M22.8 CONNECT THE MECHANISM
Learn a velocity field and integrate it into samples
What if every speck of noise simply travelled in a straight line to a picture? That idea trains today's fastest image generators. Take the steps yourself and watch the error shrink.
LESSON OVERVIEW12 min lesson
Lesson overview
What if every speck of noise simply travelled in a straight line to a picture? That idea trains today's fastest image generators. Take the steps yourself and watch the error shrink.
What you’ll explore
- Flow matching learns a time-dependent transport field along chosen probability paths; numerical solvers turn that field into samples, with step count, guidance, and evaluation defining the quality-cost tradeoff.
GO TO THE SOURCE
Original explanations, connected to the research.
Flow Matching for Generative Modeling (Lipman et al., 2022)Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow (Liu, Gong & Liu, 2022)Denoising Diffusion Implicit Models (Song, Meng & Ermon, 2020)Progressive Distillation for Fast Sampling of Diffusion Models (Salimans & Ho, 2022)Consistency Models (Song et al., 2023)Scaling Rectified Flow Transformers for High-Resolution Image Synthesis (Stable Diffusion 3; Esser et al., 2024)Suggest a correction
A precise note can make an explanation better.
Choose the scene and describe what needs attention. Download a feedback file to share through a channel you already use. This page does not send feedback or connect you with a reviewer.