Image to PSD with layers: split one picture into editable layers with AI

Upload it in the studio with Layer decomposition on. Seedream 5.0 Flash detects the separate elements, rebuilds the background behind them and returns each element as a transparent PNG with its position; YingTu stacks them into a layered PSD.

A converter wraps the whole picture in a PSD, so it is still one layer. Layer splitting separates the title, subject and decorations into their own layers and fills in the background, so you can move or replace each part.

$0.018 per output image, with the background counted as one; up to 17 outputs (background + 16 layers). Typical posters and covers land between $0.11 and $0.27.

Seedream 5.0 Flash is enough for most images at $0.018 each. On the same movie poster both models returned 5 layers; Seedream 5.0 Pro kept the ship's engine glow more completely, but cost $0.72 against $0.11 for Flash. Pick Pro when the image has a lot of glow, glass, smoke or other semi-transparent detail.

The $0.1 credit for new Gmail users covers regular Seedream generation but not layer splitting, since one split usually costs more than the credit. With your own LaoZhang key you pay per image.

Yes. The split itself returns lettering as images; choose More formats → PSD with editable text (beta) and YingTu reads each text layer (words, colors, stroke, glow, shadow, gradient) and writes a Photoshop text layer you can retype, with those effects as adjustable layer styles. Each text group keeps the original pixels (visible) next to the editable layer (hidden) — turn it on to edit. Common fonts such as Source Han Sans/Serif and Arial stand in for the original lettering; Photoshop asks to update the text layers when the file opens, so choose Update. Reading the text costs about $0.01. To change the words but keep the original lettering, select the text layer and click Retype: AI redraws the new words in the same letter shapes, colors, stroke and glow and puts them back in place ($0.018 per try on Seedream 5.0 Flash; undo or switch models if needed).

The canvas matches the rebuilt background, set by the 1K, 1.5K or 2K option. A 2K portrait image is about 1792×2400 and the PSD is usually around 20 MB.

One call returns at most 16 elements and leaves the rest painted into the background, and neighbouring elements often come back as one layer (an icon, a number and a button, say). So YingTu asks the model to separate every character, object, icon and piece of text; when a run comes back full it splits the background again automatically, up to 4 rounds, and you can stop it at any time. If a layer is still too coarse, click Split on it: the parts go into a group under it, you can keep going level by level, and everything exports together as one PSD or ZIP. Hair, smoke and glass are hard to separate cleanly.

The image is sent to the LaoZhang API for processing. The layers and the PSD are created and kept in your browser; YingTu servers do not store your images.

Yes, PSD stores layers, and no, you do not need the source file: the PSD exported here is built from your flat image, with the rebuilt background as the bottom layer and every element on its own named pixel layer in its original position. It is a standard RGB layered PSD, so it opens as separate layers in software that supports PSD layers. A plain JPG or PNG has only one layer, which is why a format converter alone cannot give you this.