See-Through: Single-image Layer Decomposition
Transform static anime illustrations into manipulatable 2.5D models
We introduce a framework that automates the transformation of static anime illustrations into manipulatable 2.5D models. Our approach decomposes a single image into fully inpainted, semantically distinct layers with inferred drawing orders — up to 23 layers including hair, face, eyes, clothing, accessories, and more.
How it Works
See-through runs a full pipeline that applies the LayerDiff 3D model for transparent layer generation and the fine-tuned Marigold model for pseudo-depth inference, stratifying the character into up to 23 semantic layers to export a layered PSD file.
LayerDiff 3D
Diffusion-based transparent layer generation using SDXL.
Marigold Depth
Pseudo-depth estimation fine-tuned for anime.
Semantic Parsing
Semantic body part segmentation for precise layer stratification.
Get Production-Ready Character Layers Today
This research demo can take minutes to run and produces experimental output. For reliable, ready-to-use layers — hair, face, clothing, and accessories — use our production Character into Layers tool.
Try Character into Layers