Pith. sign in

REVIEW 4 major objections 5 minor 78 references

The paper claims LPM is the first generative video restoration model deployed at industrial scale, processing about 45% of Kuaishou's viewing time and cutting bitrate by 20% at equal perceptual quality.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-02 05:06 UTC pith:5YD4FCZJ

load-bearing objection A credible industrial-systems report with a real evaluation gap: the headline bitrate claim is certified by the same author-built metric used to filter training data, so the economic claims need an independent quality check before being taken at face value. the 4 major comments →

arxiv 2607.13460 v1 pith:5YD4FCZJ submitted 2026-07-15 cs.CV

LPM: Industrial-Scale Generative Video Restoration

Bichuan Zhu , Fulin Li , Jiachao Gong , Jinhua Hao , Kai Zhao , Kun Yuan , Pengcheng Xu , Qiang Wang
show 9 more authors
Qiao Mo Yanlong Yuan Yizhen Shao Yuxiao Hu Zixi Tuo Ming Sun Chao Zhou Bin Chen Bin Yu
This is my paper
classification cs.CV
keywords video restorationdiffusion modelsuser-generated contentindustrial deploymentbitrate reductiontemporal consistencyconsistency modelsquality assessment
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces LPM, a diffusion-based framework for restoring user-generated videos with complex real-world degradations. It argues that by combining a large curated training dataset, a progressive image-to-video training scheme, and efficient inference, a single system can deliver high-fidelity, temporally consistent restoration for arbitrarily long videos. The central claim is that this is the first such generative model deployed at industrial scale, serving roughly 45% of Kuaishou's viewing time and reducing bitrate by more than 20% at comparable perceptual quality. The authors aim to show that generative restoration is not just a research demo but a practical, cost-effective component of a large video delivery pipeline.

Core claim

The central discovery is a working industrial deployment: LPM, a two-stage diffusion model (LPM-Image for spatial restoration, LPM-Video for temporal consistency) that, when integrated into Kuaishou's encoding pipeline, improves online QoE metrics and achieves over 20% bitrate reduction at equal perceptual quality. The paper attributes this success to several design choices: constructing a billion-scale clean dataset with strict quality and texture filtering, training with a coarse-to-fine curriculum and a fidelity refinement stage that preserves text and fine structures, using factorized 2D+1D attention and position-free temporal attention for arbitrary-length video, a temporal-pyramid infe

What carries the argument

The key machinery is the combination of a progressive image-to-video training paradigm and a temporal-pyramid inference mechanism. LPM-Image learns a spatial prior via a conditional flow-matching objective; LPM-Video extends it by inserting factorized temporal attention blocks (2D+1D), removing temporal positional encodings (NoPE), and training with mask-guided cross-clip conditioning that simulates long-video inference. At serving time, shot-level keyframes are restored jointly to form a 'storyboard' of anchors; each window is then restored conditioned on both a fixed shot-level anchor and the overlapping previous window, recursively subdividing long shots. This hierarchy shortens the depen

Load-bearing premise

The quality evidence hinges on KVQ, a no-reference metric developed and used by the authors' team, accurately reflecting human perception; if KVQ does not track human judgments, the 'comparable perceptual quality' bitrate claim and the benchmark comparisons lose their basis, since no independent human rating or external audit is supplied.

What would settle it

A blind human study comparing LPM's restored low-bitrate video to the in-house codec's higher-bitrate video on a diverse UGC sample; if raters do not judge the two as perceptually equivalent, the 20% bitrate savings claim collapses. Alternatively, running the public benchmark evaluations with an externally validated perceptual metric (e.g., human opinion scores) and checking whether LPM's lead over baselines persists would test the dependence on KVQ.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If LPM's claims hold, diffusion-based restoration becomes a standard step in large-scale video pipelines rather than a research-only post-processing tool.
  • The bitrate reduction at equal perceptual quality implies generative restoration can substitute for some codec work, yielding direct bandwidth cost savings at current scale.
  • The temporal-pyramid inference with position-free attention offers a recipe for arbitrary-length video generation or restoration without fixed training windows.
  • The acceleration stack (consistency distillation to 1-2 steps plus quantization) makes diffusion-based enhancement practical for real-time, high-throughput serving.
  • The fidelity refinement with region-level masks shows how to reduce hallucinations in text and fine structures, a known pain point of generative restoration.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If KVQ is accepted as a perceptual metric, the quality comparisons are internally consistent; an independent, human-rated benchmark would be needed to confirm that the same quality holds outside the authors' evaluation setup.
  • The 45% viewing-time share and 20% bitrate savings are both reported from internal deployment data; an external replication of the bitrate comparison would strengthen the case that generative restoration, not dataset- or metric-specific tuning, drives the gain.
  • The architecture choices (NoPE, mask guidance, temporal pyramid) may transfer to other video-to-video tasks such as inpainting, frame interpolation, or enhancement of AI-generated content.
  • A testable extension: apply the same three-stage truncated consistency distillation to a publicly available restoration model to see whether 1-2 step speedups generalize without the proprietary dataset scale.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper introduces LPM, a two-stage diffusion-based framework for real-world video restoration. LPM-Image is a DiT-based image restoration model trained on a large proprietary dataset (Kwai UltraVision) with KVQ/KTQ filtering; LPM-Video extends it with factorized 2D+1D temporal attention, position-free temporal attention, and mask-guided cross-clip conditioning. A temporal-pyramid inference scheme is proposed for long-video consistency, and acceleration via consistency distillation and quantization is described. The authors report state-of-the-art numbers on public and internally constructed benchmarks, and claim industrial deployment at Kuaishou affecting 45% of viewing time and a >20% bitrate reduction at comparable perceptual quality.

Significance. If substantiated, the industrial deployment and bandwidth-savings claims would make LPM a notable systems contribution, showing that diffusion-based restoration can be economically viable. The architectural components — factorized 2D+1D DiT, NoPE, mask guidance, temporal-pyramid inference — are well motivated and appear internally consistent. However, the strength of the evidence is undermined by the reliance on the Kuaishou-developed KVQ metric both as a training filter and as the primary evaluation criterion, and by the absence of measurement details for the production claims. The claims are therefore plausible but not yet adequately supported.

major comments (4)
  1. [§2.1.1, §3.1, Tables 3–5] KVQ is used both to filter training data (thresholds 4.0/4.7 in §2.1.1) and as a headline evaluation metric in Tables 3–5. Because KVQ is developed by the same team, the perceptual comparisons and the 'comparable perceptual quality' condition in the bitrate claim (Abstract, §1, §4) may be optimized to the evaluator. The paper does not provide an independent human study or an external audit of KVQ. Please add a human evaluation or at least report correlation of KVQ with human ratings on a held-out set, and show that the training-filtering thresholds do not artificially inflate the evaluation scores.
  2. [§1, §4] The central production claims — 45% of total viewing time and >20% bitrate reduction at comparable perceptual quality — are stated without a description of the measurement methodology, the QoE metrics used, the comparison codec settings, or any uncertainty quantification. As written these claims cannot be verified or reproduced. Please provide a detailed evaluation protocol, including the definition of 'comparable perceptual quality', the bitrate measurement procedure, and ideally confidence intervals.
  3. [§3.1, Tables 4–5] LPM-Benchmark is an internally constructed benchmark (192 videos) with no description of selection criteria, no external validation, and no release. Using it as a basis for the state-of-the-art claims is problematic. Please describe the construction process, provide human validation of the degradation types, and make the benchmark available, or restrict SOTA claims to public benchmarks.
  4. [§2.5, Table 2] The acceleration section reports a 39.5× end-to-end speedup and claims 'PSNR>43dB, SSIM>0.99, KVQ difference<0.01 relative to the unaccelerated model' without specifying the evaluation dataset, resolution, hardware, or measurement procedure. A PSNR above 43 dB on real-world restored video after 1-step distillation would be unusually high and needs detailed verification. Please clarify the evaluation setup and provide per-stage comparisons.
minor comments (5)
  1. [Eq. (2)] Equation (2) has a trailing comma and the mask convention would benefit from an explicit statement that M=1 denotes regions to be preserved. As written, the equation is correct but slightly ambiguous.
  2. [Table 1] The VAE comparison is reported on a 'challenging internal benchmark' that is not described. Please specify the dataset, the degradation types, and the number of samples so the reader can judge the comparison.
  3. [Tables 3–5] All quantitative tables report single-point metrics without standard deviations or significance tests. Given that the reported margins are often small (e.g., Table 3 MUSIQ 64.68 vs 64.33), error bars or repeated-evaluation statistics would improve confidence.
  4. [Abstract and §1] There are several typographical issues in the abstract and Section 1 (e.g., 'theLarge', 'thefirst') that should be corrected in the final version.
  5. [§1, Contribution 3] The claim of being 'the first generative video restoration model deployed at industrial scale' should be supported by a comparison with prior industrial systems or a clear definition of 'industrial scale'.

Circularity Check

1 steps flagged

Partial metric circularity: KVQ both filters LPM's training targets and certifies 'comparable perceptual quality' in the bitrate claim; public benchmarks keep the central claim partly independent.

specific steps
  1. self definitional [§2.1.1, §3.1, §3.2, Abstract/§1/§4]
    "We therefore use Kuaishou Visual Quality (KVQ) [31, 40, 67], a no-reference metric that evaluates perceptual factors such as blur, noise, sharpness, and blocking artifacts. ... We retain samples with KVQ scores above 4.0 for large-scale training and raise the threshold to 4.7 for high-quality fine-tuning. ... We additionally report MUSIQ [22], CLIP-IQA [53], and KVQ [31], which assess perceptual quality without requiring reference images. ... At comparable perceptual quality, it reduces bitrate by 20% relative to Kuaishou's in-house codec."

    KVQ is load-bearing in two coupled roles. In §2.1.1 it filters the clean-target distribution (thresholds 4.0/4.7), so LPM is trained to reproduce KVQ-favorable content. In §3.1–§3.2 the same metric is a primary no-reference evaluation score, and the production claim of 'comparable perceptual quality' (Abstract/§1/§4) is not anchored to any independent human-rated comparison in the paper. Hence the reported KVQ gains and the bitrate saving are partly self-referential: the model is fitted to a KVQ-filtered distribution and then certified by the same KVQ metric. This is only partial circularity because the flow-matching loss is not KVQ itself, because public benchmarks (RealSR, REDS, VideoLQ, YouHQ, SPMCS) and independent metrics (MUSIQ, CLIP-IQA, PSNR) support the quality claims, and because

full rationale

The only concrete circular step is the KVQ coupling: the same in-house no-reference metric is used to curate training targets and to evaluate output quality, with the production bitrate claim resting on 'comparable perceptual quality' without an explicit independent anchor. This is a genuine but partial self-referentiality. Other claims—first industrial-scale deployment, temporal-pyramid inference, 39.5x acceleration—are not derived from their inputs by construction. LPM-Benchmark is internally constructed, but it is not load-bearing because the paper also reports results on public benchmarks (RealSR, RealSet, REDS, VideoLQ, YouHQ, SPMCS) using independent metrics. The KVQ citations [31,40,67] are peer-reviewed publications, so they are real evidence rather than an unverified self-citation chain. The central quality result has independent grounding; the headline economic claim, however, would be strengthened by a human-rated or otherwise external quality comparison. Accordingly, the score reflects partial metric circularity, not a fully circular derivation.

Axiom & Free-Parameter Ledger

5 free parameters · 6 axioms · 4 invented entities

The central claims depend on the internal KVQ/KTQ metrics, the synthetic degradation approximation, and self-reported production telemetry. Each of these is an unverified premise; the model architecture itself uses standard building blocks and does not introduce a new physical entity or force.

free parameters (5)
  • KVQ quality thresholds (4.0 / 4.7) = 4.0 (pretraining), 4.7 (high-quality fine-tuning)
    Hand-chosen thresholds determine which images/videos enter the training corpus and the high-quality fine-tuning tier; changes would shift the learned output distribution and likely change benchmark scores (§2.1.1).
  • KTQ texture thresholds = unspecified
    Texture-complexity filtering uses an internal KTQ metric; no thresholds are given, and the metric is not formally defined, so the data selection is not reproducible (§2.1.2).
  • Mask-guidance HQ-reference probability = 0.5
    During high-quality fine-tuning, the LQ condition at the reference frame is replaced by the HQ frame with 50% probability; this hyperparameter controls the strength of cross-clip conditioning (§2.3.2).
  • Large-motion temporal subsampling intervals = Interval=1 or 2 (illustrated), otherwise unspecified
    High-frame-rate or low-motion clips are temporally subsampled to increase apparent motion; the chosen intervals reshape the motion distribution and are not specified as a policy (§2.3.2, Fig. 4).
  • Temporal-pyramid recursion depth / anchor sampling density = unspecified
    The number of pyramid levels and the shot keyframe density is a heuristic that trades long-range consistency against computation; not quantified (§2.4).
axioms (6)
  • domain assumption KVQ accurately measures perceived visual quality in the UGC regime.
    KVQ is used for data filtering (§2.1.1) and as a primary evaluation metric (Tables 3–5); the quality and bitrate-equivalent claims collapse if KVQ does not track human perception.
  • domain assumption KTQ texture complexity is a valid proxy for restoration difficulty/value.
    KTQ filters out low-texture data; if texture complexity does not predict perceptual benefit, the data distribution may be misaligned with the deployment target (§2.1.2).
  • domain assumption Synthesized degradations (BSRGAN + FFmpeg compression) approximate in-the-wild UGC degradations.
    All training uses synthetic LQ/HQ pairs; the transfer to real UGC depends on this approximation (§2.2.2, §2.3.2).
  • domain assumption Position-free temporal attention (NoPE) preserves temporal identity and extrapolates to arbitrary-length videos.
    The arbitrary-length claim rests on NoPE plus anchor conditioning; if absolute temporal order matters for long-range consistency, the model lacks an explicit signal to maintain it (§2.3.1).
  • domain assumption Mask-guided recursive conditioning with a 50% HQ reference substitution matches deployment-time inference.
    The model is trained to treat the last restored frame of the previous clip as a reference; deployment assumes this training simulation transfers (§2.3.2, §2.4).
  • domain assumption The internal production measurements (45% viewing time, 20% bitrate reduction) are causally attributable to LPM and correctly measured.
    These numbers are self-reported with no methodology; QoE improvements could be confounded by other platform changes (§1, §4).
invented entities (4)
  • KVQ (Kuaishou Visual Quality) metric no independent evidence
    purpose: No-reference metric used to filter training samples and to evaluate restoration quality.
    Published by Kuaishou-affiliated authors, but no external validation or human-rating correlation is provided in this paper; it is simultaneously the filter and the score (§2.1.1, Tables 3–5).
  • KTQ (Kuaishou Texture Quality) metric no independent evidence
    purpose: Filters data by texture complexity (Laplacian + DCT energy).
    Not formally defined or released; no validation that texture richness selects better restoration training targets (§2.1.2).
  • Kwai UltraVision Dataset no independent evidence
    purpose: Billion-scale internal training corpus of high-quality, high-texture images/videos.
    Not released; quality scores are internal; cannot be audited (§2.1.1).
  • LPM-Benchmark no independent evidence
    purpose: Internal 192-video benchmark for in-the-wild evaluation.
    Not released; constructed by the authors; used to support SOTA claims (§3.1).

pith-pipeline@v1.3.0-alltime-deepseek · 17339 in / 14589 out tokens · 138545 ms · 2026-08-02T05:06:52.735201+00:00 · methodology

0 comments
read the original abstract

We present the Large Processing Model (LPM), a diffusion-based generative framework for photorealistic video restoration under complex, in-the-wild degradations. To our knowledge, LPM is the first generative video restoration model deployed at industrial scale. LPM addresses the diverse degradations in user-generated content (UGC) through a unified system encompassing large-scale data engineering, foundation-model training, and efficient inference. Its enhanced architecture, progressive training strategy, and temporal-pyramid inference mechanism jointly enable high-fidelity, temporally consistent restoration of arbitrarily long videos across the broad content distribution encountered on UGC platforms. LPM has been deployed in production at Kuaishou, where videos processed by the model account for approximately 45% of total viewing time, delivering consistent improvements across key quality-of-experience metrics. Beyond perceptual enhancement, LPM delivers substantial system-level benefits: at comparable perceptual quality, it reduces bitrate by 20% relative to Kuaishou's in-house codec, yielding annual bandwidth cost savings on the order of hundreds of millions. Its low serving cost also enables integration into products such as Kling, demonstrating that generative restoration can be practical, scalable, and cost-effective for large-scale video processing.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

78 extracted references · 15 linked inside Pith

  1. [1]

    Gpt-4 technical report.arXiv preprint arXiv:2303.08774, 2023

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. Gpt-4 technical report.arXiv preprint arXiv:2303.08774, 2023

    Pith/arXiv arXiv 2023

  2. [2]

    Enhancevideoquality api documentation, 2020.https://next.api.aliyun.com/api/videoenhan/ 2020-03-20/EnhanceVideoQuality?RegionId=cn-shanghai

    Alibaba Cloud. Enhancevideoquality api documentation, 2020.https://next.api.aliyun.com/api/videoenhan/ 2020-03-20/EnhanceVideoQuality?RegionId=cn-shanghai. Accessed: October 1, 2025

    2020

  3. [3]

    Layer normalization.arXiv preprint arXiv:1607.06450, 2016

    Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E Hinton. Layer normalization.arXiv preprint arXiv:1607.06450, 2016

    Pith/arXiv arXiv 2016

  4. [4]

    Vivid-vr: Distilling concepts from text-to-video diffusion transformer for photorealistic video restoration.arXiv preprint arXiv:2508.14483, 2025

    Haoran Bai, Xiaoxu Chen, Canqian Yang, Zongyao He, Sibin Deng, and Ying Chen. Vivid-vr: Distilling concepts from text-to-video diffusion transformer for photorealistic video restoration.arXiv preprint arXiv:2508.14483, 2025

    Pith/arXiv arXiv 2025

  5. [5]

    Tuning real- world image restoration at inference: A test-time scaling paradigm for flow matching models.arXiv preprint arXiv:2603.22027, 2026

    Purui Bai, Junxian Duan, Pin Wang, Jinhua Hao, Ming Sun, Chao Zhou, and Huaibo Huang. Tuning real- world image restoration at inference: A test-time scaling paradigm for flow matching models.arXiv preprint arXiv:2603.22027, 2026

    arXiv 2026

  6. [6]

    Instantvir: Real-time video inverse problem solver with distilled diffusion prior

    Weimin Bai, Suzhe Xu, Yiwei Ren, Jinhua Hao, Ming Sun, Wenzheng Chen, and He Sun. Instantvir: Real-time video inverse problem solver with distilled diffusion prior. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 16583–16592, 2026

    2026

  7. [7]

    Plug-and-play tri-branch invertible block for image rescaling

    Jingwei Bao, Jinhua Hao, Pengcheng Xu, Ming Sun, Chao Zhou, and Shuyuan Zhu. Plug-and-play tri-branch invertible block for image rescaling. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 1826–1834, 2025

    2025

  8. [8]

    Toward real-world single image super- resolution: A new benchmark and a new model

    Jianrui Cai, Hui Zeng, Hongwei Yong, Zisheng Cao, and Lei Zhang. Toward real-world single image super- resolution: A new benchmark and a new model. InProceedings of the IEEE/CVF international conference on computer vision, pages 3086–3095, 2019

    2019

  9. [9]

    Investigating tradeoffs in real-world video super-resolution

    Kelvin CK Chan, Shangchen Zhou, Xiangyu Xu, and Chen Change Loy. Investigating tradeoffs in real-world video super-resolution. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5962–5971, 2022

    2022

  10. [10]

    Adversarial diffusion compression for real-world image super-resolution

    Bin Chen, Gehui Li, Rongyuan Wu, Xindong Zhang, Jie Chen, Jian Zhang, and Lei Zhang. Adversarial diffusion compression for real-world image super-resolution. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 28208–28220, 2025

    2025

  11. [11]

    Cassr: Activating image power for real-world image super-resolution.arXiv preprint arXiv:2403.11451, 2024

    Haolan Chen, Jinhua Hao, Kai Zhao, Kun Yuan, Ming Sun, Chao Zhou, and Wei Hu. Cassr: Activating image power for real-world image super-resolution.arXiv preprint arXiv:2403.11451, 2024

    Pith/arXiv arXiv 2024

  12. [12]

    Effective diffusion transformer architecture for image super-resolution

    Kun Cheng, Lei Yu, Zhijun Tu, Xiao He, Liyu Chen, Yong Guo, Mingrui Zhu, Nannan Wang, Xinbo Gao, and Jie Hu. Effective diffusion transformer architecture for image super-resolution. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 2455–2463, 2025

    2025

  13. [13]

    Cloudflare radar year in review 2025

    Cloudflare. Cloudflare radar year in review 2025. Interactive report, 2025. https://radar.cloudflare.com/ year-in-review/2025. Global traffic patterns through Cloudflare network; category-level trends

    2025

  14. [14]

    Deepseek-v2: A strong, economical, and efficient mixture-of-experts language model.arXiv preprint arXiv:2405.04434, 2024

    DeepSeek-AI, Aixin Liu, Bei Feng, Bin Wang, Bingxuan Wang, Bo Liu, Chenggang Zhao, Chengqi Deng, Chong Ruan, Damai Dai, Daya Guo, et al. Deepseek-v2: A strong, economical, and efficient mixture-of-experts language model.arXiv preprint arXiv:2405.04434, 2024

    Pith/arXiv arXiv 2024

  15. [15]

    Tsd-sr: One-step diffusion with target score distillation for real-world image super-resolution

    Linwei Dong, Qingnan Fan, Yihong Guo, Zhonghao Wang, Qi Zhang, Jinwei Chen, Yawei Luo, and Changqing Zou. Tsd-sr: One-step diffusion with target score distillation for real-world image super-resolution. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 23174–23184, 2025

    2025

  16. [16]

    Mobile network traffic update (q3 2025)

    Ericsson. Mobile network traffic update (q3 2025). Technical report, Ericsson, 2025.https://www.ericsson.com/ en/reports-and-papers/mobility-report/dataforecasts/mobile-traffic-update . Video traffic expected to account for approximately 75% of all mobile data traffic by end of 2025

    2025

  17. [17]

    Scaling rectified flow transformers for high-resolution image synthesis

    Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas Müller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, et al. Scaling rectified flow transformers for high-resolution image synthesis. InForty-first international conference on machine learning, 2024

    2024

  18. [18]

    Gemini: A family of highly capable multimodal models.arXiv preprint arXiv:2312.11805, 2023

    Gemini Team. Gemini: A family of highly capable multimodal models.arXiv preprint arXiv:2312.11805, 2023. 17 LPM: Industrial-Scale Generative Video Restoration

    Pith/arXiv arXiv 2023

  19. [19]

    Animatediff: Animate your personalized text-to-image diffusion models without specific tuning

    Yuwei Guo, Ceyuan Yang, Anyi Rao, Zhengyang Liang, Yaohui Wang, Yu Qiao, Maneesh Agrawala, Dahua Lin, and Bo Dai. Animatediff: Animate your personalized text-to-image diffusion models without specific tuning. In International Conference on Learning Representations (ICLR), 2024

    2024

  20. [20]

    Gaussian error linear units (gelus).arXiv preprint arXiv:1606.08415, 2016

    Dan Hendrycks and Kevin Gimpel. Gaussian error linear units (gelus).arXiv preprint arXiv:1606.08415, 2016

    Pith/arXiv arXiv 2016

  21. [21]

    Denoising diffusion probabilistic models.Advances in Neural Information Processing Systems, 33:6840–6851, 2020

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models.Advances in Neural Information Processing Systems, 33:6840–6851, 2020

    2020

  22. [22]

    Musiq: Multi-scale image quality transformer

    Junjie Ke, Qifei Wang, Yilin Wang, Peyman Milanfar, and Feng Yang. Musiq: Multi-scale image quality transformer. InProceedings of the IEEE/CVF international conference on computer vision, pages 5148–5157, 2021

    2021

  23. [23]

    Auto-encoding variational bayes

    Diederik P Kingma and Max Welling. Auto-encoding variational bayes. InInternational Conference on Learning Representations (ICLR), 2014

    2014

  24. [24]

    Hunyuanvideo: A systematic framework for large video generative models.arXiv preprint arXiv:2412.03603, 2024

    Weijie Kong, Qi Tian, Zijian Zhang, Rox Min, Zuozhuo Dai, Jin Zhou, Jiangfeng Xiong, Xin Li, Bo Wu, Jianwei Zhang, et al. Hunyuanvideo: A systematic framework for large video generative models.arXiv preprint arXiv:2412.03603, 2024

    Pith/arXiv arXiv 2024

  25. [25]

    Truncated consistency models

    Sangyun Lee, Yilun Xu, Tomas Geffner, Giulia Fanti, Karsten Kreis, Arash Vahdat, and Weili Nie. Truncated consistency models. InThe Thirteenth International Conference on Learning Representations, 2025

    2025

  26. [26]

    Srdiff: Single image super-resolution with diffusion probabilistic models.Neurocomputing, 479:47–59, 2022

    Haoying Li, Yifan Yang, Meng Chang, Shiqi Chen, Huajun Feng, Zhihai Xu, Qi Li, and Yueting Chen. Srdiff: Single image super-resolution with diffusion probabilistic models.Neurocomputing, 479:47–59, 2022

    2022

  27. [27]

    Diffvsr: Revealing an effective recipe for taming robust video super-resolution against complex degradations

    Xiaohui Li, Yihao Liu, Shuo Cao, Ziyan Chen, Shaobin Zhuang, Xiangyu Chen, Yinan He, Yi Wang, and Yu Qiao. Diffvsr: Revealing an effective recipe for taming robust video super-resolution against complex degradations. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 15319–15328, 2025

    2025

  28. [28]

    Tasr: Timestep- aware diffusion model for image super-resolution

    Qinwei Lin, Xiaopeng Sun, Yu Gao, Yujie Zhong, Zheng Zhao, Dengjie Li, and Haoqian Wang. Tasr: Timestep- aware diffusion model for image super-resolution. InProceedings of the 33rd ACM International Conference on Multimedia, pages 10034–10043, 2025

    2025

  29. [29]

    Diffbir: Toward blind image restoration with generative diffusion prior

    Xinqi Lin, Jingwen He, Ziyan Chen, Zhaoyang Lyu, Bo Dai, Fanghua Yu, Yu Qiao, Wanli Ouyang, and Chao Dong. Diffbir: Toward blind image restoration with generative diffusion prior. InEuropean conference on computer vision, pages 430–448. Springer, 2024

    2024

  30. [30]

    Flow straight and fast: Learning to generate and transfer data with rectified flow

    Xingchao Liu, Chengyue Gong, and Qiang Liu. Flow straight and fast: Learning to generate and transfer data with rectified flow. InThe Eleventh International Conference on Learning Representations (ICLR), 2023

    2023

  31. [31]

    Kvq: Kwai video quality assessment for short-form videos

    Yiting Lu, Xin Li, Yajing Pei, Kun Yuan, Qizhi Xie, Yunpeng Qu, Ming Sun, Chao Zhou, and Zhibo Chen. Kvq: Kwai video quality assessment for short-form videos. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 25963–25973, 2024

    2024

  32. [32]

    Latent consistency models: Synthesizing high-resolution images with few-step inference.arXiv preprint arXiv:2310.04378, 2023

    Simian Luo, Yiqin Tan, Longbo Huang, Jian Li, and Hang Zhao. Latent consistency models: Synthesizing high-resolution images with few-step inference.arXiv preprint arXiv:2310.04378, 2023

    Pith/arXiv arXiv 2023

  33. [33]

    Oapt: Offset-aware partition transformer for double jpeg artifacts removal

    Qiao Mo, Yukang Ding, Jinhua Hao, Qiang Zhu, Ming Sun, Chao Zhou, Feiyu Chen, and Shuyuan Zhu. Oapt: Offset-aware partition transformer for double jpeg artifacts removal. InEuropean Conference on Computer Vision, pages 38–56. Springer, 2024

    2024

  34. [34]

    Ntire 2019 challenge on video deblurring and super-resolution: Dataset and study

    Seungjun Nah, Sungyong Baik, Seokil Hong, Gyeongsik Moon, Sanghyun Son, Radu Timofte, and Kyoung Mu Lee. Ntire 2019 challenge on video deblurring and super-resolution: Dataset and study. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition workshops, pages 0–0, 2019

    2019

  35. [35]

    TensorRT-LLM: Large language model inference optimization library.https://github.com/NVIDIA/ TensorRT-LLM, 2023

    NVIDIA. TensorRT-LLM: Large language model inference optimization library.https://github.com/NVIDIA/ TensorRT-LLM, 2023

    2023

  36. [36]

    Scalable diffusion models with transformers

    William Peebles and Saining Xie. Scalable diffusion models with transformers. InProceedings of the IEEE/CVF international conference on computer vision, pages 4195–4205, 2023

    2023

  37. [37]

    Sdxl: Improving latent diffusion models for high-resolution image synthesis

    Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas Müller, Joe Penna, and Robin Rombach. Sdxl: Improving latent diffusion models for high-resolution image synthesis. InInternational Conference on Learning Representations (ICLR), 2024

    2024

  38. [38]

    Xpsr: Cross-modal priors for diffusion-based image super-resolution

    Yunpeng Qu, Kun Yuan, Kai Zhao, Qizhi Xie, Jinhua Hao, Ming Sun, and Chao Zhou. Xpsr: Cross-modal priors for diffusion-based image super-resolution. InEuropean Conference on Computer Vision, pages 285–303. Springer, 2024. 18 LPM: Industrial-Scale Generative Video Restoration

    2024

  39. [39]

    Visual autoregressive modeling for image super-resolution

    Yunpeng Qu, Kun Yuan, Jinhua Hao, Kai Zhao, Qizhi Xie, Ming Sun, and Chao Zhou. Visual autoregressive modeling for image super-resolution. InInternational Conference on Machine Learning, 2025. PMLR, vol. 267, pp. 50926–50948

    2025

  40. [40]

    Kvq: boosting video quality assessment via saliency-guided local perception

    Yunpeng Qu, Kun Yuan, Qizhi Xie, Ming Sun, Chao Zhou, and Jian Wang. Kvq: boosting video quality assessment via saliency-guided local perception. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 2150–2160, 2025

    2025

  41. [41]

    Cosmos tokenizer: A suite of image and video neural tokenizers, 2024

    Fitsum Reda, Jinwei Gu, Xian Liu, Songwei Ge, Ting-Chun Wang, Haoxiang Wang, and Ming-Yu Liu. Cosmos tokenizer: A suite of image and video neural tokenizers, 2024

    2024

  42. [42]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 10684–10695, June 2022

    2022

  43. [43]

    Glu variants improve transformer.arXiv preprint arXiv:2002.05202, 2020

    Noam Shazeer. Glu variants improve transformer.arXiv preprint arXiv:2002.05202, 2020

    Pith/arXiv arXiv 2002

  44. [44]

    Self-supervised controlnet with spatio-temporal mamba for real-world video super-resolution

    Shijun Shi, Jing Xu, Lijing Lu, Zhihang Li, and Kai Hu. Self-supervised controlnet with spatio-temporal mamba for real-world video super-resolution. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 7385–7395, 2025

    2025

  45. [45]

    Consistency models

    Yang Song, Prafulla Dhariwal, Mark Chen, and Ilya Sutskever. Consistency models. InInternational Conference on Machine Learning, pages 32211–32252. PMLR, 2023

    2023

  46. [46]

    Roformer: Enhanced transformer with rotary position embedding.Neurocomputing, 568:127063, 2024

    Jianlin Su, Murtadha Ahmed, Yu Lu, Shengfeng Pan, Wen Bo, and Yunfeng Liu. Roformer: Enhanced transformer with rotary position embedding.Neurocomputing, 568:127063, 2024

    2024

  47. [47]

    Improving the stability and efficiency of diffusion models for content consistent super-resolution.IEEE Transactions on Image Processing, 34:8421–8434, 2025

    Lingchen Sun, Rongyuan Wu, Jie Liang, Zhengqiang Zhang, Hongwei Yong, and Lei Zhang. Improving the stability and efficiency of diffusion models for content consistent super-resolution.IEEE Transactions on Image Processing, 34:8421–8434, 2025

    2025

  48. [48]

    Pixel-level and semantic- level adjustable super-resolution: A dual-lora approach

    Lingchen Sun, Rongyuan Wu, Zhiyuan Ma, Shuaizheng Liu, Qiaosi Yi, and Lei Zhang. Pixel-level and semantic- level adjustable super-resolution: A dual-lora approach. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 2333–2343, 2025

    2025

  49. [49]

    Audio/video enhancement access — tencent cloud media processing service (mps), 2025.https: //cloud.tencent.com/document/product/862/118703

    Tencent Cloud. Audio/video enhancement access — tencent cloud media processing service (mps), 2025.https: //cloud.tencent.com/document/product/862/118703. Accessed: October 1, 2025

    2025

  50. [50]

    Video/image quality enhancement — volcengine media processing documentation, 2025.https: //www.volcengine.com/docs/6448/70279?lang=zh

    Volcengine. Video/image quality enhancement — volcengine media processing documentation, 2025.https: //www.volcengine.com/docs/6448/70279?lang=zh. Accessed: October 1, 2025

    2025

  51. [51]

    Wan: Open and advanced large-scale video generative models.arXiv preprint arXiv:2503.20314, 2025

    Team Wan, Ang Wang, Baole Ai, Bin Wen, Chaojie Mao, Chen-Wei Xie, Di Chen, Feiwu Yu, Haiming Zhao, Jianx- iao Yang, et al. Wan: Open and advanced large-scale video generative models.arXiv preprint arXiv:2503.20314, 2025

    Pith/arXiv arXiv 2025

  52. [52]

    Coloring the noise: Adversarial sobolev alignment for faithful image super resolution.arXiv preprint arXiv:2605.23264, 2026

    Hongbo Wang, Huaibo Huang, Pin Wang, Jinhua Hao, Chao Zhou, and Ran He. Coloring the noise: Adversarial sobolev alignment for faithful image super resolution.arXiv preprint arXiv:2605.23264, 2026. Accepted to ICML 2026

    Pith/arXiv arXiv 2026

  53. [53]

    Exploring clip for assessing the look and feel of images

    Jianyi Wang, Kelvin CK Chan, and Chen Change Loy. Exploring clip for assessing the look and feel of images. In Proceedings of the AAAI conference on artificial intelligence, volume 37, pages 2555–2563, 2023

    2023

  54. [54]

    Exploiting diffusion prior for real-world image super-resolution.International Journal of Computer Vision, 132(12):5929–5949, 2024

    Jianyi Wang, Zongsheng Yue, Shangchen Zhou, Kelvin CK Chan, and Chen Change Loy. Exploiting diffusion prior for real-world image super-resolution.International Journal of Computer Vision, 132(12):5929–5949, 2024

    2024

  55. [55]

    Seedvr2: One-step video restoration via diffusion adversarial post-training.arXiv preprint arXiv:2506.05301, 2025

    Jianyi Wang, Shanchuan Lin, Zhijie Lin, Yuxi Ren, Meng Wei, Zongsheng Yue, Shangchen Zhou, Hao Chen, Yang Zhao, Ceyuan Yang, et al. Seedvr2: One-step video restoration via diffusion adversarial post-training.arXiv preprint arXiv:2506.05301, 2025

    arXiv 2025

  56. [56]

    Seedvr: Seeding infinity in diffusion transformer towards generic video restoration

    Jianyi Wang, Zhijie Lin, Meng Wei, Yang Zhao, Ceyuan Yang, Chen Change Loy, and Lu Jiang. Seedvr: Seeding infinity in diffusion transformer towards generic video restoration. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2161–2172, 2025

    2025

  57. [57]

    Youtube ugc dataset for video compression research

    Yilin Wang, Sasi Inguva, and Balu Adsumilli. Youtube ugc dataset for video compression research. In2019 IEEE 21st international workshop on multimedia signal processing (MMSP), pages 1–5. IEEE, 2019

    2019

  58. [58]

    Image quality assessment: from error visibility to structural similarity.IEEE transactions on image processing, 13(4):600–612, 2004

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Simoncelli. Image quality assessment: from error visibility to structural similarity.IEEE transactions on image processing, 13(4):600–612, 2004. 19 LPM: Industrial-Scale Generative Video Restoration

    2004

  59. [59]

    Seesr: Towards semantics- aware real-world image super-resolution

    Rongyuan Wu, Tao Yang, Lingchen Sun, Zhengqiang Zhang, Shuai Li, and Lei Zhang. Seesr: Towards semantics- aware real-world image super-resolution. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 25456–25467, 2024

    2024

  60. [60]

    Star: Spatial-temporal augmentation with text-to-video models for real-world video super- resolution

    Rui Xie, Yinhong Liu, Penghao Zhou, Chen Zhao, Jun Zhou, Kai Zhang, Zhenyu Zhang, Jian Yang, Zhenheng Yang, and Ying Tai. Star: Spatial-temporal augmentation with text-to-video models for real-world video super- resolution. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 17108–17118, 2025

    2025

  61. [61]

    Videovae+: Large motion video autoencoding with cross-modal video vae

    Yazhou Xing, Yang Fei, Yingqing He, Jingye Chen, Jiaxin Xie, Xiaowei Chi, and Qifeng Chen. Videovae+: Large motion video autoencoding with cross-modal video vae. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 17951–17960, 2025

    2025

  62. [62]

    Pixel-aware stable diffusion for realistic image super-resolution and personalized stylization

    Tao Yang, Rongyuan Wu, Peiran Ren, Xuansong Xie, and Lei Zhang. Pixel-aware stable diffusion for realistic image super-resolution and personalized stylization. InECCV, pages 74–91. Springer, 2024

    2024

  63. [63]

    Motion-guided latent diffusion for temporally consistent real-world video super-resolution

    Xi Yang, Chenhang He, Jianqi Ma, and Lei Zhang. Motion-guided latent diffusion for temporally consistent real-world video super-resolution. InEuropean conference on computer vision, pages 224–242. Springer, 2024

    2024

  64. [64]

    Progressive fusion video super-resolution network via exploiting non-local spatio-temporal correlations

    Peng Yi, Zhongyuan Wang, Kui Jiang, Junjun Jiang, and Jiayi Ma. Progressive fusion video super-resolution network via exploiting non-local spatio-temporal correlations. InProceedings of the IEEE/CVF international conference on computer vision, pages 3106–3115, 2019

    2019

  65. [65]

    Scaling up to excellence: Practicing model scaling for photo-realistic image restoration in the wild

    Fanghua Yu, Jinjin Gu, Zheyuan Li, Jinfan Hu, Xiangtao Kong, Xintao Wang, Jingwen He, Yu Qiao, and Chao Dong. Scaling up to excellence: Practicing model scaling for photo-realistic image restoration in the wild. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 25669–25680, 2024

    2024

  66. [66]

    Zipir: Latent pyramid diffusion transformer for high-resolution image restoration

    Yongsheng Yu, Haitian Zheng, Zhifei Zhang, Jianming Zhang, Yuqian Zhou, Connelly Barnes, Yuchen Liu, Wei Xiong, Zhe Lin, and Jiebo Luo. Zipir: Latent pyramid diffusion transformer for high-resolution image restoration. arXiv preprint arXiv:2504.08591, 2025

    Pith/arXiv arXiv 2025

  67. [67]

    Ptm-vqa: efficient video quality assessment leveraging diverse pretrained models from the wild

    Kun Yuan, Hongbo Liu, Mading Li, Muyi Sun, Ming Sun, Jiachao Gong, Jinhua Hao, Chao Zhou, and Yansong Tang. Ptm-vqa: efficient video quality assessment leveraging diverse pretrained models from the wild. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2835–2845, 2024

    2024

  68. [68]

    Resshift: Efficient diffusion model for image super-resolution by residual shifting.Advances in Neural Information Processing Systems, 36:13294–13307, 2023

    Zongsheng Yue, Jianyi Wang, and Chen Change Loy. Resshift: Efficient diffusion model for image super-resolution by residual shifting.Advances in Neural Information Processing Systems, 36:13294–13307, 2023

    2023

  69. [69]

    Arbitrary-steps image super-resolution via diffusion inversion

    Zongsheng Yue, Kang Liao, and Chen Change Loy. Arbitrary-steps image super-resolution via diffusion inversion. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 23153–23163, 2025

    2025

  70. [70]

    Shiftlut: Spatial shift enhanced look-up tables for efficient image restoration

    Xiaolong Zeng, Yitong Yu, Shiyao Xiong, Jinhua Hao, Ming Sun, Chao Zhou, and Bin Wang. Shiftlut: Spatial shift enhanced look-up tables for efficient image restoration. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 29959–29968, 2026

    2026

  71. [71]

    Degradation-guided one-step image super-resolution with diffusion priors.arXiv preprint arXiv:2409.17058, 2024

    Aiping Zhang, Zongsheng Yue, Renjing Pei, Wenqi Ren, and Xiaochun Cao. Degradation-guided one-step image super-resolution with diffusion priors.arXiv preprint arXiv:2409.17058, 2024

    Pith/arXiv arXiv 2024

  72. [72]

    Root mean square layer normalization.Advances in neural information processing systems, 32, 2019

    Biao Zhang and Rico Sennrich. Root mean square layer normalization.Advances in neural information processing systems, 32, 2019

    2019

  73. [73]

    Sageattention: Accurate 8-bit attention for plug-and-play inference acceleration

    Jintao Zhang, Jia Wei, Haofeng Huang, Pengle Zhang, Jun Zhu, and Jianfei Chen. Sageattention: Accurate 8-bit attention for plug-and-play inference acceleration. InInternational Conference on Learning Representations (ICLR), 2025

    2025

  74. [74]

    Adding conditional control to text-to-image diffusion models

    Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models. InProceedings of the IEEE/CVF international conference on computer vision, pages 3836–3847, 2023

    2023

  75. [75]

    Open-sora 2.0: Training a commercial-level video generation model in $200 k

    Zangwei Zheng, Xiangyu Peng, Yuxuan Lou, Chenhui Shen, Tom Young, Xinying Guo, Binluo Wang, Hang Xu, Hongxin Liu, Mingyan Jiang, et al. Open-sora 2.0: Training a commercial-level video generation model in $200 k. arXiv preprint arXiv:2503.09642, 2025

    Pith/arXiv arXiv 2025

  76. [76]

    Upscale-a-video: Temporal- consistent diffusion model for real-world video super-resolution

    Shangchen Zhou, Peiqing Yang, Jianyi Wang, Yihang Luo, and Chen Change Loy. Upscale-a-video: Temporal- consistent diffusion model for real-world video super-resolution. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2535–2545, 2024. 20 LPM: Industrial-Scale Generative Video Restoration

    2024

  77. [77]

    Cpga: Coding priors-guided aggregation network for compressed video quality enhancement

    Qiang Zhu, Jinhua Hao, Yukang Ding, Yu Liu, Qiao Mo, Ming Sun, Chao Zhou, and Shuyuan Zhu. Cpga: Coding priors-guided aggregation network for compressed video quality enhancement. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2964–2974, 2024

    2024

  78. [78]

    Flashvsr: Towards real-time diffusion-based streaming video super-resolution.arXiv preprint arXiv:2510.12747, 2025

    Junhao Zhuang, Shi Guo, Xin Cai, Xiaohui Li, Yihao Liu, Chun Yuan, and Tianfan Xue. Flashvsr: Towards real-time diffusion-based streaming video super-resolution.arXiv preprint arXiv:2510.12747, 2025. 21

    arXiv 2025