Slam Lora Sdxl

🚀 Sub-path Linear Approximation Model (SLAM) LoRA: SDXL

This project presents a distilled model from stabilityai/stable-diffusion-xl-base-1.0 using the proposed Sub-path Linear Approximation Model, which significantly reduces the number of inference steps to just 2 - 4 steps.

🚀 Quick Start

The checkpoint is a distilled version from stabilityai/stable-diffusion-xl-base-1.0 with our proposed Sub-path Linear Approximation Model. This model reduces the number of inference steps to only between 2 - 4 steps.

✨ Features

• Reduced Inference Steps: Significantly cuts down the inference steps to 2 - 4 steps.
• Compatibility: Compatible with LCMScheduler.

📦 Installation

First, install the latest version of the Diffusers library as well as peft, accelerate and transformers.

pip install --upgrade pip
pip install --upgrade diffusers transformers accelerate peft

💻 Usage Examples

Basic Usage

import torch
from diffusers import LCMScheduler, AutoPipelineForText2Image

model_id = "stabilityai/stable-diffusion-xl-base-1.0"
adapter_id = "alimama-creative/slam-lora-sdxl"

pipe = AutoPipelineForText2Image.from_pretrained(model_id, torch_dtype=torch.float16, variant="fp16")
pipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config)
pipe.to("cuda")

# load and fuse lcm lora
pipe.load_lora_weights(adapter_id)
pipe.fuse_lora()

prompt = "A brown teddy bear holding a glass vase in front of a grave."

image = pipe(prompt=prompt, num_inference_steps=4, guidance_scale=1.0).images[0]

Comparison

Compare with latent-consistency/lcm-lora-sdxl.

More Examples

📚 Documentation

• Paper: https://arxiv.org/abs/2404.13903
• Project Page: https://subpath-linear-approx-model.github.io/

📄 License

This project is licensed under the apache-2.0 license.