These nodes cover image generation, editing, and composition inside Workflows, from your first generated asset through final layout.
The Create Image Node
The Create Image node generates images with any model of your choice, and is the core node used in most workflows. It works the same way as Layer's 2D inference form: you can feed in reference images, add prompts, use custom trained LoRA models, and use its outputs as reference or base images for other nodes.
A Create Image node's output is always a single file format (typically PNG) at a single aspect ratio and resolution. To get multiple aspect ratios from one image, use the Composition Node, the Edit Image node, or feed the output into a new Create Image node at a different aspect ratio as a reference image.
Not every 2D model supports the same reference image types. Models like Flux.1 Dev support pose or structure references in addition to normal reference images, while others only accept normal reference images. Check the model's supported reference types before connecting anything beyond a standard reference image.
Setting | What it controls |
Model | The 2D generation model for this node. Each generator in a chain can use a different model. |
Prompt | Enter directly, or feed in a prompt node. Prompts entered directly cannot be changed when the workflow runs. |
Reference image | Upload directly or attach a workflow input. |
Aspect ratio and resolution | One combination per node. Use the Composition Node and Edit Image node for outpainting into other resolutions. |
Number of images | 1 to 16 per generator. Increasing this multiplies generations further down the workflow. |
The Edit Image Node
The Edit Image node modifies an existing image input, and is one of the most commonly used nodes on Layer, useful for most image editing steps and even for adding UI elements to a scene. It is especially powerful when paired with a reasoning model like Gemini 2.5 or Gemini 3.
Edit Image outputs are always 2D (typically PNG) and take two kinds of input:
Images to Edit: the base image being modified, such as a piece of key art, an object, or a banner background.
Reference Groups: a collection of reference images sent alongside the base image and referenced in the prompt. You can create up to 7 Reference Groups per node, with no extra unit cost regardless of how many you use.
The number of Images to Edit and the number of images per Reference Group determine how many times the node runs:
4 Images to Edit alone runs the node 4 times.
4 Images to Edit with one Reference Group containing 1 image also runs the node 4 times.
4 Images to Edit with one Reference Group containing 2 images runs the node 8 times (4 base images x 2 reference images).
4 Images to Edit with 3 separate Reference Groups of 1 image each runs the node 4 times (4 base images x 1 image per group).
As of this writing, we recommend Gemini 3 for the best results across the widest range of edit requests. For most cases, use a dynamic aspect ratio rather than a fixed one, especially if you are feeding in images of multiple aspect ratios.
The Upscale Node
The Upscale node raises an image's resolution. Some upscalers include creativity or resemblance controls; others act as a simple upscaler with no adjustable settings. SeedVR2 is our recommended general purpose upscaler, while models like Topaz Enhance offer more control over creativity and resemblance to the original image.
To downscale or set an image to a specific resolution instead, use a Composition Node set to that resolution, and feed the image in so it fills the entire canvas.
The Remove Background Node
The Remove Background node strips the background from any image fed into it and returns the result at the same aspect ratio and resolution. Use it after editing images that originally had transparency, since most generation and edit models don't support transparency and will fill that space with neutral pixels instead.
There are no model choices for this node. Simply connect an image input and drag the output to the next node.
The Composition Node
The Composition Node is one of the most powerful and flexible nodes in Workflows, built around three ideas: choosing the aspect ratio and resolution of your output, adding any number of layers, and positioning or scaling each layer independently per resolution. In short, it lets you place one or more images in a specific arrangement at a specific size.
Core components of the node:
Composition selector: choose or create compositions, each with its own aspect ratio and resolution.
Layers: any number of inputs, shared across compositions. Layers stack top to bottom in the list, with higher layers appearing on top. Use the eyeball icon to show or hide a layer per composition rather than deleting it, and the six dot icon to reorder layers.
Canvas size: preset sizes are available, or set any custom size.
Workflow inputs can be assigned to a specific layer, and more than one input can feed a single layer, useful when working with multiple characters or images across layers.
If you connect a Composition Node's output into an Edit Image or Remove Background node, Layer automatically creates a "Flatten Layers" node, which compresses the composition into a single image so it can be modified further.
Common use cases:
Goal | Approach |
Turn a square asset into a landscape image without losing content | Center the original square image on a larger canvas with transparent padding around it, then feed that into an Edit Image node with a prompt to outpaint from the center. |
Layer a logo and CTA on top of a background | Bring in a background layer plus separate logo and CTA layers (as reference inputs), positioning each and setting their stacking order so they render on top of the background. |
Preview how an app icon reads with rounded corners | Overlay a mask reference image that simulates the app store's corner rounding, so you can confirm important details still read correctly before publishing. |