Flow Matching
A modern variant framework for diffusion models that learns a deterministic velocity field to continuously map a noise distribution to a data distribution. It has been adopted by recent text-to-image models including Stable Diffusion 3 and Flux.
Core Facts
Flow Matching continuously maps a noise distribution to a data distribution by learning a deterministic velocity field, and both Stable Diffusion 3 and Flux adopt this framework.
Flow Matching was proposed by Lipman et al. in 2022 and directly learns a deterministic ODE flow that maps a noise distribution to a data distribution.
Flow matching is a generative model training method that directly learns a deterministic ODE path from the noise distribution to the data distribution, defines a linear interpolation path xt=(1-t)x₀+tx₁, and uses a neural network to learn to predict the velocity vector at every point along that path.
Timeline (last 90 days)
In the second attempt to build a joint video-action model, the team used the pretrained video prediction model Cosmos Predict 2.5-2B as the backbone network, extracting intermediate latent features during the video generation denoising process, and feeding them together with the robot's proprioceptive state and noisy action tokens into the Gemma Action Head, jointly generating denoised action control tokens through Flow Matching.
Flow matching models generate images by learning a continuous probability flow (ODE trajectory) from the noise distribution to the data distribution, and are generally more stable to train and require fewer inference steps.
π0 introduces an action decoding head based on Flow Matching, which outputs high-frequency, continuous dexterous manipulation actions while retaining language understanding capabilities.
MeanFlow is a generative modeling paradigm based on flow matching that parameterizes the average velocity of an entire trajectory as a single network output, theoretically enabling high-quality image synthesis in a single forward inference step.
DenoFlow is a method for denoising steady-state visual evoked potential (SSVEP) EEG signals, adopting a Flow Matching modeling approach.
Rectified Flow is a specific implementation of flow matching that straightens the path between the source distribution and the target distribution into a straight line, enabling fewer integration steps and higher generation efficiency.
MintFlow is a training-free constrained sampling framework that reformulates the constraint imposition process as minimal intervention on pre-trained flow trajectories.
Compared to score-function-based diffusion models, the flow matching training objective is more concise, and the trajectory can be designed as a rectified flow, enabling high-quality generation in fewer inference steps.
Alternative training objectives such as Flow Matching have demonstrated faster convergence speeds than traditional DDPM
The technical principle behind four-step generation typically comes from accelerated sampling paradigms such as Consistency Models or Flow Matching, which use distillation training to enable the model to accurately predict the denoising direction at sparse time steps.
3 more timeline events
All Facts (19)
Flow Matching continuously maps a noise distribution to a data distribution by learning a deterministic velocity field, and both Stable Diffusion 3 and Flux adopt this framework.
85%VerifiedFlow Matching was proposed by Lipman et al. in 2022 and directly learns a deterministic ODE flow that maps a noise distribution to a data distribution.
80%VerifiedFlow matching is a generative model training method that directly learns a deterministic ODE path from the noise distribution to the data distribution, defines a linear interpolation path xt=(1-t)x₀+tx₁, and uses a neural network to learn to predict the velocity vector at every point along that path.
75%Unverified新一代模型如FLUX、Z-Image由于架构改进，CFG值设为1甚至更低即可获得最佳效果，而传统Stable Diffusion模型通常使用7-12的CFG值
80%UnverifiedFlux基于流匹配（Flow Matching）技术，在写实质感和光影处理上表现突出
80%UnverifiedVeo 3基于Google的Flow Matching技术构建
70%UnverifiedThe reverse SDE of a diffusion process can be expressed as dX_t = [f(X_t,t) - g(t)²∇log p_t(X_t)]dt + g(t)dW̄_t.
60%UnverifiedIn the second attempt to build a joint video-action model, the team used the pretrained video prediction model Cosmos Predict 2.5-2B as the backbone network, extracting intermediate latent features during the video generation denoising process, and feeding them together with the robot's proprioceptive state and noisy action tokens into the Gemma Action Head, jointly generating denoised action control tokens through Flow Matching.
50%UnverifiedFlow matching models generate images by learning a continuous probability flow (ODE trajectory) from the noise distribution to the data distribution, and are generally more stable to train and require fewer inference steps.
50%Unverifiedπ0 introduces an action decoding head based on Flow Matching, which outputs high-frequency, continuous dexterous manipulation actions while retaining language understanding capabilities.
50%UnverifiedMeanFlow is a generative modeling paradigm based on flow matching that parameterizes the average velocity of an entire trajectory as a single network output, theoretically enabling high-quality image synthesis in a single forward inference step.
50%UnverifiedDenoFlow is a method for denoising steady-state visual evoked potential (SSVEP) EEG signals, adopting a Flow Matching modeling approach.
50%UnverifiedRectified Flow is a specific implementation of flow matching that straightens the path between the source distribution and the target distribution into a straight line, enabling fewer integration steps and higher generation efficiency.
50%UnverifiedMintFlow is a training-free constrained sampling framework that reformulates the constraint imposition process as minimal intervention on pre-trained flow trajectories.
50%UnverifiedCompared to score-function-based diffusion models, the flow matching training objective is more concise, and the trajectory can be designed as a rectified flow, enabling high-quality generation in fewer inference steps.
50%UnverifiedAlternative training objectives such as Flow Matching have demonstrated faster convergence speeds than traditional DDPM
50%UnverifiedThe technical principle behind four-step generation typically comes from accelerated sampling paradigms such as Consistency Models or Flow Matching, which use distillation training to enable the model to accurately predict the denoising direction at sparse time steps.
50%UnverifiedPhysical Intelligence's π0 adopts a Flow Matching-based continuous action generation scheme for dexterous manipulation tasks.
50%UnverifiedFlow Matching由Lipman等人于2022年提出，将离散时间步推广至连续时间流形，以确定性ODE求解器替代随机采样
50%