Welcome to Workflows, Layer’s feature that allows you to create many assets and perform multiple AI steps in a single flow. This short document is intended to brief you on the basic nodes in our Workflows MVP, and give you an idea of how they may be used for your own use cases.

To access workflows, simply click on the Workflows link on the left of your workspace. If you do not see this option, please ensure that you’re logged into the exact account that was granted access to workflows.

This will take you to the main Workflows page, where you can start creating your own workflows or start using a Layer-created example. We highly recommend you begin by looking at the Layer-created examples to familiarize yourself with design patterns and try out a few workflows for yourself. You may view them by clicking on Featured Workflows:

To run or edit an example workflow, you’ll want to first import it into your workspace. This will allow you to create a local version of it to edit and update however you’d like. Click on the purple “Import to Workspace” button after clicking on the Workflow you’d like to work with. Or, click on View Workflow to see its node setup.

Once you import the workflow, the button will change to Run Workflow. Once you click that, you’ll be prompted to prepare the workflow by updating or editing any prompts or reference images (if any), and then you may run your workflow. First, click Prepare Workflow:

Then, after editing any prompts or reference images, your workflow will run. Please note that workflows may take minutes and, in the case of exceptionally complex or large workflows, up to an hour to complete. While we run all of our requests in parallel, our partner services may take time to finish rendering and delivering the assets in all the ways your workflow calls for.

Once the workflow is complete, the assets are shown to you:

To create a copy of an existing workflow, click the three-dot Settings icon on the top-right and choose to duplicate the workflow.

Once you view the workflow you’re duplicated (or clicked View Workflow) on the Workflow Details page, you’ll see one of two things:

To start a draft and begin editing the workflow, you’ll need to start a new draft. This ensures that any changes you’re making are local to your own draft and don’t affect the published version, which other people may be using. See below:

Once you’re in your workflow draft, you’ll be able to edit, delete, add, change, and generally adjust the workflow however you’d like. Once you’re ready to test your workflow, you can test in two different ways: Either by running a specific node, or by running the entire workflow.

The image below shows the Test Workflow button that you may use in the graph:

Below shows the Run Node option for an individual node:

And lastly, below shows the estimated unit cost for a workflow, as well as how many actions will be taken for a full workflow run. This may be affected by the number of prompts, reference images, image generation numbers, and assets you’re creating in your workflow.

When you’ve finished your draft, you may publish it back to your workspace or create a new workflow instead. Please note that publishing a workflow and overriding the existing one will update it for any users using the workflow, so we recommend that you create a new workflow if others are using your workflow.

The 2D Generator is where you can generate images with any model of your choice. You may use Layer-provided models, choose a base model, or use your own custom models built on your LoRAs. You can choose to generate one or more images per generation, as well as the aspect ratio. NOTE: We highly recommend you only generate one image per generator to start; Otherwise you may accidentally generate many more images than you intend.

The Prompt Editor is where you may edit an image input. This is one of the most commonly-used nodes on Layer, useful in most image editing steps and even for adding UI elements to scenes.

Keep in mind that every Prompt Edit will need at least one base image to be valid, and that a prompt edit will be run for each reference image you attach. The Workflows MVP does not yet support attaching multiple reference images to a single base image request.

This simple node will remove the background from any images fed into it. It will return back the input image in the same aspect ratio or resolution that it was provided in. This is recommended to be used after editing images with transparency originally in them, since most generation and edit models do not support transparency and will add back in neutral pixels instead.

The Upscale node allows you to determine how best to upscale your images. SeedVR2 is our recommended general-purpose upscaler, but you may pick other models that allow for more creativity or prompt adherence in your upscaling. For downscaling or setting an image to a specific resolution, we recommend creating a Composition Node and using a custom resolution to set that image as a layer at a specific resolution.

The Composition Node is the most powerful and flexible node in the workflow, and the key to using Workflows to their greatest potential. The Composition Node is built around the following concepts:

The intent of the composition node can be boiled down to, “I want to put one or more images in a specific arrangement and specific aspect ratio.” It can be used multiple times in a workflow, and can be used in many different ways to resize images or compose them in a different arrangement. For example:

If you drag a Composition Node output into a Prompt Edit or Remove Background node, you will automatically create a “Flatten Layers” node. This compresses the composition’s layers into a single image so it may be modified by the Refine node.

A prompt is the basic input required for most generation nodes. Prompts are strings of words that are fed into a node and help determine what may be created by that node. They also help guide the actions of certain nodes like Prompt Edit and can be chained together to modify an image in specific ways throughout the graph.

Every node that supports prompts in workflows also supports an array of prompts, meaning you may add as many prompts as you may like to the workflow. Keep in mind that for each prompt you add, you may end up multiplying the number of results at the end of your workflow. Be careful adding more prompts until after you test your workflow with as few prompts as possible.

References are images that may be used in generators, refine nodes, or in the composition node as static images. They may be used in a variety of ways:

You may add more than one image to reference nodes, but please note that this may increase the number of final results you receive.

Will I be charged units for my generations while I’m using workflows, even when I am testing out my workflow?

Yes, you will be charged units for your node’s generations. This occurs when you’re testing a node in the workflow, when you’re testing the entire workflow, or running the workflow itself. Layer makes an effort to only charge you for the specific nodes you need to run for your test, but sometimes those nodes may involve running previous nodes before the node you’ve selected to run is in a valid state to run. This may cause particularly complex workflows to charge tens or hundreds of units per test. We recommend using the absolute minimum number of prompts, reference images, and nodes to test your workflow so that you minimize the units consumed.

Where do my generated assets live? How can I access them again after I run a workflow?

Your assets are tied to the session you’re in. If you’re running a workflow normally outside of the graph, you can view your session in the left nav bar and view your creations in your activity feed. If you’re running your workflow within the graph, you can see all of your node’s generations by clicking on each node and viewing their outputs at the bottom of the graph.

I’m not seeing all of my generations in the workflow outputs?

You need to connect a node’s outputs to the Output node

This example shows a more complex composition node. You can view this example for yourself in the “Create Mobile Ad Banners” example workflow. Since this is a more complex example, we wanted to explain why it’s organized the way it is.

There are five specific areas of the composition node that should be called out in detail:

Here’s an example of the above, but for a more constrained aspect ratio that had layers turned off and re-arranged. This resolution and aspect ratio (sidebar banner size) lives within the exact same composition node as the example above: