Stable Diffusion is a diffusion-based image-generation model that converts textual prompts into images. Enterprises can apply it to architecture, fashion, avatars, virtual try-on, medical training, e-commerce, and entertainment, while also using related workflows for image editing. section 4.0 section 4.1.4
Its logical architecture combines four main components:
- Text encoder: converts the user’s prompt into a vector representation. section 4.1.4
- UNet denoising autoencoder: reconstructs an image from latent space by removing noise through encoder–decoder processing. section 4.1.4
- Scheduler: guides the repeated reconstruction process that restores the original image. section 4.1.4
- Variational autoencoder: creates an image close to a normal distribution. section 4.1.4
Generating an image with the REST API
A Python client can call the Stability image-generation endpoint with engine ID "stable-diffusion-xl-1024-v1-0", a bearer API key, and the prompt "Laughing panda in the clouds eating bamboo." Requests use the v1 API in the demonstrated workflow; newer Stable Diffusion 3 models use a v2beta endpoint and a different engine ID. Body parameters such as image dimensions, text prompts, prompt weights, and cfg_scale steer generation. section 4.2.2
