Open‑Source AI Photoshop: Alibaba’s Qwen‑Image‑Layered Enables One‑Click Smart Layering

Model Overview

Qwen‑Image‑Layered is an open‑source diffusion model that takes a single RGB image and outputs multiple physically isolated RGBA layers, enabling native editability comparable to Photoshop.

Key Features

Photoshop‑level layering : Generates fully independent RGBA layers.

Prompt‑controlled structure : Users can specify 3‑10 layers, from coarse layout to fine details.

Unlimited decomposition : Layers can be recursively split for deeper detail.

Resources: https://huggingface.co/Qwen/Qwen-Image-Layered, https://modelscope.cn/models/Qwen/Qwen-Image-Layered, https://github.com/QwenLM/Qwen-Image-Layered, https://qwen.ai/blog?id=qwen-image-layered, https://arxiv.org/abs/2512.15603

Demonstrations

Complex scenes are automatically split into separate layers; side‑by‑side comparisons show the original image and the extracted layers. Prompt‑driven examples illustrate how textual instructions shape the layer hierarchy.

Use Cases Highlighted by the Community

1. De‑constructing PPT or presentation slides

AI‑generated slide images are split into independent pictures, text, and graphic elements for direct editing and layout adjustment in PowerPoint.

Quentin Lhoest 🤗: “It arrives just in time! Open‑source will fix Google Slide AI.”

2. Game development – rapid scene element separation

Backgrounds, characters, foreground objects, and UI text are isolated, accelerating asset creation and iteration.

@nicekate8888: “Background and text separation is amazing – perfect for game development!”

3. Design layer management

Enables precise, local adjustments without affecting the whole composition, changing the “post‑edit” AI image workflow.

Maki@Sunwood AI Labs: “Turn a single image into multiple editable layers – a new world of image editing.”

How to Use

In ComfyUI

The model is available as a ComfyUI node. Large outputs (e.g., 1152×2016) are not quantized; on an RTX 5090 inference takes roughly 500 seconds but yields very high quality.

In Figma via Plugin

Use the ComfyUi Image Generator Cloud plugin, which includes the Qwen‑Image‑Layered model. Select an image, set parameters such as layer count, and the plugin returns a grouped layer set ready for Figma.

Limitations

Community feedback indicates strong performance on cartoons, illustrations, and design drafts, but weaker results on realistic photographs where background inpainting may look unnatural.

Amogh Vaishampayan: “Great on graphic designs, but realistic photos suffer from hole‑filling issues; combining with an inpainting model improves results.”

Pairing the model with a state‑of‑the‑art inpainting model can mitigate these artifacts.

Conclusion

Qwen‑Image‑Layered demonstrates AI‑driven understanding of image structure by decomposing images into editable hierarchies. While still imperfect on complex textures, its open‑source nature and integration with tools such as ComfyUI and Figma facilitate rapid ecosystem growth and new creative workflows.