Skip to main content

Your First Graph

This walkthrough builds a tiny pipeline that feeds an explicit tensor through a couple of operations — enough to learn the Start node execution model that every CodefyUI graph relies on.

In Basic node category mode the sidebar hides Start, Reshape, Softmax and Print; add them by double-clicking the canvas and searching (see The sidebar).

1. Add an input​

Drag a TensorInput node (Data category) onto the canvas. Its shape defaults to 1,4,4, so it holds 16 values. Set its value_mode to explicit, then type the numbers you want the pipeline to see into the inline grid, or fill it with Fill 0, Fill 1 or Random (see The config panel).

2. Wire up some operations​

Connect it through a chain of tensor-op nodes. This one flattens the 16 values into one row and turns them into probabilities:

TensorInput → Reshape → Softmax → Print

Drag from each output port to the next input port. The edges validate types as you connect.

Set Reshape's shape to 1,16. A new shape must hold the same number of elements as the tensor it receives, and Reshape's default, -1,784, needs a multiple of 784, so the run would stop at Reshape. Leave Softmax's dim at -1: the 16 probabilities then add up to 1.

3. Add a Start node​

:::warning Every graph needs a Start node Drag a Start node onto the canvas and connect its trigger output (the diamond handle on the right side) to the first node you want executed — typically the TensorInput. Drop the wire anywhere on that node.

Without a Start → first-node trigger edge, the graph is treated as a draft and Run rejects it with an error toast: "No entry points defined. Drag a Start node from the palette and connect it to the node you want to start execution from." When Run finds other problems, a toast about a node names it by its title and has a Show button that selects the node and brings it into view. These toasts stay until the next Run or until you switch tabs. The executable set includes each triggered node, its downstream data flow, any upstream nodes that feed data into that set, and internal roots in any reached preset or subgraph container. :::

This trigger-based routing is what lets you keep scratch nodes on the canvas without running them. To choose between several sources for one input, use a Switch node.

4. Run it​

Click Run. Each node's card shows its status while the run goes, and the Execution Log lists each node's outcome and the Print node's output. See Running Graphs for what happens during execution.

5. Inspect what flowed​

Record node outputs is on by default (Settings → Recording & Inspection), so the run you just made already captured every node's output. Click any node to open the Teaching Inspector and see the exact tensor — shape, dtype, min/max/mean, and values — at every step.

Next steps​

  • Load a real example instead of building from scratch — see the Examples Gallery (e.g. Train CNN on MNIST).
  • Browse every node you can drop on the canvas in the Node Reference.