See-through: Layer Decomposition

See-through: Layer Decomposition

Splits one anime character illustration into separate, fully filled-in layers with depth order and gives you a layered PSD file. For illustrators, animators and researchers.

Made by 24yearsold on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: See-through: Layer Decomposition is a Gradio demo that takes one anime character illustration and splits it into separate semantic layers. Hidden parts of each layer are filled in, the layers come with a depth order, and the result is a layered PSD file. It runs on ZeroGPU hardware, so you can try it in a browser.

Who it suits: It suits illustrators, animators and researchers who want an illustration broken into editable layers without cutting it apart by hand.

Industry | Who uses it | How they use it | Example
Illustration | Artists | Turn a flat finished picture into editable layers | Upload a character and open the PSD
Animation | Animators | Get layers with depth order to start from | Move the separated parts of one character
AI research | Researchers | Try the See-through pipeline on new images | Compare the layers with the paper's examples

If your picture is not an anime character, you can skip it, since the model is made for that style.

Where to use it:

• A web browser, on the Space's page on Hugging Face
• A local copy of the files, run with Python under Windows, macOS or Linux
• An image editor that opens PSD files, for the result

Use case | Who | What they get back
Split an illustration | Artists | A layered PSD with the parts separated
Prepare for animation | Animators | Layers with depth order
Test the pipeline | Researchers | Results from the three-stage pipeline

How to set it up:

• A web browser, which is all the online page needs
• Python, for a local copy
• The Python packages listed in requirements.txt
• A character illustration, ideally square

1. Open the Space's page on Hugging Face to use it online. Nothing needs installing.
2. For a local copy, install the Hugging Face hub package and download the Space's files.
3. Move into the downloaded folder.
4. Install Gradio 6.12.0, then the packages from requirements.txt.
5. Start the app with python app.py.

pip install -U huggingface_hub
hf download 24yearsold/see-through-demo --repo-type space --local-dir see-through-demo
cd see-through-demo
pip install gradio==6.12.0
pip install -r requirements.txt
python app.py

How to use it: Upload an illustration, run it and download the PSD. Square pictures work best, and other shapes are cropped from the center.

1. Upload your character illustration.
2. Change the seed if you want a different result.
3. Press Submit and wait about one to two minutes.
4. Download the layered PSD file.

Tool | What it does
LayerDiff 3D | Generates transparent layers with a diffusion model built on SDXL
Marigold Depth | Estimates a rough depth map tuned for anime pictures
Post-processing | Sorts the layers into a stack and exports the PSD

Example: Upload a square illustration of one character and press Submit. After a minute or two, you download a PSD in which the hair, clothes and other parts sit on their own layers.

For newcomers and experts: Start with a clean, square picture of one character and leave the seed alone. Open the PSD in your editor and look at the layers.

The demo runs the full See-through pipeline, which joins layer generation, depth estimation and layer stacking. It uses a newer model checkpoint than the paper, so results may not match the paper's exactly. The repository also holds a data pipeline README and notes on PSD stratification.

You upload a character illustration
LayerDiff 3D makes the transparent layers
Marigold Depth estimates the depth order
Post-processing exports a layered PSD

Good to know: The demo uses a newer checkpoint than the paper, so it may not reproduce the paper's numbers exactly.

• License: MIT
• Cost to run: none named, and the online copy runs on ZeroGPU hardware
• Data it reaches: the illustration you upload and the models the Space names
• Not for: pictures that are not anime characters, or work that must match the paper's results exactly

On Hugging Face, 13 November 2026

• Likes: 199
• Built with: gradio
• License: mit

Links

• Space: https://huggingface.co/spaces/24yearsold/see-through-demo
• Files: https://huggingface.co/spaces/24yearsold/see-through-demo/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.