Pith. sign in

REVIEW 3 major objections 5 minor 44 references

DiffuseSlide: Training-Free High Frame Rate Video Generation Diffusion

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read DiffuseSlide generates high-frame-rate video from a single image without retraining, using latent interpolation, noise re-injection, and sliding-window denoising.

desk verdict A plausible training-free video frame-rate upsampling method, but the SOTA claim rests on an uncontrolled cross-paper comparison and a non-standard metric protocol. read the letter →

arxiv 2506.01454 v1 pith:BO3PJA2F submitted 2025-06-02 cs.CV

Geunmin Hwang , Hyun-kyu Ko , Younghyun Kim , Seungryong Lee , Eunbyung Park This is my paper
classification cs.CV
keywords highframe-ratevideogenerationtraining-freediffusionpipelineframeinterpolationnoisere-injectionslidingwindowdenoisinglatenttemporalconsistencyWebVid-10M
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper proposes DiffuseSlide, a training-free pipeline that turns a short low-frame-rate video produced by a pre-trained image-to-video diffusion model into a smooth high-frame-rate video. It does this by linearly interpolating the generated latents in time, adding controlled noise, and then denoising with a sliding window so that every subsequence is conditioned on its own keyframe rather than only the first frame. The authors report that this removes flicker and ghosting artifacts while preserving the keyframes, and that on a 2,048-clip subset of WebVid-10M the method achieves lower Fréchet Video Distance than direct inference, linear interpolation, and the reported ZeroSmooth numbers, with the highest PSNR and SSIM among the training-free baselines. If correct, any existing image-to-video diffusion model could produce 2x or 4x frame rates without retraining or architecture changes.

What carries the argument

The machinery is a two-stage latent-space pipeline built on a pre-trained latent video diffusion model. First, a low-frame-rate video is generated and linearly interpolated in latent space to the target frame count. Then the sequence is denoised with noise re-injection and sliding-window denoising: at each timestep, the latent is split into overlapping subsequences, each subsequence is denoised while conditioned on its matching keyframe, and the overlaps are fused; while the timestep is above a threshold, the denoise step is repeated M times with a small Gaussian of variance $\sigma^2_\tau - \sigma^2_{\tau-1}$ re-injected each time. The sliding window uses a keyframe set $K$, window size $w$, and stride $s$ so that every frame sits near a conditioning keyframe.

What would settle it

A decisive check is to run DiffuseSlide and ZeroSmooth on the same 2,048 WebVid-10M clips with the same FVD script; if the FVD gap narrows to noise or reverses, the reported advantage is an artifact of subset choice.

Watch

Extended reading notes

Core claim

The central discovery is that a pre-trained image-to-video diffusion model can be made to generate temporally consistent high-frame-rate video without any training by combining three operations: latent linear interpolation between keyframes, noise injection followed by repeated noise re-injection during denoising, and sliding-window denoising in which each window is conditioned on its own keyframe. The interpolated latents act as a structural prior; the noise re-injection gives the denoiser repeated chances to push the sequence back onto the video data manifold; and the sliding window prevents later frames from drifting because conditioning is refreshed at each window. On Stable Video Diffusion, the paper reports FVD of 584.63 at 2x and 636.41 at 4x, beating direct inference, linear interpolation, and ZeroSmooth as reported, and matching or beating the training-based LDMVFI on FVD, with the highest PSNR and SSIM among the training-free baselines. The method is also demonstrated for video-to-video enhancement.

Load-bearing premise

The claim that DiffuseSlide beats ZeroSmooth assumes that FVD numbers from different 2,048-video subsets of WebVid-10M are directly comparable, because the exact ZeroSmooth evaluation clips are not public.

Editorial extensions

If this is right

  • Any pre-trained image-to-video diffusion model can be adapted to 2x or 4x frame rates without retraining, as long as it can generate short keyframe sequences.
  • Sliding-window denoising prevents late-frame degradation in long sequences by refreshing the conditioning keyframe at every window instead of relying only on the first frame.
  • Noise re-injection reduces flicker and ghosting artifacts left by simple latent linear interpolation, improving temporal coherence in fast-motion scenes.
  • At 4x expansion, DiffuseSlide becomes computationally competitive with the training-based LDMVFI while reporting a lower FVD, making it practical for resource-constrained settings.
  • The same pipeline applies to video-to-video enhancement, enabling longer low-frame-rate input videos to be upsampled under GPU memory limits.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper, the sliding-window conditioning scheme suggests the pipeline should transfer to other image-to-video backbones, including text-conditioned models, though only SVD and I2VGen-XL are tested.
  • Beyond the paper, the reported 273-second inference time at 2x points to sampler acceleration as the natural next step; the paper itself lists slow diffusion inference as a limitation.
  • Beyond the paper, a direct test of the method's ceiling would be to shrink the window stride or increase the number of interpolated frames per window, since only 2x and 4x expansions are demonstrated.
  • Beyond the paper, because PSNR and SSIM are computed against keyframes and cannot penalize pure linear interpolation, a human or perceptual study on fast-motion clips would be a stronger test of the claimed smoothness.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper presents DiffuseSlide, a training-free pipeline for increasing the frame rate of videos produced by pre-trained image-to-video diffusion models. Given a low frame-rate latent generated from a single image, the method linearly interpolates additional frames in latent space, then applies a denoising procedure that alternates standard denoising steps with controlled noise re-injection. To handle sequences longer than the base model's capacity, the latent is denoised in overlapping windows, each conditioned on a different keyframe. The authors report FVD, PSNR, and SSIM for 2x and 4x frame-rate expansion using Stable Video Diffusion and I2VGen-XL on a 2,048-video subset of WebVid-10M, and claim state-of-the-art results, including an FVD improvement over ZeroSmooth and Linear Interpolation.

Significance. If the evaluation were properly controlled, the paper would make a useful contribution: it offers a simple, architecture-agnostic, training-free recipe for high frame-rate video generation, with clear pseudocode, ablations of the two main components, and qualitative results on both image-to-video and video-to-video tasks. The sliding-window multi-keyframe conditioning idea is a reasonable response to the known problem of degrading conditioning in long sequences. However, the headline SOTA claim currently rests on a non-controlled comparison with ZeroSmooth, and the reported FVD and PSNR/SSIM numbers lack the reproducibility details needed to verify them. As presented, the evidence primarily supports the more modest conclusion that DiffuseSlide improves over linear interpolation and direct inference under the paper's own protocol.

major comments (3)
  1. [Section 4.2 and Table 1] The central SOTA claim is not supported by a controlled comparison. The paper states in Section 4.2 that it sampled a new 2,048-clip subset of WebVid-10M because the specific ZeroSmooth instances are not publicly available, and the ZeroSmooth row in Table 1 uses values taken directly from the ZeroSmooth paper. FVD is well known to be sensitive to the sampled clips and random seed, so the reported gap (584.63 vs 779.6 at 2x) cannot be attributed to method quality. The authors should either evaluate ZeroSmooth on the same subset when code becomes available, report a proper matched-subset comparison for all baselines, or restrict the claim to the baselines actually evaluated under the same protocol.
  2. [Section 4.2 and Appendix C] The FVD evaluation protocol is not reproducible as described. Appendix C says that 2,048 WebVid videos are compared with generated videos, but WebVid-10M clips have heterogeneous resolutions, frame counts, and frame rates; the paper does not specify how original videos are resized or cropped to the generated resolutions (576x1024 or 704x1280), how many frames are used per clip, or how temporal lengths are matched. Without these details, and without reporting seeds or confidence intervals, the FVD differences in Table 1 (e.g., 636.41 vs 752.1 at 4x) cannot be independently checked or considered reliable.
  3. [Section 4.2 and Table 1] The PSNR/SSIM comparison is not apples-to-apples. For DiffuseSlide, PSNR/SSIM measure agreement between the generated low-fps keyframes and the high-fps output, which by construction favors a method that preserves keyframes; the paper itself notes these metrics 'do not directly assess the quality' and that Linear Interpolation is excluded from them. The ZeroSmooth values in Table 1 were computed under a standard protocol (similarity to ground-truth frames), so the table's claim that DiffuseSlide achieves the 'highest' PSNR and SSIM is not meaningful. The authors should recompute all methods under one protocol or relabel PSNR/SSIM as an internal keyframe-preservation measure.
minor comments (5)
  1. [Section 3.3, Eq. (2)] In the VE formulation, the noisy latent should be z_tau = z_0 + sigma_tau * epsilon, not z_tau = z_0 + sigma_tau^2 * epsilon; the current expression gives the added perturbation a variance of sigma_tau^4, inconsistent with the noise schedule used in Algorithm 1.
  2. [Section 3.4 and Algorithm 1] The prose says noise re-injection is applied 'afterward' (once), while Algorithm 1 reinjects noise after each denoising step within a loop of M iterations; please align the text with the algorithm.
  3. [Appendix A, Eq. (4)] The symbol z_t is used both for noisy latents in Section 3 and for frame indices in Appendix A; please use a different notation (e.g., f_t or x_t) to avoid confusion.
  4. [Section 4.2] Please specify exactly which frames are compared when computing PSNR/SSIM between the low-fps keyframes and the high-fps output; in the 4x setting each keyframe is followed by three inserted frames, so it is unclear whether all inserted frames, only the aligned keyframes, or some other selection is used.
  5. [General] There are several typos and stylistic issues: 'levarging' in Section 3.2, 'unnatrual' in Section 2, and inconsistent use of accents in 'Frechet' in Appendix C; these should be corrected.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation: FVD is externally grounded, the only self-citation is a non-load-bearing noise-injection reference, and the PSNR/SSIM concerns are transparent metric-alignment issues rather than circular steps.

full rationale

The paper's central quantitative claim rests on FVD computed against original WebVid-10M videos on a self-sampled 2,048-clip subset. FVD is an external distributional metric, and no DiffuseSlide parameter is fitted to it; the pipeline uses fixed hyperparameters (tau, delta, M in Appendix B) and pre-trained SVD/I2VGen-XL backbones, so the FVD result is not a fitted quantity renamed as a prediction. The only self-citation is [18] (DiffuseHigh), used as one of three citations for an SDEdit-style noise-addition step; it is not load-bearing because the sliding-window and noise-re-injection mechanics are introduced in this paper and are not justified by an author-specific uniqueness theorem. The PSNR/SSIM protocol explicitly measures keyframe preservation, with Section 4.2 stating that these metrics 'do not directly assess the quality of the generated high frame-rate video, but rather the changes introduced during the high frame-rate video generation,' and the method's keyframe-conditioned sliding window is designed to maintain keyframe fidelity. This makes the PSNR/SSIM comparison partly self-aligned, but the paper is transparent about that role, and the FVD claim is independent of it. The ZeroSmooth comparison is a genuine external-validity threat, because DiffuseSlide and ZeroSmooth were evaluated on different, non-public WebVid-10M subsets, but that is a data-comparability problem rather than circularity of derivation. Overall, no derivation step reduces to its inputs by construction; the score reflects only the minor, non-load-bearing self-citation.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim depends on several hand-set hyperparameters (tau, delta, M, window size/stride, motion bucket) and on assumptions about the pre-trained model's latent space and the validity of latent interpolation as initialization. No new physical or conceptual entities are introduced.

free parameters (5)
  • noising step tau = 8 (SVD), 15 (I2VGen-XL)
    Chosen by hand per backbone to control how much structure is disrupted before denoising; not derived from data or theory.
  • noise re-injection step delta = 3 (SVD), 4 (I2VGen-XL)
    Hand-set; controls how many denoising steps use the noise-reinjection loop.
  • reinjection iterations M = 5 for both backbones
    M=5 was selected over M=7 after an ablation (Table 4) because PSNR/SSIM favored it, even though FVD slightly favored M=7.
  • sliding window size w and stride s = not reported
    Algorithm 2 takes w and s as inputs, but the paper never states the values used in the experiments, so the configuration is underspecified.
  • SVD motion bucket ID = 128 for keyframe generation; 64 (2x) and 32 (4x) for expansion
    The motion bucket is reduced for high frame-rate expansion, an extra task-specific adjustment that is tuned per setting.
assumptions (4)
  • domain assumption The pre-trained image-to-video model (SVD or I2VGen-XL) produces good low-frame-rate keyframes.
    The whole pipeline starts from these keyframes, and Section 5 acknowledges that the method inherits the limits of the base model.
  • domain assumption Linear interpolation in the latent space is a valid initialization for high-frame-rate latent video.
    Used to produce z0 in Eq. (1); the paper does not justify why averaging latents is a good starting point beyond the observation that it yields ghosting artifacts.
  • ad hoc to paper The alternating denoise and noise re-injection procedure guides the latent toward the smooth video manifold.
    Stated in Section 3.4 and Figure 2 as the mechanism for artifact removal; no proof or quantitative support beyond the ablation.
  • ad hoc to paper Averaging overlapping windows in the sliding-window scheme produces a coherent global video latent.
    Algorithm 2 uses Fuse(Z) to average overlapping frames, but the paper does not analyze whether this averaging preserves temporal consistency.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DiffuseSlide: Training-Free High Frame Rate Video Generation Diffusion." pith.science (2026). https://pith.science/paper/BO3PJA2F

@misc{pith2026250601454,
  author       = {Pith},
  title        = {Pith review of: DiffuseSlide: Training-Free High Frame Rate Video Generation Diffusion},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BO3PJA2F}},
  note         = {Machine review of arXiv:2506.01454}
}
read the original abstract

Recent advancements in diffusion models have revolutionized video generation, enabling the creation of high-quality, temporally consistent videos. However, generating high frame-rate (FPS) videos remains a significant challenge due to issues such as flickering and degradation in long sequences, particularly in fast-motion scenarios. Existing methods often suffer from computational inefficiencies and limitations in maintaining video quality over extended frames. In this paper, we present a novel, training-free approach for high FPS video generation using pre-trained diffusion models. Our method, DiffuseSlide, introduces a new pipeline that leverages key frames from low FPS videos and applies innovative techniques, including noise re-injection and sliding window latent denoising, to achieve smooth, consistent video outputs without the need for additional fine-tuning. Through extensive experiments, we demonstrate that our approach significantly improves video quality, offering enhanced temporal coherence and spatial fidelity. The proposed method is not only computationally efficient but also adaptable to various video generation tasks, making it ideal for applications such as virtual reality, video games, and high-quality content creation.

Figures

Figures reproduced from arXiv: 2506.01454 by the authors.

Figure 1
Figure 1. Overall pipeline of DiffuseSlide. Given an input image, a low frame-rate video is generated using a pre-trained image-to-video model. To increase the frame rate, linear interpolation is applied, followed by noise injection and denoising to remove artifacts. Our method employs multi-step noise reinjection to better align with the data manifold. Our sliding window-denoise approach conditions on multiple frames to bett… view at source ↗
Figure 2
Figure 2. Noise re-injection adds flexibility to manifold con [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Direct denoising of the full sequence latent may result [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Qualitative results showcasing high frame-rate video generation. The images illustrate the smooth transitions and artifact reduction achieved with DiffuseSlide in 4x frame-rate experiments. “L.I” refers to Linear Interpolation. Model Method 2× 4× FVD(↓) PSNR(↑) SSIM(↑)…
Figure 5
Figure 5. Figure 5: Ablation study results showing the impact of sliding window denoising. The quality improvement across frames demonstrates the effectiveness of these components in preserving temporal consistency and reducing degradation in later frames. “S.W” refers to Sliding Window d…
Figure 6
Figure 6. Figure 6: Ablation study results showing the effect of Noise [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Detailed qualitative results for video-to-video genera￾tion. This figure highlights the detailed improvements in smooth￾ness and artifact reduction achieved by DiffuseSlide in 4× frame￾rate experiments. E. Video-to-Video Generation DiffuseSlide is a versatile framework…
Figure 8
Figure 8. Figure 8: Detailed improvements in high frame-rate video gen￾eration. The figure showcases the effectiveness of DiffuseSlide in reducing artifacts and enhancing smooth transitions for 4× frame￾rate generation [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 9
Figure 9. Figure 9: Overall qualitative results for high frame-rate video generation. This figure demonstrates the smooth transitions and reduced artifacts achieved with DiffuseSlide in 4× frame-rate scenarios. “L.I” refers to Linear Interpolation. For a closer look at specific details, s…
Figure 10
Figure 10. Figure 10: Overall qualitative results for video-to-video generation. The figure illustrates the enhanced temporal consistency and artifact reduction achieved with DiffuseSlide in 4× frame-rate video-to-video scenarios. “L.I” refers to Linear Interpolation. See [PITH_FULL_IMAGE…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

44 extracted references · 15 canonical work pages

  1. [1]

    Frozen in time: A joint video and image encoder for end-to-end retrieval

    Max Bain, Arsha Nagrani, G ¨ul Varol, and Andrew Zisser- man. Frozen in time: A joint video and image encoder for end-to-end retrieval. In Proceedings of the IEEE/CVF Inter- national Conference on Computer Vision, pages 1728–1738,

  2. [2]

    Depth-aware video frame interpolation

    Wenbo Bao, Wei-Sheng Lai, Chao Ma, Xiaoyun Zhang, Zhiyong Gao, and Ming-Hsuan Yang. Depth-aware video frame interpolation. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition , pages 3703–3712, 2019. 1

    work page 2019

  3. [3]

    Improving image generation with better captions

    James Betker, Gabriel Goh, Li Jing, Tim Brooks, Jianfeng Wang, Linjie Li, Long Ouyang, Juntang Zhuang, Joyce Lee, Yufei Guo, et al. Improving image generation with better captions. Computer Science. https://cdn. openai. com/papers/dall-e-3. pdf, 2(3):8, 2023. 2

    2023

  4. [4]

    Stable video diffusion: Scaling latent video diffusion models to large datasets

    Andreas Blattmann, Tim Dockhorn, Sumith Kulal, Daniel Mendelevitch, Maciej Kilian, Dominik Lorenz, Yam Levi, Zion English, Vikram V oleti, Adam Letts, et al. Stable video diffusion: Scaling latent video diffusion models to large datasets. arXiv preprint arXiv:2311.15127, 2023. 2, 3, 5

    arXiv 2023

  5. [5]

    Align your latents: High-resolution video synthesis with la- tent diffusion models

    Andreas Blattmann, Robin Rombach, Huan Ling, Tim Dock- horn, Seung Wook Kim, Sanja Fidler, and Karsten Kreis. Align your latents: High-resolution video synthesis with la- tent diffusion models. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 22563–22575, 2023

    work page 2023

  6. [6]

    Videocrafter1: Open diffusion models for high-quality video generation

    Haoxin Chen, Menghan Xia, Yingqing He, Yong Zhang, Xiaodong Cun, Shaoshu Yang, Jinbo Xing, Yaofang Liu, Qifeng Chen, Xintao Wang, et al. Videocrafter1: Open diffusion models for high-quality video generation. arXiv preprint arXiv:2310.19512, 2023

    arXiv

  7. [7]

    Videocrafter2: Overcoming data limitations for high-quality video diffu- sion models

    Haoxin Chen, Yong Zhang, Xiaodong Cun, Menghan Xia, Xintao Wang, Chao Weng, and Ying Shan. Videocrafter2: Overcoming data limitations for high-quality video diffu- sion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 7310– 7320, 2024. 2

    work page 2024

  8. [8]

    Ldmvfi: Video frame interpolation with latent diffusion models

    Duolikun Danier, Fan Zhang, and David Bull. Ldmvfi: Video frame interpolation with latent diffusion models. InProceed- ings of the AAAI Conference on Artificial Intelligence, pages 1472–1480, 2024. 1, 8

    work page 2024

Show all 44 references
  1. [9]

    Ex- plorative inbetweening of time and space

    Haiwen Feng, Zheng Ding, Zhihao Xia, Simon Niklaus, Vic- toria Abrevaya, Michael J Black, and Xuaner Zhang. Ex- plorative inbetweening of time and space. arXiv preprint arXiv:2403.14611, 2024. 4

    2024 arXiv

  2. [10]

    Make a cheap scaling: A self-cascade diffusion model for higher-resolution adapta- tion

    Lanqing Guo, Yingqing He, Haoxin Chen, Menghan Xia, Xiaodong Cun, Yufei Wang, Siyu Huang, Yong Zhang, Xin- tao Wang, Qifeng Chen, et al. Make a cheap scaling: A self-cascade diffusion model for higher-resolution adapta- tion. In European Conference on Computer Vision , pages 3...

    2025

  3. [11]

    Photorealistic video generation with diffusion models

    Agrim Gupta, Lijun Yu, Kihyuk Sohn, Xiuye Gu, Meera Hahn, Fei-Fei Li, Irfan Essa, Lu Jiang, and Jos ´e Lezama. Photorealistic video generation with diffusion models. In European Conference on Computer Vision, pages 393–411. Springer, 2025. 2

    2025

  4. [12]

    Latent video diffusion models for high-fidelity long video generation

    Yingqing He, Tianyu Yang, Yong Zhang, Ying Shan, and Qifeng Chen. Latent video diffusion models for high-fidelity long video generation. arXiv preprint arXiv:2211.13221 ,

    arXiv

  5. [13]

    Imagen video: High definition video generation with diffusion mod- els

    Jonathan Ho, William Chan, Chitwan Saharia, Jay Whang, Ruiqi Gao, Alexey Gritsenko, Diederik P Kingma, Ben Poole, Mohammad Norouzi, David J Fleet, et al. Imagen video: High definition video generation with diffusion mod- els. arXiv preprint arXiv:2210.02303, 2022. 2

    arXiv

  6. [14]

    Video dif- fusion models

    Jonathan Ho, Tim Salimans, Alexey Gritsenko, William Chan, Mohammad Norouzi, and David J Fleet. Video dif- fusion models. Advances in Neural Information Processing Systems, 35:8633–8646, 2022. 2

    2022

  7. [15]

    Cogvideo: Large-scale pretraining for text-to-video generation via transformers

    Wenyi Hong, Ming Ding, Wendi Zheng, Xinghan Liu, and Jie Tang. Cogvideo: Large-scale pretraining for text-to-video generation via transformers. arXiv preprint arXiv:2205.15868, 2022. 2

    2022 arXiv

  8. [16]

    Video interpolation with diffu- sion models

    Siddhant Jain, Daniel Watson, Eric Tabellion, Ben Poole, Janne Kontkanen, et al. Video interpolation with diffu- sion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 7341– 7351, 2024. 1

    2024

  9. [17]

    Ocai: Improving optical flow estimation by occlusion and consistency aware interpolation

    Jisoo Jeong, Hong Cai, Risheek Garrepalli, Jamie Menjay Lin, Munawar Hayat, and Fatih Porikli. Ocai: Improving optical flow estimation by occlusion and consistency aware interpolation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages ...

    2024

  10. [18]

    Diffusehigh: Training-free progressive high- resolution image synthesis through structure guidance.arXiv preprint arXiv:2406.18459, 2024

    Younghyun Kim, Geunmin Hwang, Junyu Zhang, and Eun- byung Park. Diffusehigh: Training-free progressive high- resolution image synthesis through structure guidance.arXiv preprint arXiv:2406.18459, 2024. 3

    2024 arXiv

  11. [19]

    Hunyuanvideo: A systematic framework for large video generative models

    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. 2

    2024 arXiv

  12. [20]

    Video frame interpolation with transformer

    Liying Lu, Ruizheng Wu, Huaijia Lin, Jiangbo Lu, and Jiaya Jia. Video frame interpolation with transformer. In Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3532–3542, 2022. 1

    2022

  13. [21]

    Frame interpolation with consecutive brownian bridge diffusion

    Zonglin Lyu, Ming Li, Jianbo Jiao, and Chen Chen. Frame interpolation with consecutive brownian bridge diffusion. In Proceedings of the 32nd ACM International Conference on Multimedia, pages 3449–3458, 2024. 1

    2024

  14. [22]

    Sdedit: Guided image synthesis and editing with stochastic differential equa- tions

    Chenlin Meng, Yutong He, Yang Song, Jiaming Song, Jia- jun Wu, Jun-Yan Zhu, and Stefano Ermon. Sdedit: Guided image synthesis and editing with stochastic differential equa- tions. arXiv preprint arXiv:2108.01073, 2021. 3

    2021 arXiv

  15. [23]

    Softmax splatting for video frame interpolation

    Simon Niklaus and Feng Liu. Softmax splatting for video frame interpolation. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition , pages 5437–5446, 2020. 1

    2020

  16. [24]

    SoRA, 2024

    OpenAI. SoRA, 2024. 1, 2

    2024

  17. [25]

    Sdxl: Improving latent diffusion mod- els for high-resolution image synthesis

    Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas M ¨uller, Joe Penna, and Robin Rombach. Sdxl: Improving latent diffusion mod- els for high-resolution image synthesis. arXiv preprint arXiv:2307.01952, 2023. 2

    2023 arXiv

  18. [26]

    Freenoise: Tuning-free longer video diffusion via noise rescheduling.arXiv preprint arXiv:2310.15169, 2023

    Haonan Qiu, Menghan Xia, Yong Zhang, Yingqing He, Xin- tao Wang, Ying Shan, and Ziwei Liu. Freenoise: Tuning-free longer video diffusion via noise rescheduling.arXiv preprint arXiv:2310.15169, 2023. 2, 5

    2023 arXiv

  19. [27]

    Hierarchical text-conditional image gener- ation with clip latents

    Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. Hierarchical text-conditional image gener- ation with clip latents. arXiv preprint arXiv:2204.06125, 1 (2):3, 2022. 2

    2022 arXiv

  20. [28]

    High-resolution image syn- thesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj¨orn Ommer. High-resolution image syn- thesis with latent diffusion models. In CVPR, pages 10684– 10695, 2022

    2022

  21. [29]

    Photorealistic text-to-image diffusion models with deep language understanding

    Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily L Denton, Kamyar Ghasemipour, Raphael Gontijo Lopes, Burcu Karagol Ayan, Tim Salimans, et al. Photorealistic text-to-image diffusion models with deep language understanding. NeurIPS, 35:36479–36494, 2022. 2

    2022

  22. [30]

    Make-a-video: Text-to-video generation without text-video data

    Uriel Singer, Adam Polyak, Thomas Hayes, Xi Yin, Jie An, Songyang Zhang, Qiyuan Hu, Harry Yang, Oron Ashual, Oran Gafni, et al. Make-a-video: Text-to-video generation without text-video data. arXiv preprint arXiv:2209.14792 ,

    arXiv

  23. [31]

    Score-based generative modeling through stochastic differential equa- tions

    Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Ab- hishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equa- tions. arXiv preprint arXiv:2011.13456, 2020. 3

    2011 arXiv

  24. [32]

    To- wards accurate generative models of video: A new metric & challenges

    Thomas Unterthiner, Sjoerd Van Steenkiste, Karol Kurach, Raphael Marinier, Marcin Michalski, and Sylvain Gelly. To- wards accurate generative models of video: A new metric & challenges. arXiv preprint arXiv:1812.01717, 2018. 5

    2018 arXiv

  25. [33]

    Phenaki: Variable length video generation from open domain textual descriptions

    Ruben Villegas, Mohammad Babaeizadeh, Pieter-Jan Kin- dermans, Hernan Moraldo, Han Zhang, Mohammad Taghi Saffar, Santiago Castro, Julius Kunze, and Dumitru Erhan. Phenaki: Variable length video generation from open domain textual descriptions. In International Conference on Le...

    2022

  26. [34]

    Mcvd-masked conditional video diffusion for prediction, generation, and interpolation

    Vikram V oleti, Alexia Jolicoeur-Martineau, and Chris Pal. Mcvd-masked conditional video diffusion for prediction, generation, and interpolation. Advances in neural informa- tion processing systems, 35:23371–23385, 2022. 1

    2022

  27. [35]

    Modelscope text-to-video technical report

    Jiuniu Wang, Hangjie Yuan, Dayou Chen, Yingya Zhang, Xiang Wang, and Shiwei Zhang. Modelscope text-to-video technical report. arXiv preprint arXiv:2308.06571, 2023. 2

    2023 arXiv

  28. [36]

    Generative inbetweening: Adapting image- to-video models for keyframe interpolation

    Xiaojuan Wang, Boyang Zhou, Brian Curless, Ira Kemelmacher-Shlizerman, Aleksander Holynski, and Steven M Seitz. Generative inbetweening: Adapting image- to-video models for keyframe interpolation. arXiv preprint arXiv:2408.15239, 2024. 4

    2024 arXiv

  29. [37]

    Zero-shot im- age restoration using denoising diffusion null-space model

    Yinhuai Wang, Jiwen Yu, and Jian Zhang. Zero-shot im- age restoration using denoising diffusion null-space model. arXiv preprint arXiv:2212.00490, 2022. 2

    2022 arXiv

  30. [38]

    Lavie: High-quality video gener- ation with cascaded latent diffusion models

    Yaohui Wang, Xinyuan Chen, Xin Ma, Shangchen Zhou, Ziqi Huang, Yi Wang, Ceyuan Yang, Yinan He, Jiashuo Yu, Peiqing Yang, et al. Lavie: High-quality video gener- ation with cascaded latent diffusion models. arXiv preprint arXiv:2309.15103, 2023. 2

    2023 arXiv

  31. [39]

    Dynamicrafter: Animating open-domain images with video diffusion priors

    Jinbo Xing, Menghan Xia, Yong Zhang, Haoxin Chen, Wangbo Yu, Hanyuan Liu, Gongye Liu, Xintao Wang, Ying Shan, and Tien-Tsin Wong. Dynamicrafter: Animating open-domain images with video diffusion priors. In Eu- ropean Conference on Computer Vision , pages 399–417. Springer, 2025. 2

    2025

  32. [40]

    Vibidsam- pler: Enhancing video interpolation using bidirectional dif- fusion sampler

    Serin Yang, Taesung Kwon, and Jong Chul Ye. Vibidsam- pler: Enhancing video interpolation using bidirectional dif- fusion sampler. arXiv preprint arXiv:2410.05651, 2024. 4

    2024 arXiv

  33. [41]

    Zerosmooth: Training-free diffuser adapta- tion for high frame rate video generation

    Shaoshu Yang, Yong Zhang, Xiaodong Cun, Ying Shan, and Ran He. Zerosmooth: Training-free diffuser adapta- tion for high frame rate video generation. arXiv preprint arXiv:2406.00908, 2024. 2, 5

    2024 arXiv

  34. [42]

    Cogvideox: Text-to-video diffusion models with an expert transformer

    Zhuoyi Yang, Jiayan Teng, Wendi Zheng, Ming Ding, Shiyu Huang, Jiazheng Xu, Yuanming Yang, Wenyi Hong, Xiao- han Zhang, Guanyu Feng, et al. Cogvideox: Text-to-video diffusion models with an expert transformer. arXiv preprint arXiv:2408.06072, 2024. 2

    2024 arXiv

  35. [43]

    Extracting motion and ap- pearance via inter-frame attention for efficient video frame interpolation

    Guozhen Zhang, Yuhan Zhu, Haonan Wang, Youxin Chen, Gangshan Wu, and Limin Wang. Extracting motion and ap- pearance via inter-frame attention for efficient video frame interpolation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 5...

    2023

  36. [44]

    I2vgen-xl: High-quality image-to-video synthesis via cascaded diffusion models

    Shiwei Zhang, Jiayu Wang, Yingya Zhang, Kang Zhao, Hangjie Yuan, Zhiwu Qin, Xiang Wang, Deli Zhao, and Jingren Zhou. I2vgen-xl: High-quality image-to-video synthesis via cascaded diffusion models. arXiv preprint arXiv:2311.04145, 2023. 2, 3 A. Linear Interpolation The interpol...

    2023 arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.