They're right that a converter can't do this. We don't convert — we rebuild.

Why "jpg to psd converter" gives you a flat file

A JPG stores one grid of pixels and nothing else — no layers, no text objects, no fonts. So a format converter has nothing to separate; it wraps your flat pixels in a .psd container as a single locked background layer and calls it done. That is why the forums are full of "how do I restore to layers?" answered with "you can't." The extension changed; the file didn't. It is an honest limitation of treating this as a file-format problem.

Reconstruction is a different job

We don't convert the file — we re-read the picture. AI segments the image into its real parts (subject, each object, text, background), and because lifting an object leaves a hole, it inpaints the background behind every part so each layer is complete and independently movable. The result is a true multi-layer PSD with named layers and groups that opens in Photoshop, Affinity, Photopea and GIMP. Honest scope: this reconstructs sensible, editable layers from the visible image — it cannot resurrect the exact layers the original designer had, because that information was destroyed the moment the file was flattened.

And the text comes back editable

This is the part converters can never touch: the text in your JPG is just colored pixels, but we detect it, read it, and write it back as a real Photoshop type layer you can re-type, translate or restyle. We verify it on every export — a written file re-opened in an independent PSD reader still reports the text as live, editable type with its size and color intact. Delicate or heavily stylized logos are flagged rather than faked, so you're never handed a wrong guess.

Unflattening is not a format conversion

People search for this as "unflatten an image", and the phrasing is worth taking seriously because it names the confusion exactly. Flattening is not a wrapper you can peel off — it is destructive, in the way baking is destructive. When a design is exported to JPG or PNG, every layer collapses into a single raster and the subject, the background, each object and every line of text become the same undifferentiated pixels. Nothing about the resulting file remembers that they were ever separate. That is why renaming it to .psd changes nothing, and why no amount of format cleverness gets the structure back: the information is not hiding in the file in a form a converter could read, it is gone. Unflattening, done honestly, means looking at the picture and rebuilding a layer structure that behaves like the one that was lost. Any flat raster is a valid starting point — JPG, PNG, WebP, a screenshot, a Canva export, a photo of a printed proof.

The 30-second honesty test

Any tool can slap a .psd extension on your image. Before you trust one, open its output and check the Layers panel: one "Background" layer means you got a renamed JPG. Real separated layers you can hide, move and edit means reconstruction actually happened. Two bonus checks that catch the halfway cases: does each lifted layer carry a mask (editable) or baked-in transparency (destructive), and when you drag a layer aside, is there filled background underneath or a hole? Below is one of our own exports read back with an independent PSD library rather than with our own writer — which is the only version of this evidence worth anything.

What is actually in the file

The PSD arrives with numbered groups — 10 BG, 20 Objects, 30 FX — a named layer per element you separated, and every lifted selection attached as a real non-destructive layer mask rather than baked into the alpha, so you can refine a cut with a brush in Photoshop instead of living with it. Regions the AI reconstructed are prefixed AI_Fill_ in the layer name, so provenance is never a guess. It opens in Photoshop, Affinity Photo, Photopea, GIMP and Krita, and if your pipeline ends in an open-source tool you can export OpenRaster instead and skip PSD entirely.

What it costs to rebuild one

The editor, the click-to-select AI, edge cleanup and the ORA and PNG-stack exports are free and unlimited, because they run on your own hardware — the JPG never leaves your machine for any of that. PSD export is free three times a day. The server steps are the metered ones and each is labelled with its price before you click: the strongest subject model is 1 credit, auto-detect starts at 1 credit and adds 1 for each object whose outline needs rebuilding by the strongest model (capped at 8 per run), and filling the background behind a lifted object starts at 2 credits — large or structured holes route through an HD pass that costs more, currently 6 credits. Credits come with a monthly plan, or as a one-off pack whose credits never expire — see the pricing page for current prices in your currency. Nothing is ever watermarked or downscaled on any tier; the free tier caps quantity, never quality.