Skip to content

SYSTEM Cited by 1 source

Stable Diffusion

Definition

Stable Diffusion (Stability AI, first public release August 2022) is an open-source latent-diffusion text-to-image model that generates images by iteratively denoising from Gaussian noise in a compressed latent space, conditioned on a text prompt via a CLIP-derived text encoder. Unlike earlier text-to-image diffusion models that operated in pixel space, Stable Diffusion operates in a compressed latent space (via a pre-trained VAE), dramatically reducing the compute required per image and making both inference and fine-tuning tractable on consumer-grade GPUs.

Stable Diffusion's open-source release + permissive license made it the de facto base model for the fine-tuning ecosystem — custom art styles, character models, product-specific models, and countless derivative checkpoints are all Stable Diffusion fine-tunes.

Why the sysdesign-wiki cares

Stable Diffusion is a canonical fine-tuning base model for DreamBooth-style personalisation. The wiki's interest is serving-infra + platform-level usage, not the ML architecture itself — the upstream papers cover the latent-diffusion architecture, noise schedule, and UNet details.

Stub page — expand as more Stable-Diffusion-internals or -serving-infra sources land on the wiki.

Seen in

Last updated · 319 distilled / 1,201 read