GPT-image-2's image generation is powerful, but what it produces is a merged "flat image"—if you want to change the text, swap the background color, or adjust layers, you can't. This open-source Skill solves that problem: it automatically splits AI-generated images into multiple layers and outputs an editable PSD file.

Hands-on results: the ChatGPT web app can only split 7 layers and leaves ghosting on text, while this Codex-based Skill split out 12 layers, with text on its own layer and a clean background with no ghosting.

Codex result: 12 layers

Preparation Before You Start

• Tools needed: a ChatGPT Plus account (for the web approach) or Codex (for the advanced approach)
• Open-source repo: github.com/binggandata/bggg-skills/tree/main/bggg-creator-image2psd
• Skill level: familiarity with basic Photoshop operations
• Estimated time: 10-15 minutes (web approach), 5 minutes (Codex Skill approach)

Option 1: ChatGPT Web App (7 layers, with ghosting)

Step 1: Connect Photoshop

In ChatGPT settings, open the connector settings and link the Photoshop app.

ChatGPT's entry point for connecting Photoshop

Click Connect to establish a channel between ChatGPT and Photoshop.

Step 2: Generate the Image and Split the Layers

Open a new chat and generate the image you need with GPT-image-2. Then use the following prompt to split the layers:

Next, I want to turn this image into a PSD to edit in Photoshop,
so first split the generated poster into several images.
Do not change their relative positions, and make the background white.

Step 3: Generate the PSD File

In the app picker, select Photoshop, then enter:

Assemble the images split above into a PSD file, one layer per image.
Note that some are graphics and some are text. Remove the white background. Give me the PSD file.

The PSD file output by ChatGPT

Tip: The web app's limitation is the 7-layer cap—complex images get ghosting under the title text, and text and background graphics may end up merged into the same layer.

Option 2: Codex Skill (12 layers, no ghosting)

Step 1: Install the Skill

Clone or download the Skill files from GitHub:

https://github.com/binggandata/bggg-skills/tree/main/bggg-creator-image2psd

Step 2: Use It in Codex

Open Codex and give it a description of what you need so it uses the Skill to process the image. Codex has a reasoning loop that automatically checks intermediate results as it runs the task; if it spots edge artifacts on a split layer, it automatically goes back and adjusts the parameters until the result is clean.

Codex self-checking and correcting during its reasoning process

Step 3: Verify the Result

After importing the final PSD output into Photoshop, check the following:

• Whether the layer count meets expectations (12 layers in testing)
• Whether there is ghosting under the title text (there should be none)
• Whether the text sits on its own layer

Comparing the Two Options

The same Skill also splits layers for e-commerce product images

More Skill Resources

This Skill and future ones will all be synced to this repository:

https://github.com/binggandata/bggg-skills

If your design workflow regularly involves turning AI-generated images into editable source files, this Skill can save you a huge amount of manual cutout and layering work.