Best Practices

These patterns help you build, reuse, and scale workflows.

Build one working path first

Add the minimum inputs and one generation node.
Check its model and provider key.
Run the node and inspect the result.
Add the next processing step.
Save the project before you expand it into a batch.

A small workflow makes missing inputs and model errors easier to locate. Use Run selected or Run from selected for focused checks.

Reuse prompts

A Prompt node can define a named variable. A Prompt Constructor uses names such as @subject and @style in a shared template. Connect the source Prompt nodes to the constructor before you run it. The constructor editor can copy the resolved preview for inspection.

Prompt: subject ─┐
                ├→ Prompt Constructor → Generate Image
Prompt: style ──┘

An LLM Generate node can turn an instruction into prompt text. Name the subject, composition, light, and changes that you want. Check the generated text before a large batch.

Compare two models

A Router passes one source to several downstream nodes. An Image Compare node shows two image results with a slider.

                  ┌→ Generate Image A ─┐
Prompt → Router ──┤                    ├→ Image Compare
                  └→ Generate Image B ─┘

Keep the prompt and references the same to compare model outputs. The multi-select toolbar can change the model across several selected generators.

Process a list with Array

Connect a Prompt or LLM Generate text output to an Array node.
Choose the delimiter, newline, or regular-expression split mode.
Check the parsed items.
Enable Batch and connect the Array to a supported generator.
Run the workflow.

Array batches run sequentially through Generate Image, Generate Video, Generate Audio, or LLM Generate. Generate 3D does not use this batch path. For separate editable branches, turn Batch off and use Make Prompts. Click a parsed item to read its full text. Each outgoing noodle shows the item it carries. An Output Gallery collects image and video results.

Repeat a workflow or selection

Open the Run menu.
Set Runs to a small count between 1 and 50.
Select Run all, Run from selected, or Run selected to start immediately.
Inspect the outputs marked Run n of N.

Groups also offer Run group with the shared Runs count. This repeats the chosen graph; Array Batch repeats its text items through a generator. Both mechanisms can increase provider usage.

If repetitions remain, the first Stop finishes the current run and prevents later runs. A second Stop stops immediately. See workflow execution for the full stopping behavior.

Split an image into cell workflows

Split Grid creates one node set for each image cell. Each axis supports up to 16 cells.

Connect a source image to Split Grid.
Set the rows and columns.
Adjust the slice lines for unequal cells.
Select Open cell editor and build the template for one cell.
Choose a grid, vertical, or horizontal node layout.
Split the image to create each cell's node set.

The cell editor supports all node types, including video generation. A shared downstream Router can collect the template's terminal outputs. See Split Grid before you change an existing materialized layout.

Make execution deliberate

A Switch enables selected branches without removing their noodles. Optional input nodes let an unused branch stay empty. Locked groups skip completed or inactive parts of the workflow. Hidden noodles still run, so hiding is only a visual change.

A loop noodle repeats an upstream path. Set a small iteration count first and inspect the result. A fallback model retries failed generation with another model and its own settings. Check that the fallback's provider key and inputs are available.

Organize a large canvas

Use node names and comments that describe each step's purpose. Groups keep related nodes together. Use Arrange nodes in the multi-select toolbar for stacking or grid placement. Its spacing slider adjusts the gaps between selected nodes. Noodle labels and bundles make shared inputs easier to trace. Double-click a hidden stack's label to rename its noodles together. Use its upstream jump to find the source on a large canvas.

Use saved ComfyUI nodes for workflows with repeated ComfyUI settings. Use Output Gallery → Extract to turn gallery items into reusable input nodes.

Control cost and preserve work

Check Models and pricing before you run batches, loops, or fallback chains. The app's cost totals do not cover every provider or model. The provider account remains the record of billed usage.

Save important changes with Cmd/Ctrl + S or Nb → Save project. Use Assets to find earlier results and their generating workflow snapshots. For portable projects, keep workflow files and external media folders together. For embedded images, enable Settings → Project → Embed images as base64. Large embedded workflows use more memory.

The desktop app provides crash recovery as well as normal project saves.

Models and Pricing Desktop App