Stable Diffusion 2.1

Input

Prompt

Steps

Scheduler

Seed

Output

output

Stable Diffusion 2.1

Stable Diffusion is a type of latent diffusion model that can generate images from text. It was created by a team of researchers and engineers from CompVis, Stability AI, and LAION. Stable Diffusion v2 is a specific version of the model architecture. It utilizes a downsampling-factor 8 autoencoder with an 865M UNet and OpenCLIP ViT-H/14 text encoder for the diffusion model. When using the SD 2-v model, it produces 768x768 px images. It uses the penultimate text embeddings from a CLIP ViT-H/14 text encoder to condition the generation process.

Weights

Stability AI currently provides the following checkpoints:

  • sd-v1-1.ckpt: 237k steps at resolution 256x256 on laion2B-en. 194k steps at resolution 512x512 on laion-high-resolution (170M examples from LAION-5B with resolution >= 1024x1024).
  • sd-v1-2.ckpt: Resumed from sd-v1-1.ckpt. 515k steps at resolution 512x512 on laion-aesthetics v2 5+ (a subset of laion2B-en with estimated aesthetics score > 5.0, and additionally filtered to images with an original size >= 512x512, and an estimated watermark probability < 0.5. The watermark estimate is from the LAION-5B metadata, the aesthetics score is estimated using the LAION-Aesthetics Predictor V2).
  • sd-v1-3.ckpt: Resumed from sd-v1-2.ckpt. 195k steps at resolution 512x512 on "laion-aesthetics v2 5+" and 10% dropping of the text-conditioning to improve classifier-free guidance sampling.
  • sd-v1-4.ckpt: Resumed from sd-v1-2.ckpt. 225k steps at resolution 512x512 on "laion-aesthetics v2 5+" and 10% dropping of the text-conditioning to improve classifier-free guidance sampling.

Getting Started

For more detailed instructions, refer to the API documentation and resources available on Github.

Github

https://github.com/Stability-AI/stablediffusion

License

Apache License 2.0