The Stable Diffusion model was developed by a team of researchers and engineers from CompVis, Stability AI, runway, and LAION. They created the StableDiffusionImg2ImgPipeline, which allows users to input a text prompt and an initial image to generate new images using Stable Diffusion. The original codebase for this model can be found at CampVis/stable-diffusion, and it is compatible with all Stable Diffusion checkpoints for Text-to-Image. This pipeline utilizes a mechanism called diffusion-denoising, which was introduced in SDEdit (a technique for guided image synthesis and editing using stochastic differential equations proposed by Chenlin Meng, Yutong He, Yang Song, Jiaming Song, Jiajun Wu, Jun-Yan Zhu, and Stefano Ermon).
The developers of the model used the LAION-5B dataset and its subsets to train the model. They filtered the training data using LAION's NSFW detector, keeping only the data with a "p_unsafe" score of 0.1 or lower (a conservative measure).
Stable Diffusion Img2Img can be used to enhance and modify images to various styles, based on the original image. The strength of the initial image will determine how close or farther they generated image will be from the original image.
For more detailed instructions, refer to the API documentation and resources available on Github.