README.md

---
license: creativeml-openrail-m
tags:
- stable-diffusion
- sdxl
- layerdiffusion
- attention-injection
- diff-patches
- comfyui
---

# LayerDiffusion-SDXL-Attn-Diff

This repository contains a converted version of the `layer_xl_transparent_attn.safetensors` model from [LayerDiffusion/layerdiffusion-v1](https://huggingface.co/LayerDiffusion/layerdiffusion-v1)

## Model Details
- **Base Model**: `layer_xl_transparent_attn.safetensors` from [LayerDiffusion/layerdiffusion-v1](https://huggingface.co/LayerDiffusion/layerdiffusion-v1)
- **Conversion**: LoRA patches (`::lora::0`, `::lora::1`) are converted to `diff` patches (`::diff::0`) using matrix multiplication (`torch.mm`).
- **License**: [CreativeML Open RAIL-M](https://huggingface.co/spaces/license/cml-openrail-m), same as the original model.
- **File**: `layer_xl_transparent_attn.safetensors` (contains precomputed `diff` patches)

## Usage in ComfyUI
1. Download `layer_xl_transparent_attn.safetensors` and place it in `ComfyUI/models/layer_model/`.
2. Install the [ComfyUI-layerdiffuse](https://github.com/huchenlei/ComfyUI-layerdiffuse) custom node.
3. In your workflow, use the `LayeredDiffusionApply` node with `config` set to `SDXL, Attention Injection`.

## Credits
This model is a derivative of the original work by Lvmin Zhang and Maneesh Agrawala, as described in their paper:

> Zhang, L., & Agrawala, M. (2024). Transparent Image Layer Diffusion using Latent Transparency. *arXiv preprint arXiv:2402.17113*. [https://arxiv.org/abs/2402.17113](https://arxiv.org/abs/2402.17113)

- **Original Model**: [LayerDiffusion/layerdiffusion-v1](https://huggingface.co/LayerDiffusion/layerdiffusion-v1) by the LayerDiffusion team.

## License
This model is distributed under the [CreativeML Open RAIL-M](https://huggingface.co/spaces/license/cml-openrail-m) license, as required by the original model. Please review the license for usage terms, including non-commercial restrictions.