Generate images in a workflow

Last updated: October 12, 2026

Add the Generate Image node to make images from a prompt, reference images or reference sets as one step of a workflow, so every run makes them the same way.

Before you start

  • Open the workflow on its canvas. Your changes are saved as a private draft until you publish; see Build and publish a workflow.
  • The node starts with a default model. Which image inputs, sizes and options it offers depend on the model you pick; see Choose a model.

Steps

  1. Click Generators in the toolbar at the bottom of the canvas and drag Generate Image onto the canvas. It starts with a default model.

  2. Drag from the output of the node that holds your prompt, such as a Text node, to the node's Prompt input. The prompt field shows the text it now reads. You can also type a prompt on the node itself.

  3. Click Image Size and pick the aspect ratio and resolution, for example 16:9 2K. Set Batch Size for how many images each prompt makes.

  4. Select the node and click Run in the toolbar above it, which shows the cost. The node shows Running while it works.

  5. When it finishes, the images show on the node.

Result

The images appear on the node and under Runs and outputs. Connect Generated Image to the next node, or pick it as a workflow output when you publish.

Production tips

  • Know what multiplies. The node makes Batch Size images for every prompt, and every connected image input and reference set multiplies that again: three prompts with two pose images and a batch of four is 24 images. Check the cost on Run first.
  • Keep the style fixed, expose the subject. Put the style in the node's prompt or a reference set and feed only the subject from a Text or Prompt Template node, so every run matches.
  • Plan sizes downstream. Every image comes out at Image Size. For several aspect ratios from one image, follow the node with Reframe Image, Resize Image or Layered Composition.
  • Mention a reference set in the prompt. Type @ in the prompt to pick a reference set by name; it is added to the node's Reference Sets.

If something goes wrong

  • Nothing is generated. The node needs a prompt or at least one image input. Connect one, or type a prompt on the node.
  • An image input has no effect, or is missing. The model you picked does not support that kind of reference, so the node hides or ignores it. Pick a model that does.
  • The run costs more than expected. Count the values on every connected input; they multiply. Reduce Batch Size while you test.

Related