AI image models can now split a flat image into layers: a single JPG goes in, and each element comes back on its own transparent layer, named and in place, with the background behind it painted in. Separating an image by hand was the last job only Photoshop could do. That job is over.
Last updated 2026-10-06
There was one thing nobody could take from Photoshop.
AI writes the copy. AI generates the image. AI picks the fonts. But when a client sent a flat JPG and said “make the logo bigger”, you still opened Photoshop, grabbed the Pen tool and lost your afternoon. Cut the headline. Layer via Copy. Patch the hole. Cut the product. Patch again. Element by element, until a dead image was editable again.
That was Photoshop's last moat. It's gone.
This coffee poster was one flat image. 100 seconds later it was 10 separate, named layers: the headline, the 20% OFF badge, the cup, every leaf. Each one exactly where it was. And behind them, the background nobody ever drew, painted in. Move the cup and there is no hole. Export it and you get a layered PSD that looks like someone spent hours building it.
Nobody did.
11 layers
Bottom event info text
Whipped cream topping with cinnamon dust
Pumpkin spice latte drink base
White coffee cup with pumpkin logo
Orange maple leaf (bottom right)
Orange maple leaf (top right)
Background (filled)
In December 2025 the Qwen team open-sourced Qwen-Image-Layered, a model that decomposes an image into separate RGBA layers. In March 2026 Canva shipped Magic Layers. Image models no longer just make pictures. They understand them well enough to take them apart.
Photoshop will still be on your machine. The irony is that we even built a Photoshop plugin for this. But the job that made you open it, the slow, manual, pixel-by-pixel separation, is over.
Thirty years of the Pen tool. Ended in 100 seconds.
As a place to retouch, typeset and finish work, no: we even ship a Photoshop plugin. What is dead is the manual job of cutting a flat image back into layers with the Pen tool, Layer via Copy and Content-Aware Fill. An AI layer splitter does that in about a minute and hands Photoshop a finished layered PSD.
Qwen-Image-Layered, an open model the Qwen team released in December 2025, decomposes an image into RGBA layers. Canva Magic Layers, launched in March 2026, does it inside Canva and rebuilds text as editable text. LayerGrab splits images into named layers with the background filled in and exports PNG, ZIP or a layered PSD.
Text comes back as an image layer rather than live type in most tools, anything fully hidden in the original is painted in rather than recovered, and fine hair is softer than a careful hand mask. If the original layered file still exists, it is still the better starting point.