Skip to main content

Executing Workflows

Learn how to run, monitor, and debug your workflows.

Running a Workflow​

Basic Execution​

There are two ways to run a workflow:

  1. Click the Run button on the Start node itself
  2. Click the Run button in the Floating Action Bar at the top of the screen

The workflow executes from the Start node, following all connections through the graph.

Auto-Save on Run

When you run a workflow, DUTO automatically saves unsaved flows to the database. This ensures every run has a flow ID for job recovery and output persistence.

Pre-Run Credit Check​

Before execution begins, the system checks whether you have enough credits:

  • The Start node displays the estimated credit cost
  • If you have insufficient credits, the Run button shows a warning and is disabled
  • The Floating Action Bar also displays credit cost and deficit information

Execution Flow​

DUTO uses a DAG (Directed Acyclic Graph) execution model:

  1. Start node triggers
  2. Orphan source nodes (Text Prompt, Upload Media, etc. without incoming edges) are processed first if they have pre-existing data
  3. Nodes execute when all upstream inputs are ready
  4. Independent branches run concurrently
  5. Execution completes when all paths finish (or an error halts a branch)
Start → A → B → Preview
         ↘ C ↗

Execution order:
1. Start
2. A
3. B and C (parallel)
4. Merge point (after both B and C complete)

Visual Feedback​

Node Status Indicators​

Each node shows a status icon during and after execution:

IconStatusMeaning
Info (gray)IdleNot yet started
Clock (orange)RunningCurrently executing
Check (green)SuccessFinished successfully
Warning (red)ErrorExecution failed

Edge Animation​

  • Edges show data flow direction during execution
  • Dotted lines indicate edges within a batch processing scope (between Batch Processor and Batch End nodes)
  • Edge statuses are reset to idle at the start of each new run
  • Visual indicators auto-clear 5 seconds after workflow completion

Activity Log​

The Workspace panel (lightbulb icon in the Floating Action Bar) contains an Activity Log tab that shows real-time execution events:

  • Node started / completed / error events
  • Credit refund notifications
  • Workflow start and completion status
  • Recovery events for resumed jobs

A red dot appears on the Workspace icon when there are unread errors.

Monitoring Execution​

Real-Time Updates​

Watch your workflow execute in real-time:

  • Node status icons change immediately as execution progresses
  • Preview Media nodes display content as it generates
  • Errors appear instantly on failed nodes

Node Inspector During Execution​

Select any node during execution to see:

  • Current status (idle, running, success, error)
  • Error messages (if failed)
  • Output preview (images, video, audio, text)
  • API payload details (for debugging)
  • Credit cost for that node
  • Upstream and downstream dependency graph

Output Persistence​

After a successful run, node outputs are persisted to the database. This means:

  • Previews survive page navigation and browser refresh
  • When you reopen a saved workflow, previously generated outputs are recovered
  • Pending jobs (still processing on the server) resume polling automatically

Handling Errors​

Error Display​

When a node fails:

  • The node shows a red warning icon
  • Error details appear in the Node Inspector panel
  • Downstream nodes that depend on the failed node do not execute
  • The workflow finishes with an "error" status

Common Errors​

ErrorCauseSolution
Insufficient creditsNot enough credits for the operationAdd credits or use budget mode models
Generation failedAI model returned an errorRetry, adjust parameters, or try a different model
Invalid inputWrong data type or missing required inputCheck connections and ensure upstream nodes produced valid output
TimeoutOperation took too longRetry or simplify the request

Error Recovery​

DUTO provides three recovery actions in the Node Inspector when a node has errored:

Run from Node​

Re-execute starting from a specific node, preserving upstream results:

  1. Select the errored node in the canvas
  2. Click Run from Node in the Inspector panel
  3. Execution starts from that node using cached inputs from upstream

This is also available on non-errored nodes to re-run a portion of the workflow.

Retry​

Clear the error and re-run just the failed node with the same inputs:

  1. Select the errored node
  2. Click Retry in the Inspector
  3. The node re-executes with its original inputs

Continue from Here​

Resume the workflow from the errored node, re-evaluating downstream dependencies like Sync Gates:

  1. Select the errored node
  2. Click Continue in the Inspector
  3. The error is cleared, downstream nodes are reset, and execution resumes

Retry Gate Node​

For automated retry handling, add a Retry Gate node before error-prone operations:

... → Retry Gate → Text to Image → ...

The Retry Gate automatically retries failed downstream nodes based on configured retry limits.

Partial Execution​

Run from Node​

Execute only part of your workflow:

  1. Select any node in the canvas
  2. Click Run from Node in the Inspector panel
  3. Upstream nodes that already have outputs are treated as completed
  4. Execution proceeds from the selected node through all downstream paths

Auto-Save to Library​

Toggle the Auto-save to Library button in the Floating Action Bar to automatically save all generated media (images, videos, audio) to your Content Library during execution. A summary toast notification appears after the workflow completes, showing how many items were saved.

Stopping Execution​

Cancel Running Workflow​

  • Click the Stop button in the Floating Action Bar

Running operations cancel, and the workflow stops at the current state.

Partial Results​

When you stop a workflow:

  • Completed nodes retain their results
  • Running nodes may have partial or no results
  • Queued nodes don't execute
  • All completed outputs are still persisted to the database

Job Recovery​

DUTO includes automatic job recovery for resilience:

  • When you load a saved workflow, the system checks for any completed jobs from previous runs and restores their outputs to the canvas
  • Pending jobs (still processing on the server) automatically resume polling for results
  • Recovery events appear in the Activity Log
  • If you start a new run, all recovery polls are cancelled to avoid conflicts

Performance Optimization​

Parallel Execution​

The DAG engine automatically runs independent branches concurrently. Design workflows to maximize parallelism:

Good: Independent branches run in parallel
Start → Text to Image → Image Upscale → Preview
      ↘ Text to Video → Preview

Poor: Unnecessary sequential chain
Start → A → B → C → D (when B doesn't depend on A's output)

Batch Processing​

For processing multiple items, use Batch Processor and Batch End nodes:

Start → Data Feed → Batch Processor → Process → Batch End

Benefits:

  • Configurable concurrency (1-10 parallel items)
  • Per-item error handling (continue, stop, or retry modes)
  • Progress tracking through batch iterations

Debugging Tips​

1. Use Preview Media Nodes​

Add Preview Media nodes at key points to verify intermediate results:

... → Suspicious Node → Preview Media → Continue → ...

2. Use the Inspector Node​

The Inspector node logs detailed information about data flowing through it -- useful for examining values between nodes.

3. Check the Activity Log​

Open the Workspace panel to view the Activity Log for a chronological record of node execution events, errors, and credit usage.

4. Inspect API Payloads​

Select a completed or errored node and expand the API Payload section in the Inspector to see the exact request sent to the AI service. You can copy this payload for debugging.

5. Simplify and Isolate​

To debug a problem:

  1. Use Run from Node to test just the problematic section
  2. Add Inspector or Preview nodes before and after the failing node
  3. Check that upstream nodes produced the expected output types

Next Steps​