Generative real-world image super-resolution (Real-ISR) can synthesize visually convincing details from severely degraded low-resolution (LR) inputs, yet its stochastic sampling makes a critical failure mode hard to avoid: outputs may look sharp but be unfaithful to the LR evidence, exhibiting semantic or structural hallucinations. Preference-based reinforcement learning (RL) is a natural fit because each LR input yields a rollout group of candidate restorations. However, effective alignment in Real-ISR is hindered by three coupled challenges: (i) the lack of an LR-referenced faithfulness signal that is robust to degradation yet sensitive to localized hallucinations, (ii) a rollout-group optimization bottleneck where scalarizing heterogeneous rewards before normalization compresses objective-wise contrasts and weakens DiffusionNFT-style reward-weighted updates, and (iii) limited coverage of real degradations, which restricts rollout diversity and preference signal quality. We propose LucidNFT, a multi-reward RL framework for flow-matching Real-ISR. LucidNFT introduces LucidConsistency, a degradation-invariant and hallucination-sensitive LR-referenced evaluator trained with content-consistent degradation pools and original-inpainted hard negatives; a decoupled reward normalization strategy that preserves objective-wise contrasts within each LR-conditioned rollout group before fusion; and LucidLR, a large-scale collection of real-world degraded images for robust RL fine-tuning. Extensive experiments show that LucidNFT improves perceptual quality on strong flow-based Real-ISR baselines while generally maintaining LR-referenced consistency across diverse real-world scenarios.
Figures & tables
Figure 1: Overview of LucidConsistency. Same-content views under different degradation levels form positive pools, while other pools serve as negatives. Original–inpainted pools are paired hard negatives: they are globally similar and spatially aligned, but differ in localized AI-generated regions. LucidConsistency extracts global and native-token representations with a frozen Qwen3-VL-Embedding backbone and trainable LoRA adapters, and optimizes global-local pool contrastive losses.
Figure 2
Figure 3: Training curves of LucidNFT and DPO fine-tuning. Curves report fast validation rewards computed with reduced inference steps. LucidNFT steadily improves perceptual rewards while keeping LucidConsistency stable; the final full evaluation in Table 2 shows improved LR-faithfulness over the corresponding backbones.
Figure 4: Visual comparison on RealLQ250 [ 2 ] . Compared with the corresponding backbones and LucidFlux(+DPO), LucidNFT variants recover more accurate text structures and finer local textures while preserving LR-supported semantics.
Table 5
Appendix figures & tables6 assets
Supplementary material from the paper’s appendix.
Appendix
Figure 5: Annotation interface for LR-faithfulness evaluation. Annotators are shown the LR image and four randomly ordered SR candidates, and rank the candidates by faithfulness to the LR evidence rather than perceptual sharpness alone.
Figure 6: Primary degradation distribution across four real-world low-quality datasets. Each image contributes to exactly one dominant degradation category. Compared with RealSR and DRealSR, which are strongly dominated by defocus blur, LucidLR presents a more balanced primary degradation distribution over multiple categories.
Figure 7: Occurrence frequency of degradation categories across four real-world low-quality datasets: LucidLR, RealLQ250, RealSR, and DRealSR. Occurrence frequency is computed in a multi-label manner, so percentages do not sum to 100%. LucidLR exhibits substantially broader degradation coverage and a richer long-tail distribution than existing benchmark datasets.
Figure 8: Additional qualitative comparisons on RealLQ250 [ 2 ] . Compared with the LucidFlux baseline, LucidFlux(+LucidNFT) recovers clearer local details while better preserving LR-supported structures. See Fig. 4 in the main paper for the primary qualitative results.
Figure 9: Optimization curves under different reward formulations on LucidFlux. LucidNFT with decoupled reward normalization achieves higher UniPercept rewards and smoother convergence than IQA-only RL and scalar-aggregated multi-reward RL.
Figure 10: Advantage separability analysis on LucidFlux using RealLQ250 [ 2 ] . (a) Distinct Advantage Group Count (DAGC) versus rollout count M ; (b) mean maximum pairwise advantage gap ∣ΔA∣ versus M ; (c) distribution of ∣ΔA∣ at M=12 . Compared with scalar-first reward aggregation under the same DiffusionNFT objective, decoupled normalization produces larger advantage gaps and more distinct advantage levels, indicating reduced advantage compression within LR-conditioned rollout groups.
Benefiting from the powerful generative priors of diffusion models, diffusion-based real-world image super-resolution (Real-ISR) methods have demonstrated impressive performance.To achieve efficient Real-ISR, several recent works have designed one-step diffusion-based models.Howerver, unmediatedly feeding LR into a diffusion model creates a distributional gap with the model's original input.A straightforward approach to reduce the distribution gap is to introduce noise to the LR latents. However, directly adding noise inevitably corrupts the content of the LR images.In this study, we propose DNF-SR, a Dual-input and Negative-aware Feature fine-tuning method for Real-ISR.Specifically, we use a dual-input strategy that concatenates the original LR image with the noisy LR input and feeds them into a diffusion-based image editing model, ensuring both high-fidelity one-step super-resolution and improved perceptual and content consistency.Additionally, the noise present in the noisy LR input introduces randomness and diversity into the outputs. We exploit this property and propose a post-training optimization method, Negative-aware Feature Fine-Tuning (NF2T), which guides the model toward producing higher-quality results.NF^2T classifies multiple outputs into positive and negative subsets and then defines implicit policy improvement directions in both the image and feature spaces, thereby further enhancing the stability of the optimization.Extensive experiments show that DNF-SR outperforms other methods.Code will be released.
Shuhao Han, Wenjie Liao, Hayden Vance +4
VCIP, CS, Nankai University · Independent Researcher
Real-world image super-resolution (Real-ISR) requires balancing structural fidelity to degraded observations with realistic detail synthesis. However, existing generative Real-ISR methods often rely on entangled conditioning mechanisms, leading to structural drift or semantically inconsistent details. To address this issue, we propose Visual In-Context Restoration (VICR), a Diffusion Transformer (DiT)-based framework that formulates Real-ISR as image completion. Specifically, we introduce a decoupled visual prior injection mechanism that derives local and global cues from the low-quality (LQ) image: local cues help recover image structures and support high-frequency detail synthesis, while global cues guide overall generation and promote semantic consistency. For ambiguous regions under severe degradation, VICR employs an inference-time agent to refine semantic prompts using visual evidence from the LQ input while keeping model parameters fixed. Experiments show that VICR achieves state-of-the-art performance across multiple Real-ISR benchmarks with only 127M trainable parameters.
Qichang Zhang, Hailong Wang, Baiang Li +4
Faculty of Science and Technology, University of Macau · 2Nullmax · 3Hefei University of Technology +1
Real-world image super-resolution (Real-ISR) aims to preserve structures supported by the degraded observation while reconstructing perceptually realistic details. However, existing Real-ISR methods largely optimize fidelity and perceptual quality within a shared network, causing the two objectives to interfere throughout training and making their balance difficult to control. Recent one-step methods reduce sampling steps, yet often inherit both this coupled optimization behavior and the expensive high-resolution backbone of their multi-step predecessors. We argue that efficient Real-ISR requires not only a shorter sampling trajectory, but also specialized modeling of faithful reconstruction and perceptual detail synthesis. Based on this insight, we propose PixelIR, a fidelity-perception decoupling framework built upon pixel-space image-residual flow matching. PixelIR first learns an image flow that maps the degraded observation to a faithful reconstruction. Then, a residual flow synthesizes the missing perceptual details from noise without repeatedly relearning or overwriting the complete restoration solution. We further distill the teacher into a deployment-oriented one-step student within a coarse-to-fine pyramid architecture. Extensive experiments show that PixelIR achieves leading PSNR, SSIM, and LPIPS on both RealSR and DRealSR. The final model completes pixel-space restoration in a single evaluation with only 32.9M parameters, 89.7G MACs, and 8.5ms latency, demonstrating a strong practical fidelity-perception-efficiency balance.
Bingtian Qiao, Yue Shi, Yong Guo +2
Shanghai Jiao Tong University · Fuzhou University · Shanghai AI Laboratory