Skip to content
Luma AI Help Center home
InboxAsk a human

Layers: Overview

Updated by Martin

What Is Layers?

Layers is a Luma feature, powered by Uni-1, that gives precision editing and control over every element of an image. Instead of treating an image as one flat file, Layers generates it as a set of independent, editable layers, and it can break down any uploaded image into layers too. The core idea: change one element and preserve everything else.

The Problem It Solves

A common workflow pain: you have an approved image (the right composition, mood, and shot) but need a small fix, like swapping a product, changing a headline, or localizing copy for new markets. Previously, that meant either regenerating the whole image and hoping the approved details survived, or exporting it to another tool to edit by hand.

Full regenerations are unreliable for this. Each pass through the model degrades the image a little more, so a small requested change can end up altering things that were never supposed to move. Luma's own in-house creative team dealt with this for years, exporting images out of Luma just to make final edits before bringing them back in.

How Layers Works

Every object, text element, and background in an image becomes its own asset. You can adjust just the one you want, whether that's swapping a product, restyling a background, or regenerating a single object, and everything else in the frame stays exactly as it was. Text changes, like copy, font, or color, take effect without touching anything else in the layout.

Because untouched layers never pass back through the model, every edit behaves like a first edit: a layout approved on day one stays pixel-identical through version ten.

Native to Generation

Other tools take a flat image and try to reverse-engineer it into editable pieces after the fact. Uni-1 works the opposite way: it generates images as independently addressable layers from the first pixel. Layering is built into how the model understands a scene, not a step applied afterward.

That same understanding applies to uploads too. When you upload an existing image (a poster, a screenshot, an old brand asset), Uni-1 reads it the same way it generates one: by identifying what each region is and how it relates to the rest of the scene, rather than just grouping similar-looking pixels. That's what makes an edit to one layer hold up once the image is recomposed.

What You Can Do With Layers

  • Edit as live text: change copy, style, font, or color. Text stays adjustable rather than being baked into pixels, including custom brand fonts.

  • Rearrange per platform: make Z-order adjustments to group, hide, or delete elements, reposition them precisely, and change aspect ratio for different platforms.

  • Extract any element: pull a single layer onto a transparent background as its own standalone asset (a product shot, a logo, a character) so it's reusable anywhere, not locked inside the original composite.

  • Mix and match across images: drag a saved layer from one image into another instead of starting from a blank canvas.

  • Recompose existing creatives: upload an old poster or brand asset with no source file, and it decomposes into layers the same way a new generation would.

  • Localize at scale: generate batch variations, like localized ad copy across dozens of assets, while layout and style stay consistent throughout.

  • Finish at delivery quality: upscale individual layers to 2K on demand.

Layers works on both Luma generations and uploaded images, and the entire process runs conversationally in Luma Agents. You can ask the agent to extract specific layers, then keep chatting to edit or modify them, all in one place.