Skip to content

Workflows#

The Workflows tab (4) is a node editor. Every generation in the other tabs is a graph underneath; here you build your own for pipelines the tabs do not offer: multi-stage refinement, custom conditioning, batch transformations, anything that chains models in an unusual order.

The workflows launchpad

Getting started#

The launchpad offers three doors: Browse Workflow Templates (bundled graphs that demonstrate the important nodes), Create a new Workflow from a blank board, and Load workflow from file (a JSON export). Start from a template that is close to what you want and modify it.

  • Add nodes with the search field or by right-clicking the board; connect an output on the right of a node to an input of the same type on the left of another.
  • Ctrl+C and Ctrl+V copy and paste nodes, Del removes them, Shift and drag selects several.
  • Nodes and fields can be renamed; the name shows in the linear view.
  • Use Cache in a node footer reuses the last result of that node when its inputs have not changed, which makes iterating on the end of a long graph fast.

Linear view#

Right-click an input label and choose Add to Linear View to expose it in a simple form beside the graph. A workflow with a good linear view can be run by someone who never opens the graph; the exposed fields travel with the saved workflow.

The nodes that matter#

Concept Nodes Note
Noise Noise Every denoise needs a noise input; the seed lives here. A Random Integer node gives a random seed per run.
Conditioning Prompt, the model's text encoder Positive and negative conditioning feed the denoise node; the encoder comes from the model loader.
Denoise Denoise Latents (or the family-specific denoiser) Steps, CFG or guidance, scheduler, denoising start and end for image-to-image.
Latents and pixels Image to Latents, Latents to Image Convert with the VAE in both directions.
Control ControlNet plus a processor node Canny, depth and the rest, wired into the denoise node.
LoRA LoRA Loader Patches the model and text encoder before denoising.
Output Save Image or any image output Writes to the gallery board of your choice.

Saving and sharing#

Workflows are saved to the library with a name and category; export as JSON to share. An image made by a workflow records it: Load Workflow in the gallery menu reopens the exact graph. On the canvas, Run Workflow on Canvas feeds the current image into any workflow with an image input.

The node set changes between versions; a workflow from an older version may need a node replaced. Start from the bundled templates after an update.