Update README.md

README.md

@@ -1,30 +1,50 @@
-# Stable Layers — Inference
+---
+license: other
+license_name: stabilityai-community
+license_link: https://stability.ai/license
+base_model: Qwen/Qwen-Image-Layered
+library_name: peft
+pipeline_tag: image-to-image
+---
 
-Project Page: https://stability-ai.github.io/stable-layers.github.io/
+<h1 align="center">Stable Layers</h1>
 
-Paper: https://arxiv.org/abs/2605.30257
+<p align="center">
+  <strong>Decomposing Images into Editable RGBA Layers</strong>
+</p>
+ 
+<!-- Add a teaser image to the repo and it will render here. -->
+<!--
+<p align="center">
+  <img src="stable_layers_teaser.jpg" alt="Stable Layers Teaser" width="100%" />
+</p>
+-->
 
-Note: whilst the paper used gemini, this model was retrained using an open VLM, we found no performance loss.
+<p align="center">
+  <span><a href="https://arxiv.org/abs/2605.30257">Paper</a></span>&nbsp;&nbsp;&nbsp;
+  <span><a href="https://stability-ai.github.io/stable-layers.github.io/">Project Page</a></span>&nbsp;&nbsp;&nbsp;
+  <span><a href="https://huggingface.co/StabilityLabs/Stable-Layers">Model (Hugging Face)</a></span>&nbsp;&nbsp;&nbsp;
+  <span><a href="https://github.com/Stability-AI/Stable-Layers">Code</a></span>
+</p>
 
----
+## About
 
-## Recommended inference settings
-
-> ### **Heun sampler · 50 steps · CFG 1.0 · 640 px · 4 layers**
+**Stable Layers** decomposes a single RGB image into a small stack of back-to-front
+RGBA layers (background plus object layers) suitable for compositing and editing.
 
-| Setting | Value | Flag |
-|---|---|---|
-| **Sampler** | **Heun (2nd order)** | always used — not configurable |
-| **Steps** | **50** | `--steps 50` |
-| **CFG / guidance** | **1.0 (off)** | `--guidance-scale 1.0` |
-| Resolution | 640 px (max dim) | `--size 640` |
-| Layers | 4 | `--num-layers 4` |
+This repository hosts the LoRA adapter over `Qwen/Qwen-Image-Layered`. Note: whilst
+the paper used Gemini, this model was retrained using an open VLM, and we found no
+performance loss.
 
-**Note:** with using a high resolution or lower steps or not heun will garble the results.
+## Model Artifacts
 
----
+This repository **is** the LoRA adapter (`adapter_config.json`,
+`adapter_model.safetensors`). The base model is pulled from Hugging Face
+automatically (`Qwen/Qwen-Image-Layered`). See the
+[code repository](https://github.com/Stability-AI/Stable-Layers) for the inference
+script.
 
-## Install
+## Installation
 
 ```bash
 pip install torch diffusers transformers peft pillow numpy
@@ -34,16 +54,22 @@ Tested with `torch 2.11`, `diffusers 0.37`, `transformers 5.5`, `peft 0.18`.
 Requires **one GPU** — the base model is ~40 GB in bf16, so an 80 GB-class card
 (A100-80 / H100 / H200) is comfortable.
 
----
+## Recommended Inference Settings
 
-## Weights
+> ### **Heun sampler · 50 steps · CFG 1.0 · 640 px · 4 layers**
 
-The base model is pulled from HuggingFace automatically
-(`Qwen/Qwen-Image-Layered`). Lora found locally.
+| Setting | Value | Flag |
+|---|---|---|
+| **Sampler** | **Heun (2nd order)** | always used — not configurable |
+| **Steps** | **50** | `--steps 50` |
+| **CFG / guidance** | **1.0 (off)** | `--guidance-scale 1.0` |
+| Resolution | 640 px (max dim) | `--size 640` |
+| Layers | 4 | `--num-layers 4` |
 
----
+**Note:** using a higher resolution, lower steps, or a non-Heun sampler will garble
+the results.
 
-## Usage
+## Quickstart
 
 ```bash
 # single image
@@ -59,9 +85,7 @@ python decompose.py --input images/ --output results/ --transparent
 python decompose.py --input photo.png --output results/ --lora ./model
 ```
 
----
-
-## Output
+## Outputs
 
 ```
 results/<image_name>/
@@ -81,9 +105,7 @@ By default layers are composited onto **white** (easy to eyeball). Pass
 `--transparent` to get **RGBA with real alpha**, which is what you want when
 importing into an editor or compositing them yourself.
 
----
-
-## Notes
+**Notes:**
 
 - **Reproducible:** noise is seeded per image as `seed + image_index`
   (`--seed 42` by default), so the same inputs give the same outputs.
@@ -92,3 +114,25 @@ importing into an editor or compositing them yourself.
   override with `--prompt` if you want.
 - **Aspect ratio** is preserved; the longest side is scaled to `--size` and both
   dimensions are rounded to multiples of 16.
+
+## License
+
+This code and model usage are subject to Stability AI Community License terms.
+
+For individuals or organizations generating annual revenue of USD 1,000,000 (or local
+currency equivalent) or more, commercial usage requires an enterprise license from
+Stability AI.
+
+- License details: https://stability.ai/license
+- Enterprise request: https://stability.ai/enterprise
+
+## Citation
+
+```
+@article{stablelayers2026,
+  author  = {},
+  title   = {Stable Layers: Decomposing Images into Editable RGBA Layers},
+  journal = {arXiv preprint arXiv:2605.30257},
+  year    = {2026}
+}
+```