Advanced · Generative AI

Consistency Model

Visual diagram · (in preparation) · Math · (in preparation) · Worked example · 3 difficulty levels.

TL;DR. A diffusion-derived model trained so any point on a noise trajectory maps to the same clean sample, enabling 1–4 step generation.

Technical Definition

A diffusion-derived model trained so any point on a noise trajectory maps to the same clean sample, enabling 1–4 step generation.

How it works

Consistency models (Song et al., 2023) train a network f(x_t, t) to be invariant along the probability-flow ODE trajectory, so f(x_t, t) ≈ f(x_t', t') for any (t, t') on the same trajectory. Once trained — either from scratch (CT) or distilled from a diffusion teacher (CD) — they generate high-quality samples in a single function evaluation, enabling real-time image and video synthesis (LCM, SDXL-Turbo).

Related Concepts

• Diffusion Model — A generative model that learns to create data by reversing a gradual noising process, producing high-quality images and audio.
• Distillation — Compressing a large AI model into a smaller one that mimics the original's performance, improving efficiency.
• Flow Matching — A generative-modeling framework that learns a velocity field transporting samples from noise to data along a chosen probability path.