Sd2.1 Base Zsnr Laionaes5

🚀 SD v2.1-base with Zero Terminal SNR (LAION Aesthetic 5+)

This is an official model for text-to-image generation, trained with zero terminal SNR on LAION aesthetic 5+ data. It's based on Stable Diffusion 2.1-base and is for research demonstration.

🚀 Quick Start

This model is the official implementation for the paper Common Diffusion Noise Schedules and Sample Steps are Flawed. It is trained using the zero terminal SNR schedule on LAION aesthetic 5+ data and finetuned from stabilityai/stable-diffusion-2-1-base. Note that this model is intended for research demonstration and not for production use.

💻 Usage Examples

Basic Usage

from diffusers import StableDiffusionPipeline
prompt = "A young girl smiling"
pipe = StableDiffusionPipeline.from_pretrained("ByteDance/sd2.1-base-zsnr-laionaes5").to("cuda")
pipe(prompt, guidance_scale=7.5, guidance_rescale=0.7).images[0].save("out.jpg")

📚 Documentation

Related Models

• bytedance/sd2.1-base-zsnr-laionaes5
• bytedance/sd2.1-base-zsnr-laionaes6
• bytedance/sd2.1-base-zsnr-laionaes6-perceptual

Cite as

@misc{lin2023common,
      title={Common Diffusion Noise Schedules and Sample Steps are Flawed}, 
      author={Shanchuan Lin and Bingchen Liu and Jiashi Li and Xiao Yang},
      year={2023},
      eprint={2305.08891},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

📄 License

The model is released under the OpenRAIL++ license.