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 2× 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 \href{https://www.flashvsr.tech}{\textcolor{FlashBlue}{**www.flashvsr.tech**}}.
Figures & tables
Figure 1: FlashVSR+ achieves high-quality and efficient video restoration. Left: We recover fine visual details while maintaining natural textures. Right: At 4K resolution, we suppress visible artifacts and enable substantially faster inference on both H200 equivalent GPU and a RTX 4090.
Figure 2: Overview of the FlashVSR+ training pipeline. Stage 1 trains a 14B full-attention multi-step teacher on image and video data. Stage 2 trains the hybrid-attention student to obtain a strong initialization. Stage 3 distills the student into a one-step generator using Ry-DMD, where reward feedback modifies the teacher term estimation without requiring reward gradients.
Figure 3: Attention locality and adaptation in causal VSR. (a) Attention concentrates in compact spatiotemporal neighborhoods. (b) Local attention mass varies across DiT blocks, and hybrid training reduces the locality of retained global-attention blocks.
Figure 4: Conceptual illustration of Ry-DMD.
Method
Fidelity
Quality
Overall
RealViFormer
−70%
−72%
−73%
DOVE
−68%
−68%
−70%
DLoRAL
−38%
−28%
−35%
SeedVR2
−53%
−55%
−56%
PS-SR
−38%
−38%
−41%
FlashVSR
−15%
−10%
−15%
Table 1: Human preference study. Preference margins relative to Ours-Base.
Figure 5: Qualitative comparison with representative VSR methods. Zoomed regions highlight the reconstruction of text, fine textures, and facial details.
Dataset
Metric
RealViFormer
DOVE
SeedVR2-3B
PS-SR
FlashVSR
Ours-Base
Ours-Turbo
YouHQ40
PSNR ↑
23.50
20.91
23.34
22.35
22.53
23.64
23.51
SSIM ↑
0.6051
0.5532
0.6206
0.5899
0.5787
0.6250
0.6097
LPIPS ↓
0.3528
0.3514
0.2768
0.3122
0.2721
0.2698
0.2728
MUSIQ ↑
60.36
60.82
57.17
66.20
65.91
64.71
63.91
CLIPIQA ↑
0.3823
0.3964
0.3703
0.5063
0.4701
0.4257
0.4322
NIQE ↓
3.490
4.581
4.213
3.72
3.385
3.560
3.737
Table 2: Quality comparison with state-of-the-art VSR methods under varying inference costs. Top-1/2/3 results are highlighted in red/orange/yellow.
Figure 6: Efficiency comparison on 89-frame videos. (a) GPU time and peak memory on NVIDIA H200 equivalent GPU. (b) End-to-end wall-clock inference time on NVIDIA RTX 4090, including CPU KV-cache offloading and asynchronous prefetching overhead.
Figure 7: Quality and efficiency of trained hybrid attention. The (3,1,25) configuration achieves the best balance between DiT cost and restoration quality.
Figure 8: Qualitative comparison of OCR-guided VSR post-training. Ry-DMD restores more recognizable text, while DMDR often introduces artifacts.
Method
OCR ↑
PSNR ↑
LPIPS ↓
MUSIQ ↑
DOVER ↑
Baseline
0.624
23.94
0.246
75.35
9.79
TextSFT
0.656
22.02
0.295
76.41
10.58
DMDR
0.527
19.01
0.374
74.17
8.48
Ry-DMD
0.776
24.86
0.240
75.80
10.90
Table 3: Quantitative results of OCR-guided post-training.
Model
NFE ↓
OCR ↑
PickScore ↑
HPSv2 ↑
ImageReward ↑
Aesthetic ↑
SDXL
50
0.125
22.2
27.5
0.87
5.48
DMD2
4
0.098
22.5
29.2
0.83
5.47
One-step DMD2
1
0.087
22.0
26.8
0.67
5.39
Ry-DMD (ours)
1
0.241
22.7
29.6
1.02
5.44
Table 4: Reward-guided text-to-image generation on SDXL.
Figure 9: Qualitative comparison of Ry-DMD on 4-step SD3.5-M. Ry-DMD substantially improves text correctness and legibility while preserving the intended scene content. Beyond text rendering, the resulting images often exhibit cleaner structures, more coherent materials and lighting, and improved overall visual quality.
Figure 10: Ablation on DMD steps.
Init.
Teacher
MUSIQ ↑
DOVER ↑
Base
1.3B
45.87
8.512
Strong
1.3B
48.20
8.738
Base
14B
51.76
9.144
Strong
14B
52.78
9.417
Table 5: Ablation on student initialization and teacher capacity.
Appendix figures & tables2 assets
Supplementary material from the paper’s appendix.
Appendix
Setting
β
λRY
J
K
OCR ↑
ImageReward ↑
Default
0.1
0.2
4
3
0.241
1.02
β
0.05
–
–
–
0.212
0.98
0.2
–
–
–
0.236
1.01
0.4
–
–
–
0.203
0.92
λRY
–
0.05
–
–
0.201
0.98
–
0.1
–
–
0.237
1.01
Appendix
Table 6: Hyperparameter analysis of Ry-DMD on SDXL. We vary one hyperparameter at a time from the default configuration (β,λRY,J,K)=(0.1,0.2,4,3) .
Method
NFE ↓
OCR ↑
HPSv2 ↑
ImageReward ↑
Aesthetic ↑
SD3.5-M (4-step)
4
0.501
30.3
1.02
5.52
+ Ry-DMD
4
0.824
31.2
1.23
5.54
Appendix
Table 7: Ry-DMD on 4-step SD3.5-M. Ry-DMD substantially improves OCR accuracy while preserving the original 4-step sampling budget and improving general image-quality and preference metrics.