stability-ai / stable-diffusion
A latent text-to-image diffusion model capable of generating photo-realistic images given any text input
-
- Version
- 22.04
- Commit
- 2286f7a162c66aad8c35c122a9f80f519d9ee20e
- Stable Diffusion 2.1. https://huggingface.co/stabilityai/stable-diffusion-2-1
- Optimized for speed with AI Template - very fast, now supports all input shapes up to 1024x1024
- fp16
- Can run on A100 GPUs, but not T4, due to lack of AI Template support
-
- Version
- 22.04
- Commit
- 2286f7a162c66aad8c35c122a9f80f519d9ee20e
- Stable Diffusion 1.5. https://huggingface.co/runwayml/stable-diffusion-v1-5
- Optimized for speed with AI Template - very fast, now supports all input shapes up to 1024x1024
- Can run on A100 GPUs, but not T4, due to lack of AI Template support
-
- Stable Diffusion 2.1. https://huggingface.co/stabilityai/stable-diffusion-2-1
- Integrates the CompVis/stable-diffusion-safety-checker to filter NSFW content.
- More Schedulers - DDIM, K_EULER, DPMSolverMultistep, K_EULER_ANCESTRAL, PNDM, KLMS
-
- Stable Diffusion 2.0. https://huggingface.co/stabilityai/stable-diffusion-2
- Removes PNDM, KLMS, K_EULER_ANCESTRAL from scheduler choices.
init_image
has been removed and moved to a separate model: https://replicate.com/stability-ai/stable-diffusion-img2imgmask
has been removed and moved to a separate model: https://replicate.com/stability-ai/stable-diffusion-inpainting
-
- Add negative prompt.
-
- Support for more image sizes.
- Support for arbitrary number of outputs up to 10.
- Fix multiple outputs not working.
-
- Use the img2img pipelines from diffusers, which produces better quality output when using the
init_image
input.
- Use the img2img pipelines from diffusers, which produces better quality output when using the
-
Stable Diffusion 1.5. The checkpoint was initialized with the weights of the Stable-Diffusion-v1-2 checkpoint and subsequently fine-tuned on 595k steps at resolution 512x512 on “laion-aesthetics v2 5+” and 10% dropping of the text-conditioning to improve classifier-free guidance sampling.
-
Stable Diffusion 1.4. The checkpoint was initialized with the weights of the Stable-Diffusion-v-1-2 checkpoint and subsequently fine-tuned on 225k steps at resolution 512x512 on “laion-aesthetics v2 5+” and 10% dropping of the text-conditioning to improve classifier-free guidance sampling.