Skip to main content
Build a parametric workflow that takes a single design concept and produces image variations from it. By the end you’ll have a working three-node workflow you can extend in any direction. Time: 10 minutes You’ll need: A free Runchat account (Google or email signup works)

​
What you’ll learn

  • Adding nodes to the canvas
  • Writing a prompt and running an image model
  • Connecting nodes with edges to pass data
  • Using an Agent node to generate prompt variations
  • Returning a list of outputs from one run

​
1. Open a new workflow

The examples browser on a new workflow, with Blank canvas at the top right
Open the Runchat editor. The examples browser opens on Start from an example; click Blank canvas. Runchat saves automatically as you work. To rename the workflow, click its name in the top-left and choose Rename. Press Ctrl/Cmd + S to save manually.

​
2. Add an image model

Media nodes run generative AI models for images, video, and 3D. Each one is titled with the model it runs. The fastest way to add one is to double-click the canvas and type the model name. Double-click the canvas to open the node picker, type Flux 2 Klein, and select Flux 2 Klein.
Searching for an image model in the node picker
You can also open the node picker with Ctrl/Cmd + K, or click + next to the agent composer and choose Browse all. See Connecting Nodes for the full reference.

​
3. Write your first prompt

In the prompt input on the Flux 2 Klein node, paste:
Image node with a prompt entered, its toolbar above
Click Run in the node header to run it. Hover Run to see what the run will cost.
Running a node costs 1 credit, plus the model. Fast models like Runchat Image are included in the free usage. Per-model pricing shows next to each model in the node picker. See Credits for full pricing.

​
4. View the output

Image node showing a generated image
The result appears on the node and the inputs collapse to a short summary. Click the summary to expand the inputs again.

​
5. Add an Input node

You’ll connect a separate Input node so you can change the design concept without editing the prompt directly. Click + next to the agent composer and choose Text to add a text input. Move it to the left of the image node and enter:
Text input placed to the left of the image node

​
6. Add an Agent node between them

Double-click the canvas, type agent, and add an Agent node. Move it between the input and the image node.
Agent node placed between the input and the image node
The Agent will read your design concept, produce a list of variations, and feed each one into the image node as its own prompt.

​
7. Wire the three nodes together

Drag from the output handle on the right side of the input to the Context input on the Agent node. Then drag from the Agent’s output to the Prompt input on the image node. You can drop a wire on the input handle or anywhere on the input row.
The three nodes wired together with edges
Edge colours indicate data type. Text is blue, images are yellow, and so on. See Data Types for the full reference.

​
8. Configure the Agent for variations

Select the Agent node, open More (…) on its toolbar, and change Output format from Markdown to List. This forces the output to a list of values instead of a paragraph. In the Agent’s prompt field, enter:
The Agent node's More menu with Output format set to List

​
9. Run the workflow

Select all three nodes (drag a selection box, or Ctrl/Cmd + A) and click Run in the selection bar. Runchat figures out the order automatically: the input first, then the Agent, then the image node runs four times in parallel.
The Agent returns four prompt variations and the image node runs once for each
When the Agent returns a list, the downstream node runs once per item. This is Runchat’s Each vs All behaviour: by default each item processes individually. To combine all items into a single run, click the Each chip on the input row and choose All. Page through the results with the strip under the image (1 of 4), or drag a thumbnail onto the canvas to pull it out as its own node.

​
Next steps