Skip to main content

Creating Workflows

Learn how to design and build effective workflows in the Flow Editor.

Workflow Anatomy​

Every workflow follows a basic structure:

Input → Processing → Output
  • Input - Data enters the workflow (prompts, uploads, data feeds)
  • Processing - Nodes transform, generate, or analyze
  • Output - Results are previewed, saved to the library, or sent to an API endpoint

Starting a Workflow​

The Start Node​

Every workflow begins with a Start node. This node:

  • Triggers when you click the Run button (on the Start node itself or in the Floating Action Bar)
  • Has no inputs (it is the entry point)
  • Has one output of type flow that connects to your first processing node
  • Shows the estimated credit cost and a warning if you have insufficient credits

Adding Your First Node​

After the Start node, add nodes based on what you want to create:

GoalTypical Node Chain
Generate an image from textStart → Text Prompt → Text to Image → Preview Media
Process an uploaded imageStart → Upload Media → Image Upscale → Preview Media
Batch process dataStart → Data Feed → Batch Processor → ... → Batch End
Create video from imageStart → Upload Media → Image to Video → Preview Media
Intelligent content creationStart → Text Prompt → Brain → Text to Image → Preview Media

Common Workflow Patterns​

Linear Workflow​

The simplest pattern -- nodes execute in sequence:

Start → A → B → C → Preview

Example: Generate image and upscale it

Start → Text Prompt → Text to Image → Image Upscale → Preview Media

Branching Workflow​

Split data into multiple paths. The DAG execution engine runs independent branches concurrently:

Start → A → B → Preview 1
         ↘ C → Preview 2

Example: Generate an image and create both a video and an upscaled version

Start → Text Prompt → Text to Image → Image to Video → Preview Media
                              ↘ Image Upscale → Preview Media

Merging Workflow​

Combine multiple inputs using a Sync Gate or multi-input node:

Path A ↘
        → Sync Gate → Next Node
Path B ↗

Example: Combine audio and video

Upload Video ↘
              → Lip Sync → Preview Media
Text to Audio ↗

Conditional Workflow​

Route data to different paths based on conditions using the Conditional Switch node:

Start → Analyze → Conditional Switch
                   ↓ Branch A → Processing A
                   ↓ Branch B → Processing B

Example: Process based on image content

Start → Upload Media → Vision Analyzer → Conditional Switch
                                          ↓ Has Face → Face Swap
                                          ↓ No Face → Image Upscale

Batch Processing​

Process multiple items through the same workflow segment. Edges within the batch scope appear as dotted lines:

Start → Data Feed → Batch Processor → Processing → Batch End → Preview

Example: Generate images for multiple prompts

Start → Data Feed → Batch Processor → Text to Image → Batch End

Batch context variables are available inside the loop:

  • $item - The current item being processed
  • $index - Current iteration index (0-based)
  • $total - Total number of items

Connecting Nodes​

Handle Types​

Nodes have typed handles for inputs and outputs. Handles are color-coded:

TypeColorDescription
imageBlueStatic images (PNG, JPG, WebP)
videoPurpleVideo files (MP4, WebM)
audioGreenAudio files (MP3, WAV)
textYellowStrings and prompts
flowPrimaryFlow trigger (Start node output)
maskAmberMask data for inpainting
anyGrayUniversal connector, accepts all types

Connection Rules​

  • Outputs connect to inputs (not output-to-output or input-to-input)
  • Types must be compatible (or use any handles)
  • One output can connect to multiple inputs (branching)
  • Some inputs accept multiple connections (e.g., Sync Gate has 3 inputs)
  • The Quick Connect Menu (appears when you drop a connection on empty space) automatically filters to show only compatible nodes

Valid Connection Examples​

[image output] → [image input] ✓
[image output] → [any input]   ✓
[any output]   → [image input] ✓
[text output]  → [image input] ✗

Using Node Configurations​

Each node has configuration options that appear in the Node Inspector panel when selected.

Node-Specific Options​

Each node type has unique options. For example:

  • Text to Image - Model selection, aspect ratio, style, seed
  • Brain - System prompt, mode (creative/research/hybrid/analysis), connected output configuration
  • Batch Processor - Mode (parallel/sequential), concurrency, error handling
  • Conditional Switch - Branch conditions and routing rules
  • Data Feed - JSON data input with row/column structure

Working with Data​

Input Sources​

Data enters workflows through:

  • Text Prompt - Manual text input
  • Upload Media - Upload images, videos, or audio from your device
  • Remote Media - Load media from a URL
  • Library - Select assets from your Content Library
  • Data Feed - Structured JSON data for batch processing

Data Transformation​

Use the Brain node for complex data transformation:

Start → Data Feed → Brain (Transform) → Text to Image

The Brain node can:

  • Extract information from structured data
  • Reformat text prompts
  • Make decisions based on content
  • Generate new data from multiple inputs
  • Operate in four modes: creative, research, hybrid, and analysis

Output Destinations​

  • Preview Media - Display results in the canvas for review
  • Text Preview - Display text output inline
  • Content Library - Automatically save generated media when Auto-save to Library is enabled
  • API Endpoint - Expose workflow results through an API (when API mode is enabled)

Best Practices​

1. Plan Before Building​

Sketch your workflow first:

  • What inputs do you need?
  • What transformations occur?
  • Where do outputs go?

Or use AI Generation in the Bottom Node Menu -- describe your workflow in natural language and the AI will build it for you.

2. Test Incrementally​

Add nodes one at a time and test:

  • Verify each node works before adding the next
  • Use Preview Media nodes to check intermediate results
  • Use the Inspector node to debug data flowing between nodes
  • Fix issues early before the workflow grows complex

3. Handle Errors Gracefully​

Add error handling for production workflows:

  • Use Retry Gate for transient failures (e.g., API timeouts)
  • Add Conditional Switch for edge cases
  • Use the Inspector node to log and examine values at any point

4. Optimize for Performance​

Consider execution efficiency:

  • Parallel branches execute concurrently automatically
  • Use Batch Processor for multiple items (recommended concurrency: 3-5)
  • Choose appropriate models -- use Budget Mode for drafts, premium models for final output

5. Document Your Workflow​

Make workflows understandable:

  • Use Post-It Note nodes to annotate complex sections
  • Name your workflow clearly when saving

Next Steps​