Organizations: School of Computer Science and Technology, Wuhan University of Science and Technology, Wuhan, China · Hubei Province Key Laboratory of Intelligent Information Processing and Real-time Industrial System · Harbin Institute of Technology Zhengzhou Research Institute, Zhengzhou, China · Guangdong Laboratory of Artificial Intelligence and Digital Economy (SZ) · Huawei Technologies Ltd.
Many classic opera videos exhibit poor visual quality due to the limitations of early filming equipment and long-term degradation during storage. Although real-world video super-resolution (RWVSR) has achieved significant advances in recent years, directly applying existing methods to degraded opera videos remains challenging. The difficulties are twofold. First, accurately modeling real-world degradations is complex: simplistic combinations of classical degradation kernels fail to capture the authentic noise distribution, while methods that extract real noise patches from external datasets are prone to style mismatches that introduce visual artifacts. Second, current RWVSR methods, which rely solely on degraded image features, struggle to reconstruct realistic and detailed textures due to a lack of high-level semantic guidance. To address these issues, we propose a Text-guided Dual-Branch Opera Video Super-Resolution (TextOVSR) network, which introduces two types of textual prompts to guide the super-resolution process. Specifically, degradation-descriptive text, derived from the degradation process, is incorporated into the negative branch to constrain the solution space. Simultaneously, content-descriptive text is incorporated into a positive branch and our proposed Text-Enhanced Discriminator (TED) to provide semantic guidance for enhanced texture reconstruction. Furthermore, we design a Degradation-Robust Feature Fusion (DRF) module to facilitate cross-modal feature fusion while suppressing degradation interference. Experiments on our OperaLQ benchmark show that TextOVSR outperforms state-of-the-art methods both qualitatively and quantitatively. The code is available at https://github.com/ChangHua0/TextOVSR.
Figures & tables
Figure 1: Frameworks for real-world video super-resolution. (a) The classical synthetic degradation pipeline (D) simulates real degradations for the VSR model. (b) Real degradation modeling extracts authentic noise from external datasets and applies a negative constraint ( Lneg ) to enhance robustness. (c) Our proposed TextOVSR introduces text-guided priors to enrich image features and model diverse degradations in the feature space. λ controls the mixing ratio.
Figure 2: Generation process of degradation- and content-descriptive texts. Degradation-descriptive text is generated according to different intensity levels in the high-order degradation pipeline, while content-descriptive text is produced directly from high-resolution inputs (HRs) using a multimodal large language model (MLLM), rather than from degraded low-resolution videos (LRs).
Figure 3: The proposed TextOVSR network and TED adopt a two-stage training scheme. In the first stage, only TextOVSR is trained. The positive branch (blue) takes content-descriptive text ( TC ) and degraded videos ( Vlr ) as input, while the negative branch (red) takes degradation-descriptive text ( TD ) and mixed-noise videos ( Vlr ). Text features are extracted using a frozen CLIP encoder and fused with image features through the proposed DRF module. In the second stage, TextOVSR serves as the generator and TED as the discriminator. Adversarial training refines texture realism by selecting reliable textual features and integrating them with reconstructed image features. Here, t−1 , t , and t+1 denote three consecutive frames, with detailed propagation described in Section 3.2 .
Figure 4: The proposed Degradation-Robust Feature Fusion (DRF) module. FIt+1 and FTt+1 denote the image and text feature vectors, respectively, and Mt represents the fused feature.
Figure 5: OperaLQ Dataset. Our OperaLQ dataset consists of real degraded opera videos with varying content and resolutions.
Figure 6: Qualitative comparison with other methods. (Zoom-in for best view.)
Table 2: Comparison of quantitative results with different components of our method on the OperaLQ dataset. w/o T,N denotes enhancing only the negative branch propagation, while w/o T denotes enhancing both positive and negative branches simultaneously. w TD and w TD&TC denote using only degradation-descriptive text and using both degradation-descriptive and content-descriptive texts.
Figure 7: Qualitative comparison of different component ablation studies. (Zoom-in for best view.)
NRQM ↑
CLIPIQA+ ↑
TOPIQ ↑
NIQE ↓
Caption
5.5552
0.5718
0.4602
3.6206
Text
5.6838
0.5667
0.4636
3.5139
Table 3: Quantitative comparison of coarse-grained and fine-grained textual descriptions , where Caption denotes coarse-grained textual descriptions and Text denotes fine-grained ones.
Figure 8: Qualitative comparison of coarse-grained and fine-grained textual descriptions. (Zoom-in for best view.)
NRQM ↑
CLIPIQA+ ↑
TOPIQ ↑
NIQE ↓
CLIP
5.0118
0.3081
0.2860
5.3520
UNet
5.6838
0.5667
0.4636
3.5139
TED
5.8184
0.5659
0.4733
3.4291
Table 4: Quantitative comparison of different discriminators.
Figure 9: Qualitative comparison of different Discriminators. (Zoom-in for best view.)
Location
NRQM ↑
CLIPIQA+ ↑
TOPIQ ↑
NIQE ↓
without
5.7761
0.3990
0.4354
4.0753
before
5.4280
0.5438
0.4316
3.7274
after
5.4949
0.5462
0.4436
3.7303
Table 5: Quantitative comparison of DRF at different positions in the negative branch (before and after image feature extraction). without indicates no DRF is added.
Figure 10: Qualitative comparison of DRF at different positions in the negative branch. (Zoom-in for best view.)
Single image super-resolution aims to reconstruct high-resolution (HR) images from low-resolution (LR) inputs. Training SR models typically requires paired HR-LR data, which is difficult to obtain in reality. As a result, most methods synthesize LR images by artificially degrading HR images with handcrafted kernels or camera ISP adjustments. However, these synthetic degradations fail to capture the complexity of real LR images, leading to poor generalization in practice. To address this, we observe that even within a single high-quality image, regions at different depths exhibit varying resolutions, where distant regions act as LR patches and closer ones as HR patches. This allows the extraction of real, degradation-induced LR patches from real images. Since these LR patches lack paired HR counterparts, we propose LA-SR (Language Assistant for SR), a novel framework for unpaired SR. The key idea of LA-SR is to redefine unpaired SR in the language space, using vision-language models to bridge the LR-HR gap. LA-SR projects images into a semantically rich space representing both content and quality, and applies two language-guided losses: linguistic content loss to preserve semantic fidelity, and linguistic quality loss to enhance perceptual realism. With this alignment, LA-SR effectively super-resolves real LR inputs, producing realistic outputs that overcome the limitations of synthetic-data-trained methods.
Joonkyu Park, Kyoung Mu Lee
Dept. of ECE&ASRI, 2IPAI, Seoul National University
Despite recent advances, single-image super-resolution (SR) remains challenging, especially in real-world scenarios with complex degradations. Diffusion-based SR methods, particularly those built on Stable Diffusion, leverage strong generative priors but commonly rely on text conditioning derived from semantic captioning. Such textual descriptions provide only high-level semantics and lack the spatially aligned visual information required for faithful restoration, leading to a representation gap between abstract semantics and spatially aligned visual details. To address this limitation, we propose GramSR, a one-step diffusion-based SR framework that replaces text conditioning with dense visual features extracted from the low-resolution input using a pre-trained DINOv3 encoder. GramSR adopts a three-stage LoRA architecture, where pixel-level, semantic-level, and texture-level LoRA modules are trained sequentially. The pixel-level module focuses on degradation removal using ℓ2 loss, the semantic-level module enhances perceptual details via LPIPS and CSD losses, and the texture-level module enforces feature correlation consistency through a Gram matrix loss computed from DINOv3 features. At inference, independent guidance scales enable flexible control over degradation removal, semantic enhancement, and texture preservation. Extensive experiments on standard SR benchmarks demonstrate that GramSR consistently outperforms existing one-step diffusion-based methods, achieving superior structural fidelity and texture realism. The code for this work is available at: https://github.com/aimagelab/GramSR.
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