Guide · Image

Split an image into layers with Layerize

A flattened poster, a product shot, a finished ad: Layerize separates them into transparent PNG layers, one per element, that you can move, retouch or send on to another model. Here is how to use it, in four steps.

6 min readBeginnerUpdated 20 October 2026
Image
Layerize
Engine
Seedream
Resolution
1K
Output
Base + up to 16 layers (Seedream)
PNG layers
The workflow in this guide: an image, the Layerize node, PNG layers.Open the studio

What Layerize does

Layerize is a canvas node. It takes a finished image and returns several images: a base image, placed under every layer, and one transparent PNG layer per element it finds. Two engines: Seedream 5 Pro (ByteDance), which takes any image, and Ad Delayer, built for flat ads (background, product, logo, text). With Ad Delayer, the base image is the input image itself.

2 to 17

output images with Seedream: the base and up to 16 layers

PNG

transparent, for every layer

0

new runs to recompose in the editor

The four steps

  1. Step 1

    Add the node and connect the image

    On the canvas, add the Layerize node and connect an image to its input: an import or the output of a model in the same graph. With Seedream, the image must hold between 512×512 and 6000×6000 pixels in total, have an aspect ratio between 1/16 and 16 and weigh 30 MB at most.

  2. Step 2

    Pick the engine and the resolution

    Seedream, the default engine, splits any image; resolution goes from 1K (the default) to 2K, and 1.5K or 2K cost nearly twice as much per layer. Ad Delayer is meant for flat ads: it works on a copy of 800 px per side at most and also outputs the text structure as JSON. A plain photo does not break down.

  3. Step 3

    Name the elements if needed, then run

    With Seedream, connect a Text node to the second input (labelled "Éléments à séparer", elements to separate) to say what to isolate; without text, the model separates the main elements on its own. The cost shown before the run is a minimum: the node shows the range, and the bill depends on how many layers come out. With Ad Delayer, the price is fixed.

  4. Step 4

    Recompose and connect the layers

    After the run, every layer gets its own output, named after its content. The Edit button opens the editor: move, resize, rotate, change opacity, hide, and choose which layers keep an output (the "x/N en sortie" counter). Saving updates the Composite output without running the model again.

The settings at a glance

SettingValuesTip
Moteur (engine)Seedream or Ad DelayerAd Delayer for a flat ad
Résolution1K, 1.5K, 2KSeedream only; 1K by default, cheapest per layer
Optimisation du promptStandard or Rapide (fast)Seedream only; Standard for quality, Rapide for speed
Éléments à séparerOptional textSeedream only, through a Text node
Textes (text)Contours vectoriels or Texte vivantAd Delayer only: vector outlines or live, editable text

Use cases

Retouch one element

Connect that layer's output to an editing model: it works on that element alone, already cut out.

Isolate a product

Its layer is a transparent PNG, ready to place on a new background.

Vary a composition

With Seedream, move, enlarge or hide elements in the editor, without generating the image again.

Rework an ad

Ad Delayer separates background, product and logo, and describes the text as JSON.

Frequently asked questions

How many layers does Layerize produce?
With Seedream, 2 to 17 images: the base image and up to 16 layers. The number depends on the image and cannot be set in advance; naming the elements you want helps keep it down, with no guarantee.
How is Layerize billed?
Seedream is billed per layer produced. To launch, your balance must cover 17 layers, which are reserved; the difference with the layers actually produced is refunded as soon as the call returns. Ad Delayer is a flat fee per request, whatever the number of elements.
Can I export the layers as a PSD?
No, there is no PSD export and no ZIP archive. Every layer is a transparent PNG with its own output on the node: you connect it to the rest of the workflow.
Do edits in the editor cost credits?
No. Saving writes the new composition without running the model again. A new successful run, on the other hand, starts over: it clears the saved composition and the output selection.
Is Layerize available in Generate?
No, it is a canvas node. A workflow that uses it can, however, be published as an app.

Try Layerize

Open the studio, add the Layerize node to the canvas and connect an image to it.

Keep reading