FlashVSR+: Efficient Video Super-Resolution via Hybrid Attention and Reward-Yield Distribution Matching

Hugging Face Daily Papers Papers
video-super-resolution diffusion-distillation hybrid-attention preference-alignment efficiency distribution-matching

Summary

FlashVSR+ 提出基于混合注意力的高效视频超分辨率架构与 Reward-Yield 分布匹配蒸馏（Ry-DMD），实现 2 倍 DiT 加速、45% 峰值显存降低，可在单张 24GB RTX 4090 上完成 4K 视频超分，并支持黑盒奖励对齐的单步生成。

Recent video super-resolution (VSR) models achieve efficient one-step generation through diffusion distillation. However, high-resolution inference remains costly, and distilled models are difficult to align with human preferences. These challenges call for improvements in both model architecture and training strategy. We present FlashVSR+ to address both limitations. On the architecture side, we introduce hybrid attention based on the strong locality observed in pretrained VSR models, while retaining sparse global interaction for long-range consistency. It achieves up to 2times DiT speedup and 45\% lower peak memory, enabling spatially untiled 4K VSR on a single 24GB RTX 4090. On the training side, we propose Reward-Yield Distribution Matching Distillation (Ry-DMD) for preference-aligned one-step generation. Ry-DMD uses reward-weighted teacher rollouts to incorporate black-box rewards into distribution matching, without backpropagating through the reward model. We further show that Ry-DMD generalizes beyond VSR to T2I generation. Together, FlashVSR+ delivers stronger restoration quality and substantially improves high-resolution efficiency, with consistent gains in user preference. We release our code, models, and training data at https://www.flashvsr.tech{FlashBlue{www.flashvsr.tech}}.
Original Article
View Cached Full Text

Cached at: 10/14/26, 10:51 AM

Paper page - FlashVSR+: Efficient Video Super-Resolution via Hybrid Attention and Reward-Yield Distribution Matching

Source: https://huggingface.co/papers/2610.17552

Abstract

Recentvideosuper-resolution(VSR)modelsachieveefficientone-stepgenerationthroughdiffusiondistillation.However,high-resolutioninferenceremainscostly,anddistilledmodelsaredifficulttoalignwithhumanpreferences.Thesechallengescallforimprovementsinbothmodelarchitectureandtrainingstrategy.WepresentFlashVSR+toaddressbothlimitations.Onthearchitectureside,weintroducehybridattentionbasedonthestronglocalityobservedinpretrainedVSRmodels,whileretainingsparseglobalinteractionforlong-rangeconsistency.Itachievesupto2timesDiTspeedupand45\%lowerpeakmemory,enablingspatiallyuntiled4KVSRonasingle24GBRTX4090.Onthetrainingside,weproposeReward-YieldDistributionMatchingDistillation(Ry-DMD)forpreference-alignedone-stepgeneration.Ry-DMDusesreward-weightedteacherrolloutstoincorporateblack-boxrewardsintodistributionmatching,withoutbackpropagatingthroughtherewardmodel.WefurthershowthatRy-DMDgeneralizesbeyondVSRtoT2Igeneration.Together,FlashVSR+deliversstrongerrestorationqualityandsubstantiallyimproveshigh-resolutionefficiency,withconsistentgainsinuserpreference.Wereleaseourcode,models,andtrainingdataathttps://www.flashvsr.tech{FlashBlue{www.flashvsr.tech}}.

View arXiv pageView PDFProject pageGitHub11Add to collection

Get this paper in your agent:

hf papers read 2610\.17552

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper1

#### ruchangyao/FlashVSR_plus Video-to-Video• Updatedabout 3 hours ago • 26 • 3

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2610.17552 in a dataset README.md to link it from this page.

Spaces citing this paper1

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

FVAttn: Adaptive Sparse Attention with Runtime Load Balancing for Video Generation

Hugging Face Daily Papers

FVAttn is a training-free sparse attention system that uses runtime load balancing to improve distributed execution efficiency of adaptive sparse attention under multi-GPU sequence parallelism for video generation, achieving up to 4.41x attention speedup and 2.11x inference speedup over FlashAttention on step-distilled Wan2.2 I2V while maintaining competitive video quality.

Velox: Spatial Memory Makes World Models Efficient

Hugging Face Daily Papers

Velox 提出一种免训练的注意力路由器，利用相机视锥切片的几何对应关系进行区域级上下文选择，从而加速长视频生成中的 KV 缓存处理。在第 48 个 chunk 时注意力加速 4.36 倍、去噪加速 2.43 倍，且在 WBench 导航基准上几乎不损失质量。