Articles on: Image creation
This article is also available in:

Workflow

Workflow


1. What is a workflow?

Workflow is designed to automate and augment your production process — making it more efficient, repeatable, and scalable.

Good use cases:

  • running the same pipeline across multiple assets (product references, a CSV of prompts, a collection of images),
  • producing the same visual in several formats (1:1, 4:5, 9:16…) without rebuilding each one,
  • re-running with tweaks — new prompt, different model, updated reference — without starting from scratch,
  • keeping outputs organized and traceable in collections.


👉 For a quick one-off generation, the Creation tab is simpler. Come to Workflow when you're thinking in volume or need a repeatable process.



2. When to use Workflow

Use Workflow when you need:

  • Batch production: generate dozens/hundreds of outputs from a Collection or a CSV of prompts
  • Speed: run a full pipeline in one click instead of repeating manual actions
  • Iteration: re-run quickly with updated settings and compare results
  • Consistency: apply the same steps to multiple assets
  • Exploration: explore creative directions by running the same pipeline with different models, prompts, or formats simultaneously


3. Getting started

There are two ways to start a workflow: from a template, or from scratch.


3.1. Choose a starting template (learn)

Available template categories:

  • Essentials
  • Fashion
  • Cosmetics
  • Furniture
  • Food
  • Other

When you open the Workflow tab, you'll see the available ready-made templates, organized by category — from essentials that cover the most common production needs (product shots, social formats, batch campaigns, creative A/B tests) to industry-specific templates for fashion, cosmetics, and more.

Pick the template closest to your use case, then customize the prompt, model, formats, and output collection to match your project.


3.2. Start a workflow from scratch (3 quick steps)


  1. Add nodes from the left library (Source / Create / Edit ).
  2. Connect nodes (output → input).
  3. Configure each node, then click Run.


4. The interface.

The Workflow interface is divided into three areas.


4.1. Left panel: Component library



This is where all available nodes live, organized by type: Source, Create, Edit, Export, and Variables. Drag any node onto the canvas to add it to your pipeline. Hover over the ⓘ icon on any node to see a quick description of what it does.


4.2. Center: Canvas


The canvas is where you build and visualize your pipeline. You can:

  • Drag nodes from the left panel to place them.
  • Connect nodes by clicking and dragging from an output handle (right side of a node) to an input handle (left side of the next).
  • Zoom in and out using the toolbar at the top or your trackpad.
  • Reposition nodes freely to keep your pipeline readable.
  • Organize nodes into sections to segment your pipeline into logical groups and run each one independently.
  • Run the full flow, which processes every connected node in order, from your source inputs through to the final output.

At the bottom of the canvas, you'll see the estimated credit cost for the current run, and the Run all button to execute the full pipeline. The dropdown arrow next to it shows a breakdown of pending outputs — images and videos — and lets you trigger the run from there.


4.2. Right panel: Inspector ( node settings )


Click any node on the canvas to open its settings in the Inspector panel. This is where you can configure everything: prompt content, model choice, upscale factor, number of generations, output collection.


👉 A good habit: configure nodes left to right, starting from your source nodes, before hitting Run all.


5. Node library.

5.1. Source

Source nodes define what goes into your pipeline. You can use one or combine several.

  • Prompt: provides the text instructions that downstream generation and edit nodes will use. Write your prompt once and connect it to as many nodes as needed.
  • LLM Prompt: a more advanced prompt node that uses a language model to generate or refine your prompt(s) dynamically. Useful when you want to produce varied or contextual prompt variations at scale, without writing each one manually.
    • Use the Instructions presets to define the general direction of what the LLM will do with your prompt and reference images:
      • Brand Visual Extractor
      • Video Cinematic
      • Product Packshot
      • Lifestyle Scene
      • Fashion Editorial
      • Video Shot
  • Reference image: adds one or more images as visual inputs. Once connected, use the Reference delivery toggle to control how references are processed in batch runs:
    • All together sends all references combined to each generation,
    • One by one pairs each reference with a separate generation.
    • You can also combine two Reference Image nodes — one set to All together and one to One by one — to match different products against the same background, for example.
  • CSV Prompts: imports a list of prompts from a CSV file for batch runs. Each row becomes a separate input, letting you generate large sets of outputs in a single run.
  • Collection: loads assets from an existing collection as inputs. Useful for running edit or export steps on assets you've already generated.


5.2. Create

  • Image: generates images from your connected sources. Configure the model, number of generations, and any additional settings in the Inspector.
  • Video: generates video clips from a prompt or image input. Connect your source nodes, set the video parameters, and run.


5.3. Edit

  • Reframe: adapts an image to new dimensions or aspect ratios. Connect an image output and set your target format. Possible ratios: 1:1, 16:9, 9:16, 4:3, 3:4, custom.
  • Upscale: increases the resolution and detail of an image. Choose your upscale factor and preset in the Inspector.
  • Replace background: replaces the background of an image while preserving the main subject. You can provide a prompt or a reference image to define the new background.
  • Remove BG: removes the background entirely and outputs a transparent cutout. Useful as an intermediate step before compositing or background replacement.


5.4. Export

  • Output: the final node in your pipeline. Connects to your result-producing nodes and lets you review, download, or save everything once the run is complete. You can download all assets in one .ZIP file.
  • Add to Collection: saves outputs directly into a selected collection as the pipeline runs. Use this to keep results organized and accessible for future workflows.


5.5. Variables

Variables let you set a base parameter once and reuse it across as many nodes as needed.

  • Model: define your model choice once and connect it to multiple generation nodes. Useful when you want to switch models across the whole pipeline without editing each node individually.
  • Number of Generations: set your sample count once and connect it to all relevant nodes, so you can adjust volume in one place.


6. Configuring and running your workflow

6.1. Before you run

Before launching a run, make sure each node is properly configured. Nodes that are missing required inputs — typically a prompt — will display a warning directly on the canvas.

A few good habits before hitting run:


👉 Start with 1–3 inputs to validate your settings and check the output quality before scaling up.


👉 Use a Collection or CSV as your source once you're happy with the results on a small batch.


👉 Keep one reference workflow per use case and duplicate it when you need variations — rather than modifying your working pipeline each time.



6.2. Shortcuts

Use these shortcuts to move faster on the canvas:

Cmd/Ctrl + Z · Cmd/Ctrl + Shift + Z — Undo / Redo

Cmd/Ctrl + A — Select all nodes

Cmd/Ctrl + Shift + drag — Select a zone

Shift + click & drag — Select multiple nodes

Cmd/Ctrl + D — Duplicate a node

Cmd/Ctrl + K — Open Resource Hub


6.3. Credit estimate

The estimated credit cost for the current run is displayed at the bottom of the canvas, just above the Run button. This updates automatically as you add nodes or change the number of generations, so you always know what a run will cost before committing.


6.4. Run : All / individual nodes / Section

The Run all button executes the full pipeline from source to output in one go. If you want more control, there are two other ways to run your workflow.

Run a section — click Run section on any section header to execute only the nodes inside that group. Useful when your pipeline has distinct stages and you want to validate or iterate on one part without triggering the rest.

Run a single node — click the ▷ button directly on any node card to run it in isolation. Useful when:

  • testing a new step without re-running the entire pipeline,
  • debugging a node that produced unexpected results,
  • iterating on one part of the flow without consuming credits on the rest.


7. Results, saving, and export

After a run, you can typically:

  • Preview outputs (often grouped by input)
  • Download a single result or all results as a ZIP
  • Save outputs into a Collection so you can re-open and re-run later


8. Re-run & iterate

Workflow is built for iteration:

  • Re-run the same flow with new parameters (formats, prompt, strength, number of variants)
  • Generate new variants without rebuilding the pipeline
  • Keep runs organized via Collections (so results stay traceable)


9. Managing your workflows


Your saved workflows are accessible from the dropdown at the top of the left panel. From there you can:

  • Create a new blank workflow using + New workflow,
  • Rename or delete any existing workflow using the edit and delete icons next to its name,
  • Import an existing workflow file using Import workflow — useful when sharing pipelines across workspaces or with your team.
  • Copy/paste multiples nodes through workflows


👉 It's worth keeping your workflow list organized as it grows. A clear naming convention — by use case, format, or campaign — will save time when you need to find and re-run a specific pipeline later.


10. Troubleshooting (common issues)

  • A step fails: retry the run. If it fails again, test with a single input to isolate the problem.
  • Outputs vary too much: rely more on reference images or more precise prompts.
  • Batch is too slow: reduce batch size, reduce variants, or simplify the pipeline.
  • Background edits look wrong: use higher‑contrast inputs, or add an Upscale step after background operations.


11. Glossary

  • Workflow: a pipeline you can run repeatedly.
  • Node: one block/step in the pipeline (source/create/edit).
  • Run: one execution of a workflow with specific inputs/settings by node or sections.
  • Collection: where inputs/outputs live so you can batch, organize, and re-use results.

Updated on: 23/06/2026

Was this article helpful?

Share your feedback

Cancel

Thank you!