Views
No views yet
| Version | Size | VRAM |
|---|---|---|
| BF16 (original) | ~55 GB | ~55 GB |
| 6-bit (this) | ~31 GB | ~29 GB |
1# Install from the PR
2git clone https://github.com/zimengxiong/mflux.git
3cd mflux
4uv sync1uv run flux-generate-qwen-layered \
2 --image input.png \
3 --layers 4 \
4 --steps 50 \
5 -q 6 \
6 --output-dir ./layerslayer_0.png, layer_1.png, etc.) with transparency.| Parameter | Description | Default |
|---|---|---|
--image | Input image path | Required |
--layers | Number of layers to decompose | 4 |
--steps | Inference steps | 50 |
-q | Quantization (4, 6, or 8-bit) | None (BF16) |
--resolution | Resolution bucket (640 or 1024) | 640 |
--output-dir | Output directory for layers | ./ |
[layer, height, width])1@misc{yin2025qwenimagelayered,
2 title={Qwen-Image-Layered: Towards Inherent Editability via Layer Decomposition},
3 author={Shengming Yin, Zekai Zhang, Zecheng Tang, Kaiyuan Gao, Xiao Xu, Kun Yan, Jiahao Li, Yilei Chen, Yuxiang Chen, Heung-Yeung Shum, Lionel M. Ni, Jingren Zhou, Junyang Lin, Chenfei Wu},
4 year={2025},
5 eprint={2512.15603},
6 archivePrefix={arXiv},
7 primaryClass={cs.CV},
8 url={https://arxiv.org/abs/2512.15603},
9}