abhirupghosh1992/stable-diffusion repository overview

⁠Stable Diffusion

Stable diffusion in a text-to-image generation model which is open source. The details can be found in https://github.com/CompVis/stable-diffusion⁠. Stable diffusion is implemented by huggingface in their pipeline and the details of which can be found in https://huggingface.co/spaces/stabilityai/stable-diffusion⁠ . The primary goal of stable diffusion is to generate images based on textual prompts given by the user. Hosting the model is a challenge since it requires dedicated CUDA enabled hardware to perform efficiently.

This docker container contains the weights and all the necessary libraries to host stable diffusion in a CUDA environment. It is also integrated with a fast API server which can respond to the following requests.

get :- http://host:port/getimage?prompt=prompt%20for%20stable%20diffusion%20model⁠

This request returns a image.

The recommended specs to host this image is a Machine with NVIDIA CUDA enabled graphics card and 16 GB of RAM and 16 GB of VRAM.

The image is currently hosted on a GCP n1-standard-4 compute engine with c0-deeplearning-common-cu110-v20220806-debian-10 as the OS Image and having a NVIDIA TESLA T4 accelerator