---
title: One-Step Generative Transport
url: https://www.emergentmind.com/topics/one-step-generative-transport
type: topic
---

# One-Step Generative Transport

Searching arXiv for the cited one-step generative transport papers to ground the article in current metadata.
arXiv search query: 2604.26065 OR 2604.06413 OR 2605.11755 OR 2605.22507 OR 2512.15657 OR 2509.22592 OR 2505.13447 OR 2606.21030 OR 2606.22752 OR 2603.10701 OR 2606.22696 OR 2603.12480 OR 2603.14798 OR 2605.21388 OR 2606.08953 OR 2405.05512
One-step generative transport denotes a family of generative methods that replace the tens to hundreds of denoising or ODE evaluations used by diffusion, score, and standard flow-matching models with a single forward pass, a single large transport step, or a directly learned endpoint map. Across the recent literature, the common objective is to transport a simple, approximate, or condition-specific source distribution to a target law without iterative sampling, while retaining multimodality, calibration, and bounded latency. The field now spans direct map learning, average-velocity and solution-map models, local-transport conditioning, constrained posterior transport, and Wasserstein- or OT-guided formulations [2505.13447], [2604.06413], [2604.26065].

## 1. Conceptual foundations

A recurring diagnosis in this literature is that one-step generation is not intrinsically inaccurate; rather, it fails when the model is asked to solve a *global* transport problem in one shot. “FlowS: One-Step Motion Prediction via Local Transport Conditioning” argues that “single-step integration is accurate when the underlying transport problem is local,” and attributes one-step failure to the combination of long displacement and multimodal mode discovery from a scene-agnostic base such as \(Z\sim\mathcal N(0,I)\) [2604.26065]. This viewpoint shifts the question from whether one step is too coarse to whether the transport geometry has been conditioned into a regime where one large step is faithful.

A complementary diagnosis appears in “ODE-free Neural Flow Matching for One-Step Generative Modeling.” There, the central obstruction is not path length but *inconsistent coupling*: if source samples and target samples are paired independently, the squared-loss minimizer becomes a regression-to-the-mean solution, and the learned endpoint collapses to \(\mathbb E_{p_1}[\mathbf x_1]\) [2604.06413]. This makes one-step transport a coupling-sensitive problem. The literature therefore converges on two broad principles: one-step transport becomes plausible when the source is already close to the target manifold, and when the transport supervision preserves coherent source–target pairings.

These principles explain why many recent one-step models depart from the classical “Gaussian noise to data” template. Some begin from learned anchors near plausible futures, some begin from the observed corrupted input itself, some restrict transport to a measurement-consistent affine subspace, and some replace white noise with a prior-aligned Gaussian reference. The shared implication is that source design is part of the transport problem, not merely an implementation detail [2603.10701], [2606.22696], [2603.14798].

## 2. Mathematical formulations

One major formulation learns the **endpoint map directly**. In OT-NFM, the flow map is parameterized as
\[
\mathbf x_t = F_\theta(t,\mathbf x_0), \qquad F_\theta(0,\mathbf x_0)=\mathbf x_0,
\]
so one-step generation is simply \(F_\theta(1,\mathbf x_0)\) for \(\mathbf x_0\sim p_0\) [2604.06413]. “Characteristic Learning for Provable One Step Generation” instead learns a characteristic map \(\hat g_{0,T}\) derived from the probability transport ODE, so inference becomes \(Z_T=\hat g_{0,T}(Z_0)\) in one evaluation [2405.05512]. “Solution Flow Models for One-Step Generative Modeling” pushes the same idea further by learning the bi-time ODE solution map
\[
f_\theta(x_t,t,s),
\]
with one-step sampling given by \(f_\theta(x_1,1,0)\) [2512.15657].

A second formulation learns a **finite-interval or average velocity** rather than an infinitesimal field. “Mean Flows for One-step Generative Modeling” defines
\[
u(z_t,r,t)=\frac{1}{t-r}\int_r^t v(z_\tau,\tau)\,d\tau,
\]
so the exact finite-interval update is \(z_r=z_t-(t-r)u(z_t,r,t)\), and one-step generation is \(x=\epsilon-u_\theta(\epsilon,0,1)\) [2505.13447]. Variants of this pattern appear in AlphaFlowTSE for mixture-to-target speech transport, in OFP for action generation, and in NullFlow for inverse problems on measurement-consistent subspaces [2603.10701], [2603.12480], [2606.22696].

A third formulation keeps the flow interpretation but makes one-step sampling feasible by **changing the transport geometry**. FlowS replaces the base \(Z\) with learned scene-conditioned anchors \(A_k\) and then predicts a semigroup-consistent displacement
\[
Y_{\mathrm{pred},k}=A_k+s_\psi(A_k,0,1,c),
\]
so the single step acts as local correction rather than global relocation [2604.26065]. NullFlow similarly confines all trajectories to
\[
\mathcal M_{\mathbf y}=\{\mathbf x:A\mathbf x=\mathbf y\},
\]
with source samples \(\mathbf x_0=A^\dagger\mathbf y+P\boldsymbol\epsilon\); because the velocity lies in \(\mathrm{null}(A)\), the flow remains measurement-consistent without a separate data-fidelity step [2606.22696].

These formulations differ in parameterization, but all replace numerical integration of a local rule by a learned finite-time transport object.

## 3. Main design patterns

The recent literature repeatedly returns to a small number of design patterns.

| Design pattern | Representative papers | Core mechanism |
|---|---|---|
| Consistent source–target pairing | [2604.06413], [2509.22592] | Replace independent couplings with OT-based pairings |
| Finite-interval transport learning | [2505.13447], [2512.15657] | Learn average velocity or solution map directly |
| Local transport conditioning | [2604.26065], [2605.12939] | Start near plausible targets so one step is local refinement |
| Structural subspace restriction | [2606.22696], [2603.14798] | Choose source/reference on the correct feasible geometry |
| Self-consistency across scales or intervals | [2604.26065], [2603.10701], [2603.12480] | Enforce large-step coherence with smaller-step or teacher targets |

The first pattern is **optimal transport or coherent coupling**. OT-NFM proves that direct map regression under the product coupling leads to mean collapse, and fixes it with optimal transport pairings, minibatch OT, or LOOM [2604.06413]. OT-MF makes the same point inside the Mean Flow framework: by replacing independent pairing with OT-based coupling, one-step generators preserve the fidelity and diversity of the original multi-step flow more faithfully [2509.22592].

The second pattern is **learning the right transport quantity**. MeanFlow argues that one-step generation should target the finite-interval quantity actually needed at inference rather than the instantaneous tangent field [2505.13447]. SoFlow learns the ODE solution function itself and avoids JVP-based consistency constraints [2512.15657]. OFP transfers this principle to robot action generation by learning interval-averaged velocity over an OT action path, then enforcing self-consistency and self-guided sharpening [2603.12480].

The third pattern is **conditioning transport to be local or straight**. FlowS formalizes local transport conditioning by requiring learned anchors \(A_k\) to lie much closer to the ground-truth future than a Gaussian sample does, so one-step Euler error is directly suppressed [2604.26065]. DirectTryOn makes a parallel argument for virtual try-on: because conditional entropy is low once person and garment are fixed, the ideal transport should be close to straight, and one-step sampling and multi-step sampling are equivalent when the condition uniquely determines the target [2605.12939].

The fourth pattern is **source design as preconditioning**. In function-space Bayesian inversion, the reference cannot be naive white noise in the infinite-dimensional limit; “Preconditioned One-Step Generative Modeling for Bayesian Inverse Problems in Function Spaces” replaces it with a prior-aligned anisotropic Gaussian reference and proves Lipschitz regularity of the resulting one-step transport [2603.14798]. For path-dependent physical fields, a geometry-based non-Gaussian empirical source reduces crossings among conditional transport paths and makes one-step Euler sampling viable without distillation [2606.22752]. NullFlow makes the same move algebraically by randomizing only in the null space of the forward operator [2606.22696].

A broader theoretical unification appears in “Self-Consistent Generative Paths via Admissible Random Variational Transport,” which treats one-step, shortcut, and MeanFlow models as “large-step degeneracies of path self-consistency” and evaluates them by a random fixed-point path residual rather than endpoint matching alone [2606.08953].

## 4. Domain-specific instantiations

In **autonomous motion prediction**, one-step transport is driven by latency constraints. FlowS uses a learned prior that emits \(K\) calibrated anchor trajectories per agent and a step-consistent displacement field. On the Waymo Open Motion Dataset, the ensemble model achieves Soft mAP \(=0.4804\) and mAP \(=0.4703\) at about 75 FPS, while the single model reaches Soft mAP \(=0.4658\) and mAP \(=0.4512\) with about 13.25 ms/scene on a single A100 [2604.26065].

In **virtual try-on**, one-step transport exploits strong conditional structure. DirectTryOn argues that VTON outputs are highly constrained by person and garment inputs, so the transport path can be straightened by pure conditional transport, garment preservation loss, self-consistency, and then one-step distillation. Under the unpaired setting, it reports VITON-HD FID \(8.59\), KID \(0.56\), and DressCode FID \(5.08\), KID \(0.95\), with \(0.48\) s end-to-end inference on a PPU-810E [2605.12939].

In **image compression and reconstruction**, one-step transport often appears as a local latent correction. FlowCodec decodes a bitrate-constrained latent \(\hat l_0\), then applies one near-terminal update
\[
\hat l_1=\hat l_0+\beta_{\text{rate}}\Delta t\,v_\theta(\hat l_0,t=1-\epsilon,c=\varnothing),
\]
using a pretrained MMDiT prior, with trainable parameters below \(0.54\%\) of the generative backbone [2606.21030]. NullFlow instead turns reconstruction into posterior transport restricted to \(\mathcal M_{\mathbf y}\), so the learned flow never leaves the measurement-consistent subspace and needs no separate data-fidelity correction [2606.22696].

In **speech and robotics**, the source is often the observed signal or a temporally correlated warm start rather than Gaussian noise. AlphaFlowTSE transports directly from the observed mixture spectrum \(Y\) to the target speech \(S\), eliminating the need for mixture-ratio prediction and yielding one-step inference \(\hat S=Y+u_\theta(Y,0,1;E)\) [2603.10701]. OFP uses interval-averaged action transport, self-consistency, self-guidance, and warm start from the suffix of the previous action chunk to minimize transport distance in control space [2603.12480].

In **scientific computing**, several papers adapt one-step transport to structured physical laws. One-Step Flow Matching for path-dependent stress fields uses a geometry-informed empirical source and a latent spatiotemporal transformer to generate full \(20\)-frame \(256\times256\) stress-field sequences in one ODE step [2606.22752]. “Two-Step Diffusion” uses a different compromise: Stage I is a Meanflow-style one-step global transport, and Stage II is a near-identity corrector trained with a mini-batch \(W_2\) objective once the geometry has become local enough for stable OT [2601.20024]. This suggests that even when strict one-shot transport is insufficient, a global-then-local decomposition can retain one-step efficiency at the dominant stage.

## 5. Empirical landscape

On **ImageNet \(256\times256\)**, the strongest reported one-step number in the provided literature is W-Flow’s FID \(1.29\), with throughput on one H100 GPU of 77.88 images/sec for W-Flow-XL/2 and an abstract-level claim of approximately \(100\times\) faster sampling than multi-step diffusion models with similar FID scores [2605.11755]. MeanFlow reports FID \(3.43\) at 1-NFE for ImageNet \(256\times256\) trained from scratch, while SoFlow improves the matched DiT comparison further, reaching FID \(2.96\) for SoFlow-XL/2 at 1-NFE and \(2.66\) at 2-NFE [2505.13447], [2512.15657].

On **synthetic transport and small-scale image generation**, OT-NFM shows that one-step map learning can be competitive when coupling is coherent: on Gaussian \(\to\) 2-Moons it reports \(W_2^2=0.0230\pm0.0046\) at 1 NFE, outperforming 100-NFE OT-CFM on that task, and it qualitatively avoids the mean-collapse seen without OT pairings [2604.06413]. OT-MF similarly improves one-step Mean Flows across 2D distributions, MNIST, image-to-image translation, and point clouds by replacing independent couplings with OT-based ones [2509.22592].

On **reconstruction tasks**, NullFlow reaches LPIPS \(0.055\) with NFE \(=1\) on FFHQ inpainting, compared with LPIPS \(0.089\) for PnP-Flow at 500 NFE and LPIPS \(0.101\) for DiffPIR at 200 NFE, while still producing posterior samples from a single network evaluation [2606.22696].

On **robot manipulation**, OFP demonstrates that one-step transport can outperform multi-step action generation. Across 56 simulated manipulation tasks, one-step OFP reaches \(71.6\%\) average success, versus \(66.4\%\) for DP3 at 100 steps and \(59.8\%\) for a 100-step flow policy, while reducing action chunk generation time to \(17.58\) ms from \(3225.67\) ms and \(1865.72\) ms, respectively [2603.12480].

On **Bayesian inverse problems in function spaces**, the operator-learning one-step sampler with a prior-aligned reference produces a \(64\times64\) posterior sample in \(3.50\times10^{-4}\) s/sample and matches posterior mean and variance summaries within roughly \(10\%\) relative \(L^2\) error across Darcy, Advection, Reaction–Diffusion, and Navier–Stokes examples [2603.14798].

These results do not show a single dominant mechanism; rather, they show that one-step transport is empirically strongest when the task structure makes transport short, straight, conditionally low-entropy, or geometrically preconditioned.

## 6. Limitations, controversies, and open directions

A first limitation is that **not all one-step claims are equally strict**. “Generative Modeling by Value-Driven Transport” explicitly says that exact one-step generation is possible “in principle” because optimal OT paths are straight, but also states that the level of precision needed for such one-step generation is hard to achieve in practice; its strongest empirical result is efficient few-step transport, especially reducing \(100\) training steps to about \(10\) sampling steps with little loss [2605.22507]. This distinguishes true one-NFE endpoint-map methods from methods whose theory admits one-step sampling but whose evidence mainly supports few-step use.

A second limitation is that **source and coupling choice are often indispensable rather than optional**. OT-NFM proves that independent pairing induces mean collapse in direct map learning [2604.06413]. The function-space Bayesian paper shows that white-noise references become incompatible with the infinite-dimensional limit and that prior-aligned anisotropic references are needed for mesh-stable transport [2603.14798]. This suggests that one-step transport is unusually sensitive to the geometry of its reference law.

A third limitation is that **computational savings at inference may shift difficulty into training**. Exact or approximate OT couplings can be expensive [2604.06413], [2509.22592]. Two-Step Diffusion makes this explicit by postponing explicit \(W_2\) optimization until after a Meanflow initializer has simplified the geometry [2601.20024]. The broader implication is that one-step transport often relies on a separation between an easy-to-sample learned map and a harder offline procedure that shapes its geometry.

A fourth issue is **diagnostics**. Endpoint quality can hide path inconsistency, teacher dependence, or compression artifacts. The SCGP framework formalizes this by bounding one-step generation quality with a residual term \(\widehat R_{\rm one}\), teacher quality, distillation error, and proxy error, thereby reframing one-step evaluation as terminal path self-consistency testing rather than endpoint matching alone [2606.08953].

Finally, recent theory suggests a more favorable mathematical picture than early pessimism implied. For PDE-induced target measures, “On the Regularity and Generalization of One-Step Wasserstein-guided Generative Models for PDE-Induced Measures” proves doubling conditions and Hölder continuity of the optimal transport map from a uniform source to the target, and derives excess-risk and robustness estimates for one-step Wasserstein-guided learners such as DeepParticle [2605.21388]. A plausible implication is that one-step generative transport is most principled when the population transport map itself is regular, the source is geometrically compatible, and the learned correction is tested against an admissible refinement.

In that sense, the contemporary literature no longer treats one-step generative transport as merely an acceleration trick. It is increasingly framed as a problem of choosing the correct transport object, the correct source law, and the correct notion of self-consistency so that a single finite-time map can stand in for an otherwise long generative path.

Source: https://www.emergentmind.com/topics/one-step-generative-transport