Pith. sign in

REVIEW 2 major objections 5 minor 53 references

EquiEdit improves and balances temporal consistency with editability in text-guided video editing while staying faithful to the input video.

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 · grok-4.5

2026-07-11 09:27 UTC pith:YC2TGDEL

load-bearing objection Solid one-shot T2I-to-video editing paper that pairs a fused temporal Mamba scan with Fourier-guided noise injection and reports consistent gains on LOVEU-TGVE; the balance claim is real on the given data but rests on hand-tuned scalars and a five-person study. the 2 major comments →

arxiv 2607.05056 v1 pith:YC2TGDEL submitted 2026-07-06 cs.CV

Consistent and Editable: A Balanced Framework for Text-Guided Video Editing

Tao Jin , Li Xiao This is my paper
classification cs.CV
keywords text-guided video editingdiffusion modelstemporal consistencyeditabilityMambastate space modelsFourier noise injectionone-shot fine-tuning
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.

Text-guided video editing with diffusion models usually forces a trade-off: stronger temporal consistency across frames tends to weaken how freely the video can be rewritten by a new text prompt, and the reverse. EquiEdit claims both can be raised together. It fine-tunes a pretrained image diffusion model in one shot, adding a temporal Mamba module that scans a fused visual-temporal sequence along four spatial-first directions so frames stay coherent, and a Fourier-based noise injection that carefully adds editable noise while locking high-frequency structure from the original video. The result is edited videos that look natural, match the prompt more closely, and remain faithful to the input motion and detail. A sympathetic reader cares because short-form video is the dominant medium and a method that no longer forces users to choose between smoothness and creative control would make everyday text-driven editing practical.

Core claim

The paper establishes that a single one-shot framework, EquiEdit, can coordinatively raise both temporal consistency and text-driven editability of diffusion video edits, rather than trading one for the other, while preserving fidelity to the input video. The two mechanisms are a temporal Mamba module with temporal-aware scanning of fused sequences and a spectral noise-injection strategy that protects high-frequency structure in the initial latent noise.

What carries the argument

Temporal Mamba module with temporal-aware scanning (four designed spatial-first directions over a fused visual-temporal token sequence) plus Fourier-based noise injection that selectively preserves high-frequency structure from the inverted latent while adding attention-guided Gaussian noise.

Load-bearing premise

The four hand-designed scanning directions and the fixed Fourier-mask and noise-weight settings work well enough to generalize beyond the 76-clip evaluation set and the small user study.

What would settle it

Re-run the same CLIP frame-consistency and text-alignment metrics plus the user preference votes on a larger, held-out video set (different domains, longer clips, or different resolutions) with the same fixed hyperparameters; if EquiEdit no longer simultaneously leads both consistency and editability scores, the central balance claim fails.

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

If this is right

  • One-shot fine-tuning of an image diffusion backbone can produce temporally coherent video edits without training a full text-to-video model from scratch.
  • Mamba-style state-space scanning can replace or complement temporal attention for long video sequences at lower memory cost.
  • Spectral preservation of high-frequency structure in the inverted latent reduces flicker and blur while still allowing prompt-driven change.
  • Users can obtain edits that are both more faithful to input motion and better aligned with new text prompts than prior one-shot or training-free baselines.
  • The same two modules can be ablated independently, showing each contributes to the reported balance.

Where Pith is reading between the lines

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

  • If the fused-sequence scanning truly enlarges the effective receptive field of Mamba, the same design may transfer to other long-sequence video tasks such as generation or prediction without quadratic attention cost.
  • The attention-guided spectral injection suggests a general recipe for controlled noise addition in any latent diffusion editing pipeline where fidelity and flexibility must be traded carefully.
  • Because the method is one-shot and parameter-light, it could be adapted to personal or edge-device video editing once the scanning and mask hyperparameters are made adaptive rather than fixed.

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

2 major / 5 minor

Summary. The paper proposes EquiEdit, a one-shot fine-tuning framework for text-guided video editing built on a pre-trained latent diffusion T2I model. It claims to coordinatively improve temporal consistency and editability (normally a trade-off) while preserving fidelity to the input video. Consistency is addressed by a temporal Mamba module that constructs a fused sequence of visual tokens Xs and temporally pooled tokens Xt and scans it along four hand-designed spatial-first directions. Editability is addressed by a Fourier-based noise injection strategy (Eqs. 1–5) that injects Gaussian noise into the inverted latent while spectrally preserving high-frequency structure from the input. Experiments on 76 LOVEU-TGVE clips report higher CLIP frame-consistency and text-alignment scores than five baselines, a five-participant preference study, an ablation of the scanning design (Tab. 3), component ablations (Fig. 6), and a γ-sensitivity check (Fig. 8).

Significance. If the reported gains hold under broader evaluation, EquiEdit would be a useful practical contribution: it is the first application of Mamba-style SSMs to one-shot video editing, offers a lightweight alternative to quadratic temporal attention for full-sequence modeling, and supplies a parameter-free (at inference) spectral noise-injection path that simultaneously boosts editability and structure preservation. The empirical package—CLIP scores, targeted ablations of the fused scanning and of the two modules, and qualitative comparisons—is coherent and points in the same direction. The work is incremental rather than foundational, but the combination of temporal Mamba with structure-preserving noise injection is a concrete, reproducible design that other one-shot editors can build on.

major comments (2)
  1. Implementation Details and Tabs. 1–2: all quantitative claims rest on a single 76-clip LOVEU-TGVE subset and a five-participant preference study with no error bars, confidence intervals, or statistical tests. The free parameters (c=0.2, r_thr=1, α=0.3, γ=0.5) and the four scanning directions are fixed by hand; without multi-seed variance or a held-out split, it is impossible to judge whether the reported 0.2 CLIP consistency gain over FLATTEN and the user-vote margins generalize. A minimal addition of standard deviations across seeds or a second dataset would make the central “balance” claim load-bearing rather than suggestive.
  2. §Temporal Mamba Module and Fig. 4: the four spatial-first directions and the fused Xs/Xt construction are presented as the key design that overcomes Mamba’s long-context forgetting. Tab. 3 shows only a modest gain (95.607 → 95.863) when the fusion is removed. The paper does not compare against a standard bidirectional or multi-directional Mamba scan of Xs alone, nor against a temporal-attention baseline of comparable parameter count. Without that control, the claim that the specific fused four-direction design is responsible for the consistency improvement remains under-supported.
minor comments (5)
  1. Eq. (2): the mask definition uses r < r_thr with r_thr=1; the units of the radius (normalized frequency? pixel frequency?) are never stated, making the threshold hard to reproduce.
  2. Fig. 3 caption and pipeline: “attention guidance” arrows appear twice with no equation linking them to A_down / A_mid; a short clarification would help.
  3. User-study protocol (Evaluation): five participants and forced-choice voting rates are reported without inter-rater agreement or total number of pairwise comparisons; a sentence on the protocol would strengthen the preference numbers.
  4. Related Work: several concurrent or closely related frequency-aware and Mamba-video works (e.g., FRAG, DNI, VideoMamba) are cited but not discussed comparatively; a short paragraph situating the spectral mask relative to FRAG would be useful.
  5. Typographical: “V ote” in Tabs. 1–2, “finetuning” vs “fine-tuning” inconsistency, and occasional missing spaces after periods.

Circularity Check

0 steps flagged

No significant circularity: empirical architecture and fixed-hyperparameter modules evaluated on external CLIP/user metrics.

full rationale

EquiEdit is a standard one-shot fine-tuning method paper. The temporal Mamba module (four hand-designed spatial-first scans of the fused X_s/X_t sequence) and the Fourier noise-injection path (Eqs. 1–5 with fixed scalars c=0.2, r_thr=1, α=0.3, γ=0.5) are architectural choices, not quantities defined in terms of the later CLIP frame-consistency or text-alignment scores. Those scores and the five-participant preference votes are external benchmarks; ablations (Tab. 3, Fig. 6, Fig. 8) simply remove or vary the modules and re-measure the same external metrics. No equation reduces a claimed prediction to a fitted free parameter by construction, no uniqueness theorem is imported from the authors’ prior work, and no self-citation is load-bearing for the central claim. The derivation chain is therefore self-contained against the reported experiments.

Axiom & Free-Parameter Ledger

4 free parameters · 3 axioms · 2 invented entities

The central performance claim rests on a small set of hand-chosen scalars, standard diffusion and SSM assumptions, and two newly defined algorithmic entities (the fused temporal-aware scan and the spectral noise injector). No new physical entities are postulated; the free parameters are the usual hyper-parameters of a one-shot video editor.

free parameters (4)
  • γ (noise-injection strength) = 0.5
    Controls the amount of additional editable information mixed into the inverted latent; set to 0.5 after informal sweep (Eq. 5, Implementation Details).
  • c (Fourier low-frequency scale) = 0.2
    Scales low-frequency components inside the spectral mask; set to 0.2 (Eq. 2).
  • r_thr (frequency threshold) = 1
    Radius cutoff for the Fourier mask; set to 1 (Eq. 2).
  • α (attention-map mix) = 0.3
    Balances down-block vs mid-block cross-attention maps for noise guidance; set to 0.3 (Eq. 4).
axioms (3)
  • domain assumption A pre-trained latent diffusion T2I model (Stable Diffusion v1.4) already contains sufficient spatial knowledge that can be adapted to video by light fine-tuning of attention projections and a temporal module.
    Stated in Methodology and Implementation Details; inherited from Tune-A-Video lineage.
  • domain assumption Selective state-space models (Mamba) can model long video sequences more efficiently than full temporal attention while still capturing global temporal dependencies when scanned in multiple directions.
    Invoked in the Temporal Mamba Module section and Related Work.
  • domain assumption High-frequency leakage during denoising is a primary cause of flicker and blur; preserving high-frequency structure in the initial latent therefore improves consistency and fidelity.
    Cited from prior FRAG work and used to justify the Fourier mask (Noise Injection Strategy).
invented entities (2)
  • temporal Mamba module with temporal-aware fused scanning no independent evidence
    purpose: Inject global temporal context into each visual token by scanning a concatenated visual+pooled-temporal sequence in four spatial-first directions.
    Defined in §Temporal Mamba Module and Fig. 4; no independent external validation outside this paper’s ablations.
  • Fourier-structure-preserving noise injection (NIS) no independent evidence
    purpose: Add controllable editability to the inverted latent while retaining high-frequency structure of the source video via 3-D FFT masking and attention-guided mixing.
    Defined by Eqs. 1–5; evaluated only inside the paper’s own experiments.

pith-pipeline@v1.1.0-grok45 · 16446 in / 2781 out tokens · 26904 ms · 2026-07-11T09:27:38.062065+00:00 · methodology

0 comments
read the original abstract

Recently, diffusion models have achieved considerable success in the text-guided video editing domain. However, existing works often struggle to balance the trade-off between temporal consistency and editability in video editing, with consistency and editability typically being inversely related. To address this, we propose a high-quality video editing framework enhanced for consistency and editability, named EquiEdit, which improves coordinatively the temporal consistency and editability of the edited videos while achieving a balance between the two. In terms of temporal consistency, the proposed temporal Mamba module with a tailored temporal-aware scanning scans fused video sequences following four designed directions, effectively enhancing the inter-frame consistency of edited videos. For editability, we design a noise injection strategy based on the spectral transformation to increase editing flexibility, where the Fourier transform is used to preserve the hidden structure in the initial latent noise used for editing, ensuring inter-frame consistency of the edited video and fidelity to the input video. Extensive qualitative and quantitative experiments demonstrate the effectiveness of our method in terms of temporal consistency and editability, as well as its great fidelity to the input video itself.

Figures

Figures reproduced from arXiv: 2607.05056 by Li Xiao, Tao Jin.

Figure 1
Figure 1. Figure 1: Illustration of temporal inconsistency and infidelity [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: EquiEdit can improve and balance temporal consistency and editability, ensuring seamless and natural edits. [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: An illustration of EquiEdit. Given a text-video pair, our method leverages a pre-trained T2I model for video editing. During fine-tuning, we only update the projection matrices in the attention blocks and the parameters in the temporal Mamba module. During inference, we introduce Gaussian noise into the initial latent noise which is the output of inversion, while combining spectral transformations to prese… view at source ↗
Figure 4
Figure 4. Figure 4: Visualization of the four spatial-first scanning directions (a) and temporal-aware scanning (b) in the temporal Mamba [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative comparison between evaluated methods. EquiEdit demonstrates a balance between temporal consistency and [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Ablation study of EquiEdit. tency and text alignment. Finally, we analyze the effect of temporal-aware scanning. Frame Consistency. We calculate per-frame global CLIP cosine similarities and report the mean of these values for frame consistency evaluation. We also conduct a user study to evaluate user preferences regarding frame consistency. Participants are asked to select the edited videos with the best … view at source ↗
Figure 7
Figure 7. Figure 7: Visualization of the corresponding denoising process. [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: The impact of additional editable information. As [PITH_FULL_IMAGE:figures/full_fig_p007_8.png] view at source ↗

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

53 extracted references · 21 linked inside Pith

  1. [1]

    Avrahami, O.; Fried, O.; and Lischinski, D. 2023. Blended latent diffusion.ACM transactions on graph- ics (TOG), 42(4): 1–11

    2023

  2. [2]

    Bain, M.; Nagrani, A.; Varol, G.; and Zisserman, A

  3. [3]

    InProceedings of the IEEE/CVF international conference on computer vi- sion, 1728–1738

    Frozen in time: A joint video and image en- coder for end-to-end retrieval. InProceedings of the IEEE/CVF international conference on computer vi- sion, 1728–1738

  4. [4]

    Chai, W.; Guo, X.; Wang, G.; and Lu, Y . 2023. Stable- video: Text-driven consistency-aware diffusion video editing. InProceedings of the IEEE/CVF International Conference on Computer Vision, 23040–23050

    2023

  5. [5]

    Chen, W.; Ji, Y .; Wu, J.; Wu, H.; Xie, P.; Li, J.; Xia, X.; Xiao, X.; and Lin, L. 2023. Control-a-video: Control- lable text-to-video generation with diffusion models. arXiv preprint arXiv:2305.13840

    Pith/arXiv arXiv 2023

  6. [6]

    Cohen, N.; Kulikov, V .; Kleiner, M.; Huberman- Spiegelglas, I.; and Michaeli, T. 2024. Slicedit: Zero-Shot Video Editing With Text-to-Image Diffu- sion Models Using Spatio-Temporal Slices.arXiv preprint arXiv:2405.12211

    Pith/arXiv arXiv 2024

  7. [7]

    Cong, Y .; Xu, M.; Simon, C.; Chen, S.; Ren, J.; Xie, Y .; Perez-Rua, J.-M.; Rosenhahn, B.; Xiang, T.; and He, S. 2023. Flatten: optical flow-guided attention for consistent text-to-video editing.arXiv preprint arXiv:2310.05922

    Pith/arXiv arXiv 2023

  8. [8]

    Feng, R.; Weng, W.; Wang, Y .; Yuan, Y .; Bao, J.; Luo, C.; Chen, Z.; and Guo, B. 2024. Ccedit: Creative and controllable video editing via diffusion models. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 6712–6722

    2024

  9. [9]

    H.; Chechik, G.; and Cohen-Or, D

    Gal, R.; Alaluf, Y .; Atzmon, Y .; Patashnik, O.; Bermano, A. H.; Chechik, G.; and Cohen-Or, D. 2022. An image is worth one word: Personalizing text- to-image generation using textual inversion.arXiv preprint arXiv:2208.01618

    Pith/arXiv arXiv 2022

  10. [10]

    Geyer, M.; Bar-Tal, O.; Bagon, S.; and Dekel, T. 2023. Tokenflow: Consistent diffusion features for consistent video editing.arXiv preprint arXiv:2307.10373

    Pith/arXiv arXiv 2023

  11. [11]

    Gu, A.; and Dao, T. 2023. Mamba: Linear-time se- quence modeling with selective state spaces.arXiv preprint arXiv:2312.00752

    Pith/arXiv arXiv 2023

  12. [12]

    Gu, A.; Goel, K.; and R ´e, C. 2021. Efficiently model- ing long sequences with structured state spaces.arXiv preprint arXiv:2111.00396

    Pith/arXiv arXiv 2021

  13. [13]

    Gu, S.; Chen, D.; Bao, J.; Wen, F.; Zhang, B.; Chen, D.; Yuan, L.; and Guo, B. 2022. Vector quantized dif- fusion model for text-to-image synthesis. InProceed- ings of the IEEE/CVF conference on computer vision and pattern recognition, 10696–10706

    2022

  14. [14]

    J.; Norouzi, M.; and Salimans, T

    Ho, J.; Saharia, C.; Chan, W.; Fleet, D. J.; Norouzi, M.; and Salimans, T. 2022. Cascaded diffusion models for high fidelity image generation.Journal of Machine Learning Research, 23(47): 1–33

    2022

  15. [15]

    Ho, J.; and Salimans, T. 2022. Classifier-free diffusion guidance.arXiv preprint arXiv:2207.12598

    Pith/arXiv arXiv 2022

  16. [16]

    M.; and Yanardag, P

    Kara, O.; Kurtkaya, B.; Yesiltepe, H.; Rehg, J. M.; and Yanardag, P. 2024. Rave: Randomized noise shuf- fling for fast and consistent video editing with diffusion models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 6507– 6516

    2024

  17. [17]

    Kingma, D.; Salimans, T.; Poole, B.; and Ho, J. 2021. Variational diffusion models.Advances in neural in- formation processing systems, 34: 21696–21707

    2021

  18. [18]

    Kumagai, H.; Yamaki, R.; and Naganuma, H. 2023. Story-to-Images Translation: Leveraging Diffusion Models and Large Language Models for Sequence Im- age Generation. InProceedings of the 2nd Workshop on User-centric Narrative Summarization of Long Videos, 57–63

    2023

  19. [19]

    Li, B.; Xue, K.; Liu, B.; and Lai, Y .-K. 2022. Vqbb: Image-to-image translation with vector quantized brownian bridge.arXiv preprint arXiv:2205.07680, 2

    Pith/arXiv arXiv 2022

  20. [20]

    Li, K.; Li, X.; Wang, Y .; He, Y .; Wang, Y .; Wang, L.; and Qiao, Y . 2024. Videomamba: State space model for efficient video understanding.arXiv preprint arXiv:2403.06977

    Pith/arXiv arXiv 2024

  21. [21]

    Ling, P.; Bu, J.; Zhang, P.; Dong, X.; Zang, Y .; Wu, T.; Chen, H.; Wang, J.; and Jin, Y . 2024. MotionClone: Training-Free Motion Cloning for Controllable Video Generation.arXiv preprint arXiv:2406.05338

    Pith/arXiv arXiv 2024

  22. [22]

    Liu, S.; Zhang, Y .; Li, W.; Lin, Z.; and Jia, J. 2024. Video-p2p: Video editing with cross-attention control. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, 8599–8608

    2024

  23. [23]

    H.; Azadi, S.; Zhang, G.; Chopikyan, A.; Hu, Y .; Shi, H.; Rohrbach, A.; and Darrell, T

    Liu, X.; Park, D. H.; Azadi, S.; Zhang, G.; Chopikyan, A.; Hu, Y .; Shi, H.; Rohrbach, A.; and Darrell, T. 2023. More control for free! image synthesis with semantic diffusion guidance. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vi- sion, 289–299

    2023

  24. [24]

    Mao, Q.; Chen, L.; Gu, Y .; Fang, Z.; and Shou, M. Z

  25. [25]

    InProceedings of the 32nd ACM International Confer- ence on Multimedia, 6842–6850

    Mag-edit: Localized image editing in complex scenarios via mask-based attention-adjusted guidance. InProceedings of the 32nd ACM International Confer- ence on Multimedia, 6842–6850

  26. [26]

    Meng, C.; He, Y .; Song, Y .; Song, J.; Wu, J.; Zhu, J.- Y .; and Ermon, S. 2021. Sdedit: Guided image synthe- sis and editing with stochastic differential equations. arXiv preprint arXiv:2108.01073

    Pith/arXiv arXiv 2021

  27. [27]

    Mou, C.; Wang, X.; Xie, L.; Wu, Y .; Zhang, J.; Qi, Z.; and Shan, Y . 2024. T2i-adapter: Learning adapters to dig out more controllable ability for text-to-image dif- fusion models. InProceedings of the AAAI Conference on Artificial Intelligence, volume 38, 4296–4304

    2024

  28. [28]

    Parmar, G.; Kumar Singh, K.; Zhang, R.; Li, Y .; Lu, J.; and Zhu, J.-Y . 2023. Zero-shot image-to-image trans- lation. InACM SIGGRAPH 2023 conference proceed- ings, 1–11

    2023

  29. [29]

    Perazzi, F.; Pont-Tuset, J.; McWilliams, B.; Van Gool, L.; Gross, M.; and Sorkine-Hornung, A. 2016. A benchmark dataset and evaluation methodology for video object segmentation. InProceedings of the IEEE conference on computer vision and pattern recogni- tion, 724–732

    2016

  30. [30]

    W.; Hallacy, C.; Ramesh, A.; Goh, G.; Agarwal, S.; Sastry, G.; Askell, A.; Mishkin, P.; Clark, J.; et al

    Radford, A.; Kim, J. W.; Hallacy, C.; Ramesh, A.; Goh, G.; Agarwal, S.; Sastry, G.; Askell, A.; Mishkin, P.; Clark, J.; et al. 2021. Learning transferable visual mod- els from natural language supervision. InInternational conference on machine learning, 8748–8763. PMLR

    2021

  31. [31]

    Rombach, R.; Blattmann, A.; Lorenz, D.; Esser, P.; and Ommer, B. 2022. High-resolution image synthe- sis with latent diffusion models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 10684–10695

    2022

  32. [32]

    Ruiz, N.; Li, Y .; Jampani, V .; Pritch, Y .; Rubinstein, M.; and Aberman, K. 2023. Dreambooth: Fine tuning text-to-image diffusion models for subject-driven gen- eration. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 22500– 22510

    2023

  33. [33]

    L.; Ghasemipour, K.; Gontijo Lopes, R.; Karagol Ayan, B.; Salimans, T.; et al

    Saharia, C.; Chan, W.; Saxena, S.; Li, L.; Whang, J.; Denton, E. L.; Ghasemipour, K.; Gontijo Lopes, R.; Karagol Ayan, B.; Salimans, T.; et al. 2022. Photo- realistic text-to-image diffusion models with deep lan- guage understanding.Advances in neural information processing systems, 35: 36479–36494

    2022

  34. [34]

    J.; and Norouzi, M

    Saharia, C.; Ho, J.; Chan, W.; Salimans, T.; Fleet, D. J.; and Norouzi, M. 2022. Image super-resolution via it- erative refinement.IEEE transactions on pattern anal- ysis and machine intelligence, 45(4): 4713–4726

    2022

  35. [35]

    Song, J.; Meng, C.; and Ermon, S. 2020. De- noising diffusion implicit models.arXiv preprint arXiv:2010.02502

    Pith/arXiv arXiv 2020

  36. [36]

    Wang, J.; Yuan, H.; Chen, D.; Zhang, Y .; Wang, X.; and Zhang, S. 2023. Modelscope text-to-video technical report.arXiv preprint arXiv:2308.06571

    Pith/arXiv arXiv 2023

  37. [37]

    Wang, W.; Jiang, Y .; Xie, K.; Liu, Z.; Chen, H.; Cao, Y .; Wang, X.; and Shen, C. 2023. Zero-shot video edit- ing using off-the-shelf image diffusion models.arXiv preprint arXiv:2303.17599

    Pith/arXiv arXiv 2023

  38. [38]

    B.; and Cui, Z

    Wang, Y .; Li, Y .; Zhang, X.; Liu, X.; Dai, A.; Chan, A. B.; and Cui, Z. 2023. Edit Temporal-Consistent Videos with Image Diffusion Model.arXiv preprint arXiv:2308.09091

    Pith/arXiv arXiv 2023

  39. [39]

    B.; and Cui, Z

    Wang, Y .; Li, Y .; Zhang, X.; Liu, X.; Dai, A.; Chan, A. B.; and Cui, Z. 2024. Edit temporal-consistent videos with image diffusion model.ACM Transactions on Multimedia Computing, Communications and Ap- plications, 20(12): 1–16

    2024

  40. [40]

    Wu, H.; He, L.; Zhang, M.; Chen, D.; Luo, K.; Luo, M.; Zhou, J.-Z.; Chen, H.; and Lv, J. 2024. Diffusion Posterior Proximal Sampling for Image Restoration. In Proceedings of the 32nd ACM International Confer- ence on Multimedia, 214–223

    2024

  41. [41]

    Z.; Ge, Y .; Wang, X.; Lei, S

    Wu, J. Z.; Ge, Y .; Wang, X.; Lei, S. W.; Gu, Y .; Shi, Y .; Hsu, W.; Shan, Y .; Qie, X.; and Shou, M. Z

  42. [42]

    InPro- ceedings of the IEEE/CVF International Conference on Computer Vision, 7623–7633

    Tune-a-video: One-shot tuning of image dif- fusion models for text-to-video generation. InPro- ceedings of the IEEE/CVF International Conference on Computer Vision, 7623–7633

  43. [43]

    Z.; Li, X.; Gao, D.; Dong, Z.; Bai, J.; Singh, A.; Xiang, X.; Li, Y .; Huang, Z.; Sun, Y .; et al

    Wu, J. Z.; Li, X.; Gao, D.; Dong, Z.; Bai, J.; Singh, A.; Xiang, X.; Li, Y .; Huang, Z.; Sun, Y .; et al. 2023. Cvpr 2023 text guided video editing competition.arXiv preprint arXiv:2310.16003

    Pith/arXiv arXiv 2023

  44. [44]

    Xing, Z.; Dai, Q.; Hu, H.; Wu, Z.; and Jiang, Y .-G

  45. [45]

    InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion, 7827–7839

    Simda: Simple diffusion adapter for efficient video generation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion, 7827–7839

  46. [46]

    Xing, Z.; Feng, Q.; Chen, H.; Dai, Q.; Hu, H.; Xu, H.; Wu, Z.; and Jiang, Y .-G. 2023. A survey on video dif- fusion models.ACM Computing Surveys

    2023

  47. [47]

    Ye, Z.; Xia, K.; Fu, Y .; Dong, X.; Hong, J.; Yuan, X.; Diao, S.; Kautz, J.; Molchanov, P.; and Lin, Y . C. 2025. LongMamba: Enhancing Mamba’s Long Context Ca- pabilities via Training-Free Receptive Field Enlarge- ment.arXiv preprint arXiv:2504.16053

    Pith/arXiv arXiv 2025

  48. [48]

    W.; and Yoo, C

    Yoon, S.; Koo, G.; Hong, J. W.; and Yoo, C. D

  49. [49]

    DNI: Dilutional Noise Initialization for Diffu- sion Video Editing.arXiv preprint arXiv:2409.13037

    Pith/arXiv arXiv

  50. [50]

    Yoon, S.; Koo, G.; Kim, G.; and Yoo, C. D. 2024. FRAG: Frequency Adapting Group for Diffusion Video Editing.arXiv preprint arXiv:2406.06044

    Pith/arXiv arXiv 2024

  51. [51]

    Zhang, G.; Zhang, T.; Niu, G.; Tan, Z.; Bai, Y .; and Yang, Q. 2024. CAMEL: CAusal Motion Enhance- ment tailored for Lifting Text-driven Video Editing. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, 9079–9088

    2024

  52. [52]

    Zhang, Z.; Li, B.; Nie, X.; Han, C.; Guo, T.; and Liu, L. 2024. Towards consistent video editing with text-to- image diffusion models.Advances in Neural Informa- tion Processing Systems, 36

    2024

  53. [53]

    Zhu, L.; Liao, B.; Zhang, Q.; Wang, X.; Liu, W.; and Wang, X. 2024. Vision mamba: Efficient visual repre- sentation learning with bidirectional state space model. arXiv preprint arXiv:2401.09417

    Pith/arXiv arXiv 2024