How to Turn a Flat AI-Generated Poster into Editable Layers for Figma and Photoshop

Your image generator just gave you a decent event poster. Then the client asks for a different date, a 25% badge instead of 20%, and the cup moved a little to the left. The file you have is one flat PNG: the headline, the badge, the cup and the paper texture are all the same pixels, so none of those changes is a simple edit.

The fix is to split the image into layers, not to cut one thing out of it. An AI layer split returns every element on its own transparent layer at its original position, plus the background with the areas behind those elements painted in. You then edit the poster in Figma or Photoshop as if it had been built there.

Disclosure: I built LayerGrab. LayerGrab is an AI tool that splits one image into up to 16 separate transparent element layers, or pulls out just the one element you describe; it works on layergrab.com and as Photoshop, Figma and Sketch plugins and a Chrome extension. An AI assistant helped draft this post from LayerGrab's source code and public sample splits; the labels, layer names and file names below were checked against both on October 29, 2026.

What you need

• A flat image: PNG, JPG or WebP, up to 20 MB, from ChatGPT, Midjourney, Gemini or any other generator. The example here is a coffee-shop poster generated with gpt-image-2.
• A layergrab.com account to split your own image (Google sign-in). The sample splits on the home page open without one.
• Figma (desktop or browser) for the Figma route, or Photoshop or Photopea for the PSD route.

Background removal vs real layers

A background remover answers one question: which pixels are the subject? It keeps those, makes everything else transparent and hands you one PNG. That is right for a product listing or a sticker. For a poster you want to edit, it loses most of what you need:

• The headline, the badge and the decorations are either thrown away with the background or fused into the "subject".
• The background itself is not returned. If you want the scene without the subject, you get a hole, because nothing behind the subject was ever stored in the file.
• An element partly covered by another one comes out with a piece missing.

A layer split answers a different question: what is every element, and what was behind each of them? It returns a stack:

• one transparent layer per element, at its original position and size;
• a background layer with the gaps painted in, so hiding or moving an element never exposes an empty patch;
• elements completed where something overlapped them.

The coffee poster shows the difference. Split with every element, it came back as 10 named layers over a filled background. The latte and the whipped cream are layers of their own, so the cup layer is a whole, empty cup: the rim and the inside that the drink covered were painted in. Hide all ten layers and what is left is plain paper texture with nothing missing.

The rule of thumb: if the deliverable is one subject on transparency, a background remover is the simpler tool. If you need to change text, move things, or swap one element and keep the rest, you need layers.

Step 1: Upload the largest version you have

Open layergrab.com; the home page is the tool. Drop the image on the Image panel, paste it, or browse to it.

Upload the biggest copy the generator gives you. The split works from the pixels you upload, so a 1-megapixel render gives you a 1-megapixel layer stack. If the poster is headed for print, upscale before you split, not after.

Step 2: Choose what to separate, then split

The box under the image, What to focus on, decides how the image is cut. It has two presets, and anything you type replaces them:

• Main parts is the default. It pre-fills the box with this rule, which you can read and edit:

  Group related items: keep each main subject, each text block and each set of similar decorations on one layer, and do not split an object into its parts.

Use it when you want a short, manageable stack: the rule asks for similar decorations, such as the four maple leaves, to share one layer.

• Every element empties the box, and every main element gets its own layer, up to 16. Use it when you will animate or rearrange the pieces one by one.
• Your own words take over completely: "only the headline and the 20% OFF badge", "the cup and the drink". Naming a single element returns that element plus the filled background.

Then click Split into layers. The panel shows the current step and the seconds elapsed, with a Cancel button. Most splits take 60 to 90 seconds; busy photos and illustrations with many elements can take up to about five minutes. Keep the tab open.

Billing is per layer returned, with the filled background counting as one, and a failed split is refunded. If your balance cannot cover the whole result, nothing is charged and the result stays locked until you unlock it.

Step 3: Review the layer names

Every layer comes back with an English name describing what it shows, written by the model during the split. These are the names from the coffee poster's every-element split, as the result panel lists them:

AUTUMN LATTE WEEK headline text
20% OFF red badge
Yellow maple leaf (top left)
Red maple leaf (bottom left)
Orange maple leaf (top right)
Orange maple leaf (bottom right)
White coffee cup with pumpkin logo
Pumpkin spice latte drink base
Whipped cream topping with cinnamon dust
Bottom event info text
Background (filled)

The names matter more than they look, because they travel with the layers into every export:

• in the layered PSD, each layer carries its name, and the background is Background (filled);
• in the ZIP, the files are numbered and named after them: 00-background.png, 01-autumn-latte-week-headline-text.png, 02-20-off-red-badge.png and so on;
• in Figma, the plugin uses them as the layer names inside its group.

Use this step to catch problems before you export: hide a layer with its eye icon and check that the element is whole and the background under it is clean. If two things you need separately came back as one layer, or you got more pieces than you want, go back to Step 2 and run again with the other preset or your own description. Two runs on the same image can split it slightly differently.

The web tool does not rename layers. If your team has a convention such as txt/headline, rename after import in Figma or Photoshop.

Step 4a: Bring the layers into Figma

The most direct route is the LayerGrab plugin on Figma Community:

1. Run the plugin and click Log in with Google. Your browser opens layergrab.com; sign in and click Approve. The plugin then uses the same account and credits as the website.
2. Select one image, frame or group on the canvas, optionally type what to separate, and click Split into Layers. You can keep designing while it runs.
3. A group called LayerGrab is inserted directly above your selection in the same parent, so it lines up exactly with the original. The filled background sits at the bottom and the named elements above it.

Each element is a rectangle with the separated image as its fill, so you can move, resize, recolour or delete it like any other layer. They are images, not vector paths or text objects.

Already split on the website? Copy layer puts one element on the clipboard for Figma, and All layers (ZIP) gives you every PNG. Those PNGs are cropped to their element, so you position them by hand; the plugin and the PSD keep everything aligned.

Step 4b: Bring the layers into Photoshop

Click PSD in the result panel. You get a layered PSD at the size of the split: Background (filled) at the bottom and every element above it at its original position, each with its name. It opens in Photoshop and in Photopea.

From there the client's edits are ordinary layer work: hide AUTUMN LATTE WEEK headline text and set the new headline in a real font on the paper that is already behind it, nudge the cup layer left, and replace the badge with your own.

There is also a LayerGrab Photoshop plugin that adds the split as a group above your layers, but at the time of writing it is still in testing, so the PSD export is the way into Photoshop today.

Verify the split

Before you hand the file on, run three checks:

1. Hide every element layer. The background alone should be a complete picture: no holes and no ghost outlines where elements used to be.
2. Compare with the original. In Figma, hide the LayerGrab group to see the original underneath; in Photoshop, drag the original into the PSD as the bottom layer. Each element should sit exactly over its spot.
3. Count the layers. The PSD should have one layer per element plus Background (filled). For the coffee poster that is 11.

When not to use LayerGrab

• You need one subject on transparency and nothing else. A background remover is the simpler tool for that.
• You need the text to stay editable as type. LayerGrab returns text as pixels on its own layer; you hide it and retype. If you finish your designs inside Canva, Canva Magic Layers rebuilds text as editable Canva text and saves that step.
• The image must not leave your machine. LayerGrab is a hosted service, so the image is uploaded to its servers. An open model such as Qwen-Image-Layered can run on your own GPU instead.
• The shadows matter. A shadow an object casts on the background belongs to the background and does not move with the object; add a new one where you place it.

Common problems

• "Use a PNG, JPG or WebP image." The browser could not read the file as an image, for example a PDF. Export it as PNG first.
• "That image is over 20 MB. Try a smaller copy." Save a JPG or WebP copy at the same pixel size; you do not need to downscale.
• "The split failed. Your credits were refunded." Run it again. If you typed a long instruction, try a shorter one.
• The layers look soft in print. The stack has the resolution of the upload. Upscale the original and split again.

What is the first thing you would fix in a flat AI poster once it comes apart into layers: the text, the product shot, or the layout?

---

I build LayerGrab. For how it compares with seven other layer splitters on layer count, picking one object and export formats, see The 8 Best AI Image Layer Splitters in 2026, Compared on the LayerGrab blog. The tool itself is at layergrab.com.