rectified flow
In plain wordsA generator trained to move random noise toward real data along straight lines, which can then be straightened further so that a picture comes out in one or a few steps.
A training method for generators defined by an ordinary differential equation, built on straight lines: each source sample x₀ (usually noise) is paired with a target sample x₁ (usually data), and a velocity field is fitted so that at any point on the segment between them it points along x₁ − x₀, the same objective as linear flow matchingflow matching, n.: A way to train generators that learns, for every point between random noise and real data, which direction to move, so that noise can be steered smoothly into new examples.. Because the learned flow averages over many crossing straight lines, its trajectories are curved; the 'reflow' procedure regenerates noise–data pairs with the trained model and retrains on them; because straight lines between these pairs rarely cross, each round makes the flow straighter. A nearly straight flow needs very few integration steps, sometimes only one, which also eases distillation into one-step generators (as in InstaFlow, 2023). Stable Diffusion 3 and FLUXFLUX, prop. n.: A family of image-generating AI models from the German start-up Black Forest Labs, founded by researchers behind Stable Diffusion; some versions can be downloaded and run freely. describe themselves as rectified flow transformers, using the straight-path objective with a timestep sampling that emphasizes intermediate noise levels.
Origin
Introduced by Xingchao Liu, Chengyue Gong and Qiang Liu of the University of Texas at Austin in 'Flow Straight and Fast' (posted September 2022; ICLR 2023). Patrick Esser and colleagues' 'Scaling Rectified Flow Transformers for High-Resolution Image Synthesis' (March 2024), the Stable Diffusion 3 paper, brought it to large-scale text-to-image generation.
Related
- flow matchingflow matching, n.: A way to train generators that learns, for every point between random noise and real data, which direction to move, so that noise can be steered smoothly into new examples.
- diffusion transformerdiffusion transformer, n.: A diffusion image or video generator whose core network is a transformer — the architecture behind chatbots — instead of the older image-specific design, which lets it improve more predictably as it grows.
- Stable DiffusionStable Diffusion, prop. n.: A widely used image-generating AI whose model files were released for anyone to download, so people could run it on their own computers and adapt it.
- FLUXFLUX, prop. n.: A family of image-generating AI models from the German start-up Black Forest Labs, founded by researchers behind Stable Diffusion; some versions can be downloaded and run freely.
- consistency modelconsistency model, n.: A generator trained to jump from noise straight to a finished image in one step, instead of the dozens of small steps a diffusion model normally takes.
- diffusion distillationdiffusion distillation, n.: Training a fast image generator to imitate a slow diffusion model, so that a picture that once took fifty steps can be made in one to four.
Sources
- Liu, X., Gong, C. & Liu, Q. (2023). Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow. ICLR 2023. arXiv:2209.03003. arxiv.org/abs/2209.03003
- Esser, P. et al. (2024). Scaling Rectified Flow Transformers for High-Resolution Image Synthesis. arXiv:2403.03206. arxiv.org/abs/2403.03206
Mentioned in
Stable DiffusionStable Diffusion, prop. n.: A widely used image-generating AI whose model files were released for anyone to download, so people could run it on their own computers and adapt it., flow matchingflow matching, n.: A way to train generators that learns, for every point between random noise and real data, which direction to move, so that noise can be steered smoothly into new examples., FLUXFLUX, prop. n.: A family of image-generating AI models from the German start-up Black Forest Labs, founded by researchers behind Stable Diffusion; some versions can be downloaded and run freely., consistency modelconsistency model, n.: A generator trained to jump from noise straight to a finished image in one step, instead of the dozens of small steps a diffusion model normally takes., diffusion distillationdiffusion distillation, n.: Training a fast image generator to imitate a slow diffusion model, so that a picture that once took fifty steps can be made in one to four., noise schedulenoise schedule, n.: The plan for how much static is mixed into training examples at each step of a diffusion model, from almost none to total noise; its shape affects quality and speed.