Diffusion Model
The core idea behind AI image generation — the AI starts with pure noise and refines it into a picture.
A diffusion model is the type of AI behind image generation on PixAI. Unlike a human artist starting with a blank canvas, the AI starts with pure random noise — like TV static — and step by step removes that noise until your image emerges, guided the whole way by your Prompt.
Why this is worth knowing
This one idea explains most of the Advanced settings:
Sampling-Steps — how many noise-removal rounds the AI performs. More rounds, more refinement.
Sampler — the technique used for each round; different techniques produce different looks.
Seed_(concept) — which starting noise you get. Same starting noise + same everything else = same image.
CFG-Scale — how strongly your prompt steers each round.
Two flavors on PixAI
All PixAI models are diffusion models; they differ in their internal design. Models based on Stable-Diffusion (SD 1.5, SDXL) power most of the catalog, while PixAI's newest in-house models, Tsubaki and Serin, use the newer Diffusion Transformer design — see DiT-Model.