---
license: openrail++
base_model: stabilityai/stable-diffusion-xl-base-1.0
library_name: diffusers
pipeline_tag: text-to-image
tags:
  - lora
  - sdxl
  - sdxl-lightning
---

# SDXL-Lightning 4-step LoRA (unmodified mirror)

This repository contains exactly one adapter checkpoint, copied without changes
from [ByteDance/SDXL-Lightning](https://huggingface.co/ByteDance/SDXL-Lightning).
It is a LoRA adapter, not a standalone image-generation model.

- Upstream revision: `c9a24f48e1c025556787b0c58dd67a091ece2e44`.
- File: `sdxl_lightning_4step_lora.safetensors`.
- Size: `393854592` bytes.
- SHA-256: `bf56cf2657efb15e465d81402ed481d1e11c4677e4bcce1bc11fe71ad8506b79`.
- Base model: `stabilityai/stable-diffusion-xl-base-1.0`.

Copyright (c) 2024 Bytedance Inc. Copyright (c) 2023 Stability AI.
The original CreativeML Open RAIL++-M license and its restrictions are preserved
in LICENSE.md. This mirror is not an official ByteDance repository.

## Loading

Attach this repository as a LoRA to an SDXL pipeline. For Diffusers:

```python
pipe.load_lora_weights(
    "YOUR_USERNAME/sdxl-lightning-4step-lora",
    weight_name="sdxl_lightning_4step_lora.safetensors",
)
pipe.fuse_lora()
```

Use 4 inference steps, guidance_scale=0, and EulerDiscreteScheduler with
timestep_spacing="trailing" and prediction_type="epsilon".
Keep the upstream base-model and adapter license terms.