection noodles for a cleaner card presentation. The node graph shows the full wiring. Per-node Show on canvas controls only whether a card appears in the compact view.
- Run scope follows dependencies: A node run resolves connected upstream values. Run one step while authoring; use Run flow after the graph, routes, approvals, and destinations have been reviewed.
Steps
- Open a Flowgen project and use Add to search or browse the node catalog. Use Discover to find a ready-made flow when you want a reusable starting graph.
- Place the source and operation nodes you need. Give important cards clear names such as “Brief”, “Product reference”, or “Hero image”.
- Select a node to see its actions and properties sidebar. Read the node card and sidebar together; the card emphasizes common inputs/actions while the sidebar exposes the full supported settings.
- Keep flow-specific prompt wording in a connected Text source. When values already live in the project’s linked Brand, bind the correct compatible Brand field; use typed media/reference sockets for files and images.
- Switch to node-graph view. Drag from a source output to the exact target field, or drop a noodle onto the field. For multiple sources, confirm the field chips and order before running.
- Turn on Card mode to hide noodles while scanning cards, then turn it off again to inspect or change graph wiring. Show on canvas hides a node’s compact card without removing its node or connections.
- Use Quick connect or the node’s Add source action when the canvas offers a guided path; inspect the resulting source chip and edge to confirm the destination field.
- For row-by-row workflows, connect an Items/List node to List Iterator and collect typed results in a Results List. Do not use deprecated Group Repeater, loop-enabled Groups, or Batch to author new repetition.
- Run the single node or group you are testing. Review the result, warnings, selected output/version, and downstream stale inputs before running later stages.
- Open an editor-backed node only when the result needs direct visual/timeline editing. Save or resolve the editor state, then rerun downstream work if its input changed.
- Use Run flow for the whole graph only after checking required inputs, route conditions, approval pauses, external writes, and any paid model quote.
Read a card from top to bottom
- Identity and model/status identify the node’s purpose and selected generation model.
- Input fields display compatible source chips and actions to add/bind a source.
- Output sections expose handles/results that can feed later nodes.
- Run and editor actions are node-specific; not every card is runnable.
The right properties panel
- Select a node to open its properties panel.
- Use the panel for exact model, prompt-source, generation, editor, data, or node-specific configuration not shown on the compact card.
- Set concrete properties in their named controls; prose belongs in a connected Text source and structured Brand values belong in live Brand bindings.
What a wire means
- Source output type must be accepted by the target input.
- A prompt Text connection is source content, not a hidden inline card value.
- A Brand field connection reads the linked Brain record and follows its later edits.
- A reference connection guides appearance; a placement or edit connection has a different job even when both carry an image.
Tips
- Use graph mode whenever provenance matters or a node has multiple same-type inputs.
- Break a long workflow into named groups; groups organize and expose reusable inputs/outputs but do not replace List Iterator for repeated rows.
- Run a small test branch before creating a large batch or publishing an automation.
Limitations and important notes
- Valid fields, output handles, run semantics, and editor actions depend on the selected node type.
- Card mode changes presentation only; it does not disable the graph.
- The existence of a Run button does not mean every node runs without prerequisites or that a paid operation has been approved.
Troubleshooting
A node does not receive the prompt/reference I expected.
Switch to graph view, inspect the destination field chips, and verify the noodle terminates on the intended named input. Check socket types and source order.
Connections disappeared when I cleaned up the canvas.
Turn off Card mode to show the graph. If only one card is absent, enable its Show on canvas checkbox; neither control deletes the underlying node or edge.
The whole flow fails because one step is incomplete.
Run the smallest node or group first and resolve the missing input/model/route there before using Run flow.