Diffusion- and flow-based models have advanced Real-world Image Super-Resolution (Real-ISR), but their multi-step sampling makes inference slow and hard to deploy. One-step distillation alleviates the cost, yet often degrades restoration quality and removes the option to refine with more steps. We present Mean Flows for Super-Resolution (MFSR), a new distillation framework that produces photorealistic results in a single step while still allowing an optional few-step path for further improvement. Our approach uses MeanFlow as the learning target, enabling the student to approximate the average velocity between arbitrary states of the Probability Flow ODE (PF-ODE) and effectively capture the teacher's dynamics without explicit rollouts. To better leverage pretrained generative priors, we additionally improve original MeanFlow's Classifier-Free Guidance (CFG) formulation with teacher CFG distillation strategy, which enhances restoration capability and preserves fine details. Experiments on both synthetic and real-world benchmarks demonstrate that MFSR achieves efficient, flexible, and high-quality super-resolution, delivering results on par with or even better than multi-step teachers while requiring much lower computational cost.
Figures & tables
Figure 1: We present MFSR, a high-capacity one- or few-step super-resolution model that delivers photorealistic restoration of real-world low-resolution images. The number of diffusion inference steps is indicated by ‘s’.
Figure 2: Overview of the MeanFlow Distillation (MFD) pipeline. Left: Training pipeline where the student model is initialized from the teacher. Student model needs two time steps as input. Right: Derivation of the MFD loss. The dotted orange line denotes the teacher’s and student’s PF-ODE, the blue line shows the teacher’s predicted instantaneous velocity, the solid orange line shows the student’s predicted average velocity and the purple line indicates the student’s predicted average velocity over a shorter interval. Taking the limit Δt→0 yields the MFD loss LMFD .
Datasets
Metric
OSEDiff-1s
AddSR-1s
SinSR-1s
CTMSR-1s
S3Diff-1s
TSDSR-1s
MFSR-1s
MFSR-2s
DRealSR
PSNR ↑
27.92
27.77
28.27
28.66
27.53
26.19
24.15
24.29
LPIPS ↓
0.2966
0.3196
0.3730
0.3232
0.3109
0.3116
0.3660
0.3689
FID ↓
135.39
150.18
182.28
162.29
118.49
130.70
143.12
139.56
NIQE ↓
6.4381
6.9321
7.0246
6.1426
6.2142
5.7643
6.0241
6.2711
MUSIQ ↑
64.67
60.85
55.55
59.84
63.94
66.11
64.47
64.45
MANIQA ↑
0.5898
0.5490
0.4907
0.4865
0.6124
0.5820
0.6148
0.6354
Table 1: Quantitative comparison with the state-of-the-art one-step methods across both synthetic and real-world benchmarks. The number of diffusion inference steps is indicated by ‘s’. The best and second best results of each metric are highlighted in red and blue , respectively. S3Diff uses CFG during inference, which actually has 2 number of function evaluations (NFEs).
Figure 3: Qualitative comparison with state-of-the-art methods. All methods perform 1-step inference. Our MFSR is capable of generating vivid details without artifacts or remaining degradations.
Table 5
Figure 5: Qualitative comparison across different sampling steps and our teacher model, DiT4SR. The number of diffusion inference steps is indicated by ’s’.
Figure 6: Visual comparison from ablation study of our CFG strategy.
Figure 7: Effect of sampling steps.
Instantaneous Velocity
w
PSNR ↑
LPIPS ↓
DISTS ↓
FID ↓
NIQE ↓
MUSIQ ↑
MANIQA ↑
CLIPIQA ↑
z1−z0
-
24.21
0.3210
0.2276
121.78
5.5784
65.32
0.6035
0.6474
Original MeanFlow CFG
6
21.20
0.3478
0.2453
120.81
5.8691
67.33
0.6091
0.6951
Ours null
6
25.38
0.2931
0.2151
109.75
5.7146
65.27
0.6184
0.6551
Ours neg
1
24.49
0.2983
0.2237
109.23
5.2234
66.84
0.6273
0.6747
Ours neg
4
22.94
0.3021
0.2255
110.83
5.2317
67.02
0.6317
0.6773
Ours neg
6
21.51
0.3158
0.2295
110.14
5.2421
67.95
0.6389
0.6968
Table 3: Ablation studies for CFG strategy and its scale w .
Appendix figures & tables8 assets
Supplementary material from the paper’s appendix.
Appendix
r
NIQE ↓
MUSIQ ↑
MANIQA ↑
CLIPIQA ↑
0
5.2529
67.23
0.6283
0.6819
0.25
5.2311
67.58
0.6301
0.6863
0.5
5.2421
67.95
0.6389
0.6968
0.75
5.1046
67.76
0.6358
0.6895
Appendix
Table 4: Ablation studies for hyperparameter ratio r .
Figure 8: The LR images used in user study.
Figure 9: Qualitative comparison across different sampling steps and our teacher model, DiT4SR. The number of diffusion inference steps is indicated by ’s’.
Figure 10: Qualitative comparison with state-of-the-art methods. All methods perform 1-step inference. Please zoom in for a better view.
Figure 11: Qualitative comparison with state-of-the-art methods. All methods perform 1-step inference. Please zoom in for a better view.
Figure 12: Qualitative comparison with state-of-the-art methods. All methods perform 1-step inference. Please zoom in for a better view.
Figure 13: 4 × SR results on AI-Generated Content using 3-step sampling.
Figure 14: Result of old photo restoration using 3-step sampling.
Image super-resolution (ISR) has witnessed remarkable progress with diffusion models and flow matching. The dominant text-to-image (T2I) based approaches leverage large-scale foundation models as generative priors, achieving impressive perceptual quality but at the cost of massive model sizes and prohibitive training expenses. Recent flow-matching-based vision-only approaches have made significant strides; however, they adopt standard flow formulations that transport from a pure Gaussian prior to the data distribution, discarding the rich structural information already present in the low-quality (LQ) input. Furthermore, existing single-step acceleration techniques often forfeit the model's multi-step inference capability. In this paper, we propose Residual Flow Matching for Image Super-Resolution (RFMSR), a vision-only framework that centers the source distribution at the LQ latent, reducing transport distance and preserving structural priors throughout the flow trajectory. We further introduce a two-phase training strategy: Phase I pretrains the velocity field via conditional flow matching, while Phase II applies end-to-end supervision to the single-step prediction while retaining the velocity loss across all timesteps, achieving high-quality single-step generation without sacrificing multi-step refinement. Extensive experiments demonstrate that RFMSR achieves comparable or even superior perceptual quality compared to state-of-the-art (SOTA) methods. The source code is available at https://github.com/Faze-Hsw/RFMSR.
Shuwei Huang, Tianyao Luo, Jicheng Liu +2
Huazhong University of Science and Technology · Wuhan University
Pre-trained text-to-image (T2I) diffusion models have shown strong potential for real-world image super-resolution (Real-ISR), owing to their noise-started generation process that enables realistic texture synthesis and captures the one-to-many nature of super-resolution. However, diffusion-based Real-ISR methods still face a fundamental efficiency-quality trade-off. Multi-step methods generate high-quality results by iteratively denoising random Gaussian noise under LR conditioning, but suffer from slow sampling. Recent one-step methods greatly improve efficiency, yet they typically replace noise-started generation with direct LR-to-HR restoration, which weakens stochasticity and limits realistic detail synthesis. To address this issue, we propose SMFSR, a noise-started one-step Real-ISR framework via LR-conditioned SplitMeanFlow and GAN refinement. SMFSR preserves the random-noise starting point of diffusion models and learns a direct noise-to-HR mapping conditioned on the LR image. To this end, Interval Splitting Consistency distills the multi-step generative trajectory into a single average-velocity prediction, enabling efficient one-step generation. To compensate for the reduced opportunity for progressive refinement, we further introduce a GAN refinement stage, where a DINOv3-based discriminator enhances realistic texture synthesis and variational score distillation aligns the generated outputs with the natural image distribution under a frozen diffusion teacher. Extensive experiments demonstrate that SMFSR achieves state-of-the-art perceptual quality among one-step diffusion-based Real-ISR methods while retaining fast single-step inference.
Wei Zhu, Kai Zhang, Yu Zheng +3
Nanjing University of Science and Technology · Nanjing University · Huawei
Diffusion models (DMs) have demonstrated remarkable success in real-world image super-resolution (SR), yet their reliance on time-consuming multi-step sampling largely hinders their practical applications. While recent efforts have introduced few- or single-step solutions, existing methods either inefficiently model the process from noisy input or fail to fully exploit iterative generative priors, compromising the fidelity and quality of the reconstructed images. To address this issue, we propose FlowSR, a novel approach that reformulates the SR problem as a rectified flow from low-resolution (LR) to high-resolution (HR) images. Our method leverages an improved consistency learning strategy to enable high-quality SR in a single step. Specifically, we refine the original consistency distillation process by incorporating HR regularization, ensuring that the learned SR flow not only enforces self-consistency but also converges precisely to the ground-truth HR target. Furthermore, we introduce a fast-slow scheduling strategy, where adjacent timesteps for consistency learning are sampled from two distinct schedulers: a fast scheduler with fewer timesteps to improve efficiency, and a slow scheduler with more timesteps to capture fine-grained texture details. Extensive experiments demonstrate that FlowSR achieves outstanding performance in both efficiency and image quality. Code: this https URL.
Jiaqi Xu, Wenbo Li, Haoze Sun +8
The Chinese University of Hong Kong · Huawei Noah’s Ark Lab · HKUST (GZ) +1