Pith. sign in

REVIEW 5 major objections 5 minor 38 references

Improving Rectified Flow with Boundary Conditions

T0 review · 5 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper shows that forcing a Rectified Flow velocity to satisfy v(x,1)=x improves generation quality and stabilizes stochastic sampling.

desk verdict A simple, plausible fix for a real problem in rectified flow: enforce the terminal velocity v(x,1)=x via a subtraction parameterization; the reported gains are suggestive but rest on single-seed numbers and a suspicious mode-sampling baseline. read the letter →

arxiv 2506.15864 v2 pith:PN2PGUNG submitted 2025-06-18 cs.LG

Xixi Hu , Runlong Liao , Keyang Xu , Bo Liu , Yeqing Li , Eugene Ie , Hongliang Fei , Qiang Liu This is my paper · ORCID
classification cs.LG
keywords rectifiedflowmatchingboundaryconditionsvelocityfieldstochasticsamplingscorefunctionimagegenerationgenerativemodeling
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Rectified Flow turns sampling into solving an ODE whose velocity field should move noise to data. The paper identifies a hidden failure mode: when the velocity is learned by an unconstrained network, it does not satisfy the endpoint identity v(x,1)=x that the interpolation implies, and this makes the score function used in stochastic samplers blow up near the final time. It proposes two nearly-bulletproof parameterizations that enforce this boundary condition by construction, and reports that on ImageNet 256x256 the better variant drops FID from 6.87 to 6.32 under Euler sampling, with similar gains under SDE sampling. The result matters because it removes a systematic error at almost no implementation cost and makes stochastic sampling stable enough to use larger noise scales.

What carries the argument

The central object is the terminal velocity constraint v(x,1)=x. For the linear interpolation X_t=(1-t)X0+tX1, the conditional expectation v*(x,t)=E[X1-X0|Xt=x] simplifies at t=1 to x-E[X0]=x when X0 has mean zero. The mask-based model writes v(x,t)=g(t)(C-x)+f(t)x+h(t)m_theta(x,t) with g(1)=f(0)=h(0)=h(1)=0 and g(0)=f(1)=1, using trigonometric g, f, h; the subtraction-based model sets v(x,t)=x+m_theta(x,t)-m_theta(x,1), which forces v(x,1)=x exactly regardless of the network. This identity prevents the score estimate nabla log rho_t(x) = (t v(x,t)-x)/(1-t) from diverging near t=1, which is the mechanism the paper credits for stable stochastic sampling.

What would settle it

Compute the per-channel mean of the VAE latent codes on the ImageNet training set. If any channel mean is substantially nonzero, retrain the subtraction-based Boundary RF Model with the latent mean explicitly subtracted from X1, and compare FID against vanilla RF; the claimed improvement should persist only if the boundary constraint matches the data's actual mean.

Watch

Extended reading notes

Core claim

In Rectified Flow, the ideal velocity field satisfies v*(x,0)=E[X1]-x and v*(x,1)=x when the noise X0 is zero-mean and independent of the data X1. A neural network trained with the standard flow-matching loss does not obey these constraints at the boundaries, especially at t=1. The paper constructs two parameterizations—a mask-based model and a subtraction-based model—that enforce the right boundary by design, and reports consistent FID gains: on ImageNet 256x256, the subtraction-based model reaches FID 6.32 vs 6.87 for vanilla RF under Euler sampling, with analogous improvements under an SDE sampler. The enforced boundary also keeps the Tweedie score estimate finite as t approaches 1, which the authors argue is why stochastic samplers produce sharper, less over-smoothed images.

Load-bearing premise

The enforced right boundary v(x,1)=x is correct only if the noise X0 is zero-mean and independent of X1, and if the data representation is centered; if the latents have a nonzero mean, the constraint is systematically wrong.

Editorial extensions

If this is right

  • Any Rectified Flow model can be retrofitted with the subtraction parameterization, so a drop-in FID gain should appear on other datasets and architectures.
  • Because the Tweedie score stays bounded as t approaches 1, stochastic samplers can be run with larger noise scales without over-smoothing, which should translate to fewer required steps at equal quality.
  • The boundary violation is measurable: computing the mean of ||v(x,1)-x|| on validation data gives a diagnostic for convergence of any trained RF model.
  • The same enforced boundary should transfer to other flow-matching formulations built on linear or affine interpolation schedules.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial extension: The enforced t=1 constraint presumes a centered latent space; if the VAE latents used for ImageNet have nonzero per-channel means, the subtraction model could be systematically biased. A quick check is to subtract the empirical latent mean and see whether the FID gain changes.
  • Editorial extension: The same boundary analysis could be run for the t=0 endpoint under non-Gaussian or correlated priors, where v*(x,0)=E[X1]-x would no longer hold; a testable extension is whether a learned prior mean restores the benefit.
  • Editorial extension: The paper's own visualization shows a production text-to-image model also violates the boundary; this suggests the fix may matter more at scale, where endpoint errors accumulate over many steps, but the paper only demonstrates it on class-conditional ImageNet and CIFAR-10.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The paper identifies that neural-network parameterizations of Rectified Flow velocity fields do not automatically satisfy the theoretical boundary conditions v(x,1)=x and v(x,0)=C-x, and argues that this violation degrades deterministic and especially stochastic sampling. It proposes two reparameterizations: a mask-based model that explicitly enforces both boundaries through scalar schedules f,g,h, and a subtraction-based model of the form v(x,t)=x+m(x,t)-m(x,1) that enforces the right boundary by construction. Experiments on CIFAR-10, ImageNet 256x256, ImageNet 512x512, and DiT-B/L/XL configurations report consistent FID improvements over a vanilla RF baseline, with the headline result being ImageNet 256x256 FID of 6.32 for Subtraction-based Boundary RF versus 6.87 for vanilla RF under Euler sampling.

Significance. The mathematical derivation of the boundary conditions from the conditional-expectation definition of the optimal velocity is correct and is presented cleanly in Section 3.1. The proposed modification is minimal, architecture-agnostic, and does not introduce fitted constants to match the reported FID numbers; the subtraction-based variant is especially simple. If the reported gains are reproducible, the method could be a useful drop-in improvement for training rectified flows and flow matching, and the stabilization of stochastic sampling is a plausible and interesting mechanism. However, the empirical foundation is currently thin: the headline difference is a single-run comparison without error bars, the mode-sampling baseline is both misdescribed and anomalously weak, and the subtraction variant is confounded with an extra forward pass. These issues must be resolved before the central claim can be accepted.

major comments (5)
  1. [Section 5.2, Table 1] The central empirical claim rests on single-run FID comparisons with no error bars, multiple seeds, or bootstrap intervals. ImageNet-256 DiT-scale FID is known to vary by several tenths across seeds, so the reported 6.32 versus 6.87 gap may be within run-to-run noise. The paper should provide at least three seeds (or a statement of variance) for the main comparisons, or otherwise justify that the difference is statistically meaningful.
  2. [Section 5.2, Table 1 (RF with Mode Sampling)] The description of mode sampling is factually wrong and the reported result is anomalous. The paper states that mode sampling concentrates probability 'towards the boundaries of the timestep range (i.e., near 0 and 1)', whereas the cited SD3 method [7] uses logit-normal timestep sampling that concentrates near a central mode around t=0.5. Moreover, the reported RF-with-Mode-Sampling FID (7.48) is far worse than the vanilla RF baseline (6.87), which is not expected for a properly tuned training schedule. This suggests either a misimplementation of the baseline or an under-optimized vanilla setting, and it undermines the claim that boundary enforcement is the cause of the improvement.
  3. [Section 3.4, Eq. (5)] The claim that enforcing v(x,1)=x 'prevents the score function from diverging' is not rigorously justified. Tweedie's formula gives ∇log ρ_t(x) = (t v(x,t) − x)/(1 − t); at t=1 the numerator and denominator both vanish when v(x,1)=x, leaving a 0/0 limit whose finiteness depends on the rate at which v(x,t) approaches x, not merely on the boundary value. For example, v(x,t)=x+√(1−t) satisfies the boundary condition but still yields a divergent score. The stability argument needs either a quantitative bound on the score or an empirical measurement of score norms near t=1.
  4. [Section 3.2 and Section 5.3] The mask-based model uses the constant C=E[X1] in its left boundary condition v(x,0)=C−x, but the paper never states how C is obtained in practice: whether it is the exact population mean, a training-set estimate for the ImageNet VAE latents, a running batch estimate, or simply set to zero. This omission prevents reproduction of the mask-based results and matters because VAE latents need not be centered; an incorrect C would systematically bias the left boundary and could affect the reported mask-based FID numbers.
  5. [Section 5.3, Table 2] The subtraction-based model differs from the mask-based model not only in enforcing the right boundary but also in being a two-pass residual network with an additional forward pass at t=1. The ablation table labels the subtraction row '2-Pass Model' but does not include a two-pass model without boundary enforcement (e.g., v(x,t)=x+m(x,t) evaluated with two passes or analogous capacity increase). Consequently, the additional improvement of the subtraction variant over the mask variant (6.32 vs 6.63) cannot be attributed solely to boundary enforcement; it may reflect the extra computation or the residual parameterization rather than the boundary condition.
minor comments (5)
  1. [Abstract and Section 5.2] Reporting FID differences as percentages ('8.01% improvement') is nonstandard and can be misleading; absolute FID changes with uncertainty would be more informative.
  2. [Section 5.1, Appendix A.1] The text says 'we maintain consistent random seeds' but does not specify how many seeds are used or whether the reported results are from a single run. Please clarify whether 'consistent seeds' means one seed per method or multiple seeds averaged.
  3. [Section 5.2, text after Table 1] The sentence 'across both datasets and evaluation metrics' appears twice in close succession; remove the duplicate.
  4. [Section 3.3 and Section 5.2] The subtraction-based variant doubles the inference cost due to the extra m(x,1) forward pass, yet the main tables report no runtime or throughput comparison. Since the paper emphasizes practical ease, a wall-clock comparison would help readers weigh the trade-off.
  5. [Appendix A.1, Table 5] For CIFAR-10 the model is listed as DDPM++ while the main text says all models are DiT-based; please clarify which architecture is used for the CIFAR-10 experiments.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: boundary conditions are derived from the conditional-expectation definition of the RF velocity and are enforced by construction, with no fitted constants or self-citation chain carrying the central claim.

full rationale

The paper's derivation chain is self-contained rather than circular. The claimed boundary conditions come directly from the definition of the optimal velocity as a conditional expectation: v*(x,t)=E[X1-X0 | Xt=x] under straight-line interpolation. The right boundary v*(x,1)=x follows from E[X0]=0 in the standard setting X0~N(0,I), as stated in Section 3.1, and the left boundary v*(x,0)=E[X1]-x follows from independence of X0 and X1. The mask-based parameterization Eq. (7) imposes these identities by construction through the coefficient functions g(0)=f(1)=1 and g(1)=f(0)=h(0)=h(1)=0; the subtraction-based parameterization Eq. (10) enforces v(x,1)=x by the algebraic identity m_theta(x,1)-m_theta(x,1)=0. These are not fitted inputs renamed as predictions: the neural network m_theta is still trained with the same MSE objective Eq. (9) against the interpolation slope, and no constant is tuned to the reported FID numbers. The improvement in FID is an external empirical result, not a logical consequence of the construction alone. The stabilization claim for stochastic sampling also follows algebraically from Tweedie's formula Eq. (5), where the score has denominator (1-t); enforcing v(x,1)=x makes the numerator vanish at t=1. This is a derivation, not an assumed conclusion. Self-citations appear but are not load-bearing: [14] is cited for the Langevin/overshooting sampler and [24] for the stochastic curved Euler sampler, both external sampling tools whose use does not imply the boundary-condition result. No uniqueness theorem is imported from the authors' prior work, and no ansatz is smuggled in via self-citation. The reader's centering concern about E[X1] is a potential correctness issue for the mask-based left boundary, not a circularity in the derivation of the right boundary. Overall, the central contribution is an architectural constraint that provably satisfies derived identities, and the empirical claims are benchmarked against a vanilla baseline, so the circularity burden is essentially absent.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claim rests on standard Rectified Flow assumptions (Gaussian independent noise, linear interpolation) plus one unstated practical assumption about data centering. The boundary functions are manual design choices, and the mean constant C is left unspecified. No new physical or mathematical entities are introduced.

free parameters (2)
  • Boundary schedule functions f(t), g(t), h(t) = cos(pi*t/2), sin(pi*t/2), sin(pi*t) in the default setting
    Chosen by hand to satisfy the endpoint constraints; the ablation in Section 5.3 shows that performance depends on this choice, making it a tunable design parameter rather than a derived quantity.
  • Mean data constant C = E[X1] = not specified; implicitly treated as zero in the subtraction model
    The mask-based model in Eq. (7) requires C for the left boundary v(x,0)=C-x, but the paper never states how C is estimated or whether preprocessing centers the data. The subtraction model dodges this by only enforcing the right boundary.
assumptions (4)
  • domain assumption Noise X0 is standard Gaussian with zero mean and is independent of the data X1.
    Used in Section 3.1 to derive v*(x,1)=x because E[X0]=0. This is the standard Rectified Flow setting, but it must hold exactly for the boundary constraint to be correct.
  • domain assumption The interpolation path is the straight line Xt=(1-t)X0+tX1.
    This defines the slope X1-X0 and the loss in Eq. (3). The boundary conditions follow from this specific interpolation, not from a general flow.
  • domain assumption Tweedie's formula in Eq. (5) gives the correct score function from the learned velocity.
    The paper's argument that boundary enforcement stabilizes stochastic sampling relies on the score being computed via Eq. (5), which assumes the conditional structure of the linear Gaussian interpolation.
  • ad hoc to paper The data or latent representation is centered so that the right boundary is exactly x and the mask model can set C without error.
    The paper never verifies this centering assumption for ImageNet or CIFAR-10 latents. If the data mean is nonzero, the mask model needs a separately estimated C and the subtraction model enforces a slightly wrong boundary.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Improving Rectified Flow with Boundary Conditions." pith.science (2026). https://pith.science/paper/PN2PGUNG

@misc{pith2026250615864,
  author       = {Pith},
  title        = {Pith review of: Improving Rectified Flow with Boundary Conditions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PN2PGUNG}},
  note         = {Machine review of arXiv:2506.15864}
}
read the original abstract

Rectified Flow offers a simple and effective approach to high-quality generative modeling by learning a velocity field. However, we identify a limitation in directly modeling the velocity with an unconstrained neural network: the learned velocity often fails to satisfy certain boundary conditions, leading to inaccurate velocity field estimations that deviate from the desired ODE. This issue is particularly critical during stochastic sampling at inference, as the score function's errors are amplified near the boundary. To mitigate this, we propose a Boundary-enforced Rectified Flow Model (Boundary RF Model), in which we enforce boundary conditions with a minimal code modification. Boundary RF Model improves performance over vanilla RF model, demonstrating 8.01% improvement in FID score on ImageNet using ODE sampling and 8.98% improvement using SDE sampling.

Figures

Figures reproduced from arXiv: 2506.15864 by the authors.

Figure 1
Figure 1. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Toy example: Boundary RF Model stabilizes stochas￾tic sampling and score function. We visualize the behavior of vanilla RF model and Boundary RF Model when learning to map from noise π0 to data π1. From left to right, we visualize the fol￾lowings: 1) Euler Sampling (Deterministic): Both models learn effective ODE trajectories, generating similar samples via deter￾ministic Euler sampler. 2) Stochastic Sampling: vanil… view at source ↗
Figure 3
Figure 3. Performance comparison of Boundary RF Model and [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Qualitative comparison of image generation results on ImageNet 256 × 256 dataset. We present paired examples generated by vanilla RF model, Mask-based Boundary RF Model and Subtraction-based Boundary RF Model. We use the same random seed during training and evaluation …
Figure 5
Figure 5. Figure 5: Ablation study on boundary functions. Quantitative [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Boundary Condition Violation in FLUX.1-dev Model. Visualization of the predicted velocity field at t=1 , demonstrating deviation from the expected data distribution and violation of the right boundary condition v(\mathbf {x}, t) = \mathbf {x} . Vanilla RF Model Mask-ba…
Figure 7
Figure 7. Figure 7: Qualitative comparison of image generation results on ImageNet 256×256 dataset. We present paired examples generated by vanilla RF model (first row), Mask-based Boundary RF Model (second column) and Subtraction-based Boundary RF Model (last column). We use the same ran…
Figure 8
Figure 8. Figure 8: Qualitative comparison of boundary functions. Generated samples from double-boundary models using different choices of [PITH_FULL_IMAGE:figures/full_fig_p013_8.png]
Figure 9
Figure 9. Figure 9: Qualitative comparison of stochastic sampling with Overshooting sampler. Qualitative samples using the overshooting sampler reveal that Boundary RF Model effectively preserves high-frequency details, contrasting with the over-smoothed results from vanilla RF model unde…
Figure 10
Figure 10. Figure 10: Qualitative comparison of image generation results on ImageNet 256 × 256 dataset with DiT-L/2 and DiT-XL/2 models. We present paired examples generated by vanilla RF model (first row), Mask-based Boundary RF Model (second column) and Subtraction￾based Boundary RF Mode…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 15 canonical work pages

  1. [7]

    Scaling recti- fied flow transformers for high-resolution image synthesis

    Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas M ¨uller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, et al. Scaling recti- fied flow transformers for high-resolution image synthesis. InForty-first international conference on machine learning,

  2. [1]

    Building nor- malizing flows with stochastic interpolants.arXiv preprint arXiv:2209.15571, 2022

    Michael S Albergo and Eric Vanden-Eijnden. Building nor- malizing flows with stochastic interpolants.arXiv preprint arXiv:2209.15571, 2022. 1, 2, 4

    arXiv 2022

  3. [2]

    Imagen 3

    Jason Baldridge, Jakob Bauer, Mukul Bhutani, Nicole Brich- tova, Andrew Bunner, Kelvin Chan, Yichang Chen, Sander Dieleman, Yuqing Du, Zach Eaton-Rosen, et al. Imagen 3. arXiv preprint arXiv:2408.07009, 2024. 4

    arXiv 2024

  4. [3]

    Improving image generation with better captions.Computer Science

    James Betker, Gabriel Goh, Li Jing, Tim Brooks, Jianfeng Wang, Linjie Li, Long Ouyang, Juntang Zhuang, Joyce Lee, Yufei Guo, et al. Improving image generation with better captions.Computer Science. https://cdn. openai. com/papers/dall-e-3. pdf, 2(3):8, 2023. 4

    work page 2023

  5. [4]

    Jax: composable transformations of python+ numpy programs

    James Bradbury, Roy Frostig, Peter Hawkins, Matthew James Johnson, Chris Leary, Dougal Maclau- rin, George Necula, Adam Paszke, Jake VanderPlas, Skye Wanderman-Milne, et al. Jax: composable transformations of python+ numpy programs. 2018. 5

    work page 2018

  6. [5]

    Imagenet: A large-scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In2009 IEEE conference on computer vision and pattern recognition, pages 248–255. Ieee, 2009. 5

    2009

  7. [6]

    Diffusion models beat gans on image synthesis.Advances in neural informa- tion processing systems, 34:8780–8794, 2021

    Prafulla Dhariwal and Alexander Nichol. Diffusion models beat gans on image synthesis.Advances in neural informa- tion processing systems, 34:8780–8794, 2021. 4

    2021

  8. [8]

    Flux that plays music.arXiv preprint arXiv:2409.00587, 2024

    Zhengcong Fei, Mingyuan Fan, Changqian Yu, and Jun- shi Huang. Flux that plays music.arXiv preprint arXiv:2409.00587, 2024. 4

    arXiv 2024

Show all 38 references
  1. [9]

    Murphy, and Tim Salimans

    Ruiqi Gao, Emiel Hoogeboom, Jonathan Heek, Valentin De Bortoli, Kevin P. Murphy, and Tim Salimans. Diffusion meets flow matching: Two sides of the same coin. 2024. 2, 4, 5

    2024

  2. [10]

    Iter- ativeα-(de) blending: A minimalist deterministic diffusion model

    Eric Heitz, Laurent Belcour, and Thomas Chambon. Iter- ativeα-(de) blending: A minimalist deterministic diffusion model. InACM SIGGRAPH 2023 Conference Proceedings, pages 1–8, 2023. 1, 2, 4

    2023

  3. [11]

    Gans trained by a two time-scale update rule converge to a local nash equilib- rium.Advances in neural information processing systems, 30, 2017

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilib- rium.Advances in neural information processing systems, 30, 2017. 5

    2017

  4. [12]

    Denoising dif- fusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising dif- fusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020. 1, 2, 4

    2020

  5. [13]

    Video dif- fusion models.Advances in Neural Information Processing Systems, 35:8633–8646, 2022

    Jonathan Ho, Tim Salimans, Alexey Gritsenko, William Chan, Mohammad Norouzi, and David J Fleet. Video dif- fusion models.Advances in Neural Information Processing Systems, 35:8633–8646, 2022. 4

    2022

  6. [14]

    Amo sampler: Enhancing text rendering with overshooting

    Xixi Hu, Keyang Xu, Bo Liu, Qiang Liu, and Hongliang Fei. Amo sampler: Enhancing text rendering with overshooting. arXiv preprint arXiv:2411.19415, 2024. 2, 3, 4, 5, 6, 8, 11

    2024 arXiv

  7. [15]

    Elucidating the design space of diffusion-based generative models.Advances in neural information processing systems, 35:26565–26577, 2022

    Tero Karras, Miika Aittala, Timo Aila, and Samuli Laine. Elucidating the design space of diffusion-based generative models.Advances in neural information processing systems, 35:26565–26577, 2022. 5

    2022

  8. [16]

    Understanding diffu- sion objectives as the elbo with simple data augmentation

    Diederik Kingma and Ruiqi Gao. Understanding diffu- sion objectives as the elbo with simple data augmentation. Advances in Neural Information Processing Systems, 36: 65484–65516, 2023. 4, 5

    2023

  9. [17]

    Learning multiple layers of features from tiny images

    Alex Krizhevsky, Geoffrey Hinton, et al. Learning multiple layers of features from tiny images. 2009. 5

    2009

  10. [18]

    Flux.https://github.com/ black-forest-labs/flux, 2023

    Black Forest Labs. Flux.https://github.com/ black-forest-labs/flux, 2023. GitHub repository. 1, 11

    2023

  11. [19]

    V oicebox: Text-guided multilingual universal speech generation at scale.Advances in neural information processing systems, 36, 2024

    Matthew Le, Apoorv Vyas, Bowen Shi, Brian Karrer, Leda Sari, Rashel Moritz, Mary Williamson, Vimal Manohar, Yossi Adi, Jay Mahadeokar, et al. V oicebox: Text-guided multilingual universal speech generation at scale.Advances in neural information processing systems, 36, 2024. 4

    2024

  12. [20]

    Improving the training of rectified flows.Advances in neural information processing systems, 37:63082–63109, 2024

    Sangyun Lee, Zinan Lin, and Giulia Fanti. Improving the training of rectified flows.Advances in neural information processing systems, 37:63082–63109, 2024. 5

    2024

  13. [21]

    Flow matching for generative mod- eling.arXiv preprint arXiv:2210.02747, 2022

    Yaron Lipman, Ricky TQ Chen, Heli Ben-Hamu, Maximil- ian Nickel, and Matt Le. Flow matching for generative mod- eling.arXiv preprint arXiv:2210.02747, 2022. 1, 2, 4

    2022 arXiv

  14. [22]

    Flow matching guide and code.arXiv preprint arXiv:2412.06264, 2024

    Yaron Lipman, Marton Havasi, Peter Holderrieth, Neta Shaul, Matt Le, Brian Karrer, Ricky TQ Chen, David Lopez- Paz, Heli Ben-Hamu, and Itai Gat. Flow matching guide and code.arXiv preprint arXiv:2412.06264, 2024. 4, 5

    2024 arXiv

  15. [23]

    Rectified flow: A marginal preserving approach to optimal transport.arXiv preprint arXiv:2209.14577, 2022

    Qiang Liu. Rectified flow: A marginal preserving approach to optimal transport.arXiv preprint arXiv:2209.14577, 2022. 1, 2

    2022 arXiv

  16. [24]

    Qiang Liu.Let us Flow Together. Dr. Qiang Liu’s Website, 2025.https://www.cs.utexas.edu/ ˜lqiang/ PDF/flow_book.pdf. 2, 3, 4, 5, 11

    2025

  17. [25]

    Flow straight and fast: Learning to generate and transfer data with rectified flow

    Xingchao Liu, Chengyue Gong, and Qiang Liu. Flow straight and fast: Learning to generate and transfer data with rectified flow. InThe Eleventh International Conference on Learning Representations, 2022. 1, 2, 4

    2022

  18. [26]

    Dpm-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps.Advances in Neural Information Processing Systems, 35:5775–5787,

    Cheng Lu, Yuhao Zhou, Fan Bao, Jianfei Chen, Chongxuan Li, and Jun Zhu. Dpm-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps.Advances in Neural Information Processing Systems, 35:5775–5787,

  19. [27]

    On distillation of guided diffusion models

    Chenlin Meng, Robin Rombach, Ruiqi Gao, Diederik Kingma, Stefano Ermon, Jonathan Ho, and Tim Salimans. On distillation of guided diffusion models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, pages 14297–14306, 2023. 5

    2023

  20. [28]

    Improved denoising diffusion probabilistic models

    Alexander Quinn Nichol and Prafulla Dhariwal. Improved denoising diffusion probabilistic models. InInternational conference on machine learning, pages 8162–8171. PMLR,

  21. [29]

    Scalable diffusion models with transformers

    William Peebles and Saining Xie. Scalable diffusion models with transformers. InProceedings of the IEEE/CVF Inter- national Conference on Computer Vision, pages 4195–4205,

  22. [30]

    Sdxl: Improving latent diffusion mod- els for high-resolution image synthesis.arXiv preprint arXiv:2307.01952, 2023

    Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas M ¨uller, Joe Penna, and Robin Rombach. Sdxl: Improving latent diffusion mod- els for high-resolution image synthesis.arXiv preprint arXiv:2307.01952, 2023. 4

    2023 arXiv

  23. [31]

    Movie gen: A cast of media foundation models.arXiv preprint arXiv:2410.13720,

    Adam Polyak, Amit Zohar, Andrew Brown, Andros Tjandra, Animesh Sinha, Ann Lee, Apoorv Vyas, Bowen Shi, Chih- Yao Ma, Ching-Yao Chuang, et al. Movie gen: A cast of media foundation models.arXiv preprint arXiv:2410.13720,

    arXiv

  24. [32]

    Photorealistic text-to-image diffusion models with deep language understanding.Advances in neural information processing systems, 35:36479–36494, 2022

    Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily L Denton, Kamyar Ghasemipour, Raphael Gontijo Lopes, Burcu Karagol Ayan, Tim Salimans, et al. Photorealistic text-to-image diffusion models with deep language understanding.Advances in neural information ...

    2022

  25. [33]

    Assessing generative models via precision and recall.Advances in neural information pro- cessing systems, 31, 2018

    Mehdi SM Sajjadi, Olivier Bachem, Mario Lucic, Olivier Bousquet, and Sylvain Gelly. Assessing generative models via precision and recall.Advances in neural information pro- cessing systems, 31, 2018. 5

    2018

  26. [34]

    Improved techniques for training gans.Advances in neural information processing systems, 29, 2016

    Tim Salimans, Ian Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, and Xi Chen. Improved techniques for training gans.Advances in neural information processing systems, 29, 2016. 5

    2016

  27. [35]

    Bespoke solvers for genera- tive flow models.arXiv preprint arXiv:2310.19075, 2023

    Neta Shaul, Juan Perez, Ricky TQ Chen, Ali Thabet, Albert Pumarola, and Yaron Lipman. Bespoke solvers for genera- tive flow models.arXiv preprint arXiv:2310.19075, 2023. 2

    2023 arXiv

  28. [36]

    Denoising diffusion implicit models.arXiv preprint arXiv:2010.02502, 2020

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models.arXiv preprint arXiv:2010.02502, 2020. 5

    2010 arXiv

  29. [37]

    Generative modeling by esti- mating gradients of the data distribution.Advances in neural information processing systems, 32, 2019

    Yang Song and Stefano Ermon. Generative modeling by esti- mating gradients of the data distribution.Advances in neural information processing systems, 32, 2019. 1, 2, 4

    2019

  30. [38]

    Score-based generative modeling through stochastic differential equa- tions.arXiv preprint arXiv:2011.13456, 2020

    Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Ab- hishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equa- tions.arXiv preprint arXiv:2011.13456, 2020. 1, 2, 3, 4, 5 10 A. Appendix A.1. Experiment Details Model Co...

    2011 arXiv

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.