SDXL Lightning ⚡: A Swift Advancement Over SDXL Turbo
You can generate high quality image within 1 to 8 steps
In late 2023, SDXL Turbo made its debut. Today, we herald a superior and swifter checkpoint: SDXL Lightning. This innovative model enables image generation in a fraction of the time, offering options ranging from 1 to 8 steps. With the incorporation of LoRA and Unet. SDXL Lightning empowers users to create images with unparalleled efficiency.
What is SDXL Lignhtning ⚡?
SDXL Lightning represents a breakthrough in text-to-image generation, boasting lightning-fast speeds and exceptional quality. Capable of producing high-resolution 1024px images in just a few steps, this model is detailed in our research paper, SDXL-Lightning: Progressive Adversarial Diffusion Distillation, which is openly accessible as part of our research initiative.
These models are derived from stabilityai/stable-diffusion-xl-base-1.0, a repository housing checkpoints for 1-step, 2-step, 4-step and 8-step distilled models. The image generation quality achieved by SDXL Lightning is truly remarkable, with the 1-step model offering an experimental edge.
Furthermore, both full UNet and LoRA checkpoints are provided. While the full UNet models prioritize image quality, the LoRA models offer versatility, allowing integration with other base models for diverse applications.
Online Demo ⚡
You can try it online in huggingface website.
Input a prompt and receive an image within seconds, followed by a high-quality output. For instance, simply input 1girl as a prompt.
Local Testing with ComfyUI ⚡
You can download Model for different steps or methods.
If you using normal SDXL Lightning checkpoint with 4 steps, download sdxl_lightning_4step.safetensors
If you need Unet SDXL Lightning checkpoint with 8 steps, download sdxl_lightning_8step_unet.safetensors
If you need SDXL LoRA with 2 steps, download sdxl_lightning_2step_lora.safetensors
You also can download workflow from official huggingface website.
For Lightning Checkpoint or Unet Checkpoint, you can use following workflow to generate images.
For Lightning LoRA, can use following workflow.
In Civitai already have many Model training base on SDXL Lightning, you can simply download it and using in WebUI like Automatic1111.
In summary, the evolution of Stable Diffusion has been marked by significant advancements with the introduction of LCM, SDXL Turbo and now SDXL Lightning. With each iteration, the speed of Stable Diffusion has steadily increased. This progress signifies that soon, generating high-quality images will be effortlessly achievable on any computer.
For All Stable Diffusion Lovers
Thank you for being a part of the Stable Diffusion community!
