Pith. sign in

REVIEW 4 major objections 5 minor 20 references

FlexiClip: Locality-Preserving Free-Form Character Animation

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read FlexiClip claims that clipart animation artifacts come from temporal noise in per-frame deformations, and it removes that noise by adding a learned temporal Jacobian field integrated over time with a probability-flow ODE and trained with…

desk verdict A useful residual-Jacobian trick buried under unsupported pfODE and GFlowNet labels, with a headline claim contradicted by its own geometry metric. read the letter →

arxiv 2501.08676 v2 pith:NFUBLEJ7 submitted 2025-01-15 cs.CV cs.GR

Anant Khandelwal This is my paper · ORCID
classification cs.CVcs.GR
keywords clipartanimationtemporalJacobianprobabilityflowODEmatchinglossscoredistillationsamplingBéziertrajectoriesvideodiffusionpriorsGFlowNet
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

FlexiClip aims to show that the jerky, distorted motion in clipart animation is not an inevitable limit of Bézier-based deformation but a temporal-noise problem that can be corrected. It proposes adding a learned temporal Jacobian to each frame's spatial Jacobian, integrating that correction over time with a probability-flow ODE, and training the whole pipeline with a GFlowNet-inspired flow matching loss that forces corrections to shrink as motion stabilizes. On the AniClipart benchmark it reports higher visual identity (CLIP 0.9563 vs 0.9401), better text-video alignment (X-CLIP 0.2102 vs 0.2075), and smoother deformations (DS 13.76 vs 18.49), while acknowledging slightly higher geometric deviation (GD 52.34 vs 50.98) because it abandons ARAP rigidity. If it holds, the recipe makes text-driven clipart animation both smoother and more faithful to the prompt without requiring natural-video training data.

What carries the argument

The load-bearing object is the temporal Jacobian $J_t^R$, a per-triangle corrective matrix added to the spatial Jacobian from Bézier trajectories. It is paired with two mechanisms: the pfODE reformulation in Eqs. (12)–(13), which integrates the correction over continuous time using attention over past windows in place of a score function, and the flow matching loss $L_{\text{flow}} = \mathbb{E}_{t',t}\|\nabla_X \log p_{t'}(X,J_t) - \nabla_X \log p_{t'}(X,J_t^P)\|^2 + \|J_t - J_t^P\|^2$, which pulls the score of fully corrected frames toward the score of spatial-only frames so that the correction term is driven to zero once temporal noise is removed.

What would settle it

Check whether the learned temporal Jacobian actually converges toward zero for a stationary first frame and only grows where motion occurs, by recording $\|J_t^R\|$ over training and over the 24 frames; and run an ablation that replaces the ODE integrator with an ordinary residual MLP of the same capacity. If the ODE variant does not beat the residual variant on DS and TC, or if the corrections do not shrink as claimed, the pfODE reformulation is not load-bearing.

Watch

Extended reading notes

Core claim

The paper's claim is that temporal coherence in clipart animation can be separated from spatial deformation: each frame's total Jacobian is $J_t = J_t^P + J_t^R$, where $J_t^P$ comes from Bézier keypoint trajectories and $J_t^R$ is a learned correction that starts at zero and evolves under an ODE $\mathrm{d}J_t^R/\mathrm{d}t = f_R(J_0^P, C_W^P, C_{W-1}^R, t; \theta_R)$, integrated with Euler's method. The ODE is presented as a reformulation of the probability-flow ODE in which the usual score function is replaced by attention-encoded windows of spatial and temporal Jacobians, and a flow matching loss enforces that $J_t^R$ cancels only temporal noise while leaving the spatial pose intact. FlexiClip then demonstrates that this design beats AniClipart on identity, alignment, and smoothness metrics and wins a 30-participant user study on those same criteria.

Load-bearing premise

The paper assumes that the neural ODE in Eqs. (12)–(13) is a valid probability-flow ODE reformulation of Eq. (10), so that integrating the learned MLP over time removes temporal noise and drives the correction to zero; if that equivalence fails, the temporal Jacobian is just a residual network and the claimed theoretical grounding collapses.

Editorial extensions

If this is right

  • Text-to-clipart animation can be made smoother by learning per-frame Jacobian corrections rather than predicting each frame's pose independently.
  • The flow matching loss, by comparing score estimates of corrected versus spatial-only frames, can serve as a temporal denoising regularizer for any Jacobian-based deformation model.
  • The pipeline supports rotation, multiple text conditions, multiple interacting objects, and layered animations in a single framework, so the correction mechanism generalizes beyond simple keypoint motions.
  • Dropping ARAP rigidity in favor of learned temporal corrections lowers geometric deviation quality slightly (GD 52.34 vs 50.98) but improves deformation smoothness and motion energy, indicating a deliberate trade-off between rigidity and expressiveness.
  • Because the method needs only one input image and a text prompt, it extends SDS-based animation to clipart without retraining the video diffusion model.

Reading between the lines

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

  • Editorial inference: the same spatial/temporal Jacobian split could apply to natural-image or 3D shape animation, wherever a per-frame deformation predictor produces temporal jitter; the flow matching loss would then act as a universal smoothness regularizer.
  • Editorial inference: if Eqs. (12)–(13) are not a genuine pfODE (no score function appears), then the paper's theoretical framing reduces to adding a residual correction net with a flow-matching regularizer; a strong ablation comparing ODE vs residual integration would settle this.
  • Editorial inference: a testable extension is using a lightweight denoiser instead of the pretrained video diffusion model in Eq. (17), which would reveal how much of the quality gain comes from temporal correction versus SDS guidance.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes FlexiClip, a text-driven clipart animation method that builds on AniClipart's cubic Bézier keypoint trajectories and spatial Jacobian field deformation. Its claimed contributions are (i) temporal Jacobians that incrementally correct spatial Jacobians, (ii) a continuous-time reformulation via probability-flow ODEs (pfODEs), and (iii) a GFlowNet-inspired flow matching loss to reduce temporal noise. The system is trained with video Score Distillation Sampling, and experiments compare FlexiClip against AniClipart and several T2V/I2V models on bitmap metrics, animation metrics, and a user study.

Significance. The idea of adding a learned temporal correction to spatial Jacobians is plausible, and the qualitative results suggest the implemented system can produce appealing animations. However, the paper's central theoretical contributions are not established: the asserted pfODE reformulation in Eqs. (12)-(13) lacks any derivation or connection to Eq. (10), and the 'flow matching loss' is not shown to correspond to GFlowNet detailed balance or to any standard flow matching objective. More seriously, Table 2 shows that FlexiClip has a higher (worse) Geometric Deviation than AniClipart, directly contradicting the 'Locality-Preserving' claim in the title and the abstract's promise of geometric integrity. The quantitative evaluation reports single numbers without variance or significance tests. If the theoretical framing were removed, the remaining technical contribution is an incremental residual correction with L2 regularization.

major comments (4)
  1. [Sec. 3.3 (Eqs. 12-13)] The claim that Eq. (12)-(13) is a reformulation of the pfODE in Eq. (10) is not supported. The pfODE in Eq. (10) is driven by the score function ∇_x log p_t(x) with explicit schedules C(t) and A(t); none of these appear in Eq. (12), whose right-hand side depends only on the base Jacobian, attention features, and time, with no dependence on the current state J^R_t. The text identifies C^P_W with C(t) and C^R_{W-1} with A(t), but these are attention-encoded features rather than a covariance or scale schedule, and no probability density over Jacobian fields is specified. Appendix C derives the Gaussian-kernel pfODE and ends at G G^T = Ċ(t), but never connects this derivation to temporal Jacobians or to Eq. (12). Thus the central 'continuous-time pfODE' contribution is unsubstantiated; Eq. (12)-(13) is essentially an integrated MLP. The statement that the correction 'reduced to zero' is also not a property of Eq. (13); the only mechanism enforcing smallness is the ad-hoc penalty added in Eq. (18).
  2. [Sec. 3.4 (Eqs. 17-18)] The 'flow matching loss' is claimed to implement a detailed-balance objective from GFlowNets, but no mapping to GFlowNet states, transitions, or forward/backward policies is provided. Eq. (17) is an L2 difference of two score-function evaluations, and Eq. (18) appends a penalty on the temporal Jacobian magnitude; there is no derivation showing that this corresponds to detailed balance, to flow matching in the vector-field sense, or to any GFlowNet training objective. The terminology appears to be a post-hoc analogy rather than a formal connection. The authors should either supply the formal derivation or abandon the pfODE/GFlowNet framing.
  3. [Table 2 and Secs. 4.3, 4.5] The title and abstract claim 'locality-preserving' deformation, yet Table 2 reports Geometric Deviation (GD, lower is better) of 52.34 for FlexiClip versus 50.98 for AniClipart. Sec. 4.5 admits this is because FlexiClip lacks ARAP deformation, which 'inherently minimizes shape distortion.' The sentence in Sec. 4.3 that 'GD has increased but DS decreased showing smoother animations' does not resolve the contradiction; a higher GD means the deformation is less rigid and less local, directly undercutting the central claim. This is a load-bearing inconsistency, not a presentation issue.
  4. [Secs. 4.1-4.3 and Table 3] All quantitative results in Tables 1-4 are reported as single numbers with no variance, number of independent runs, or significance tests. Given the small margins on several metrics (e.g., TC 8.14 vs 8.51, X-CLIP 0.2102 vs 0.2075) and the worse GD, the claimed improvements over AniClipart are not established. The user study in Table 3 reports only mean percentages without confidence intervals or statistical tests, so the claim of significant outperformance is unsupported.
minor comments (5)
  1. [Sec. 2.3] The GFlowNet preliminaries are duplicated nearly verbatim in two consecutive paragraphs; please condense to a single description.
  2. [Eq. (6)] The displayed Jacobian matrix for a triangular face has an inconsistent size and layout; since J0,f is said to be in R^{2×2}, the partial-derivative table should be a 2×2 matrix and should be written unambiguously.
  3. [Eq. (19) and Sec. 3.4] The overall loss is described as a weighted sum of 'Eq.14, 16, 17', but Eq. (16) is a definition of the denoising objective, not a loss term; the actual loss terms are Eq. (17) (with the penalty from Eq. (18)) and Eq. (14). Please clarify which equations enter the total loss and define λ consistently with its earlier use in Eq. (2).
  4. [Appendix A] The text refers to 'Fig.6' and 'Fig.7' when the corresponding items are captioned as Table 6 and Table 7; the cross-references should be corrected.
  5. [Sec. 2.2 and References] The pfODE in Eqs. (7)-(9) is attributed to 'Song et al. (Lim et al., 2023)', but the cited work is by Lim et al.; the probability flow ODE was introduced by Song et al. (2021). The citation should be corrected, and the derivation in Appendix C should include the rescaling term A(t) that appears in Eq. (10).

Circularity Check

2 steps flagged · score 6.0 of 10

Temporal-Jacobian 'pfODE reformulation' is a renaming plus an L2 penalty, so the central theoretical claims partially reduce to their own loss terms.

  1. self definitional [Sec. 3.3, Eqs. (12)-(13) and the paragraph following Eq. (13)]
    "Furthermore, we are required to remove only temporal noise from spatial Jacobians and wanted to add only temporal correction term (temporal Jacobian), we propose to reformulate the ODE given in (Eq.10) given as: dJR_t/dt = fR(JP_0, C^P_W, C^R_{W-1}, t; θ_R) ... In comparison to (Eq.10) we model the time-varying noise C(t) with C^P_W and rescaling A(t) with C^R_{W-1}."

    Equation (10) is a pfODE because C(t) is the covariance of injected Gaussian noise, A(t) is a scale schedule, and the right-hand side is driven by the score ∇x log pt(x). Equation (12) has none of these: its right-hand side depends on base Jacobian JP0, attention features C^P_W and C^R_{W-1}, and time, with no score term and no dependence on the current state JR_t. The text's 'C^P_W → C(t)' and 'C^R_{W-1} → rescaling A(t)' define the attention features to be the pfODE schedules, so the claimed reformulation is established by definitional renaming, not by any derivation from Eq. (10) or from the Fokker-Planck equation in App. C. The theoretical grounding of the temporal Jacobian is therefore inherited from the name, not from the math.

  2. fitted input called prediction [Sec. 3.3 convergence claim and Eq. (18) in Sec. 3.4]
    "For convergence, as in (Eq.10) as soon as the corrections applied starts to contract it balance the opposing force from the error in spatial Jacobians and finally the corrections term reduced to zero. ... Moreover, we wanted the parameterized Bézier to capture temporal movement of keypoints as much as possible and hence we want the temporal Jacobian (corrective term) to be as small as possible, hence we add the correction minimizing loss given as: ... + Et‖Jt − JP_t‖2"

    Since Jt = JP_t + JR_t by Eq. (11), the loss term Et‖Jt − JP_t‖2 in Eq. (18) is exactly Et‖JR_t‖2, a penalty that directly drives the temporal Jacobian to zero. The paper earlier states that 'finally the corrections term reduced to zero' as a convergence property of Eq. (10); but the only mechanism in the method that enforces that reduction is this explicit regularizer. Reporting the smallness of the correction as validation of the pfODE-based convergence, when it is minimized by the loss by construction, is a fitted input presented as a predicted outcome.

full rationale

The paper's empirical comparison is self-contained: Tables 1-3 and the ablation studies evaluate against external methods using standard metrics such as CLIP, X-CLIP, MV, TC, GD, DS, and AE, and those numbers are not derived from the model's own assumptions. However, the theoretical contribution is partially circular. In Sec. 3.3, Eq. (12)-(13) is presented as a reformulation of the pfODE Eq. (10), but the identification C^P_W -> C(t) and C^R_{W-1} -> A(t) is a definitional renaming of attention features; no score function or state-dependent noise process appears in Eq. (12), and App. C derives only the standard Gaussian-kernel pfODE, ending at GG^T = C_dot(t), without ever connecting to temporal Jacobians. The convergence statement that 'the corrections term reduced to zero' is then credited to Eq. (10), yet the only mechanism enforcing smallness is the explicit L2 penalty Et‖Jt - JP_t‖2 = Et‖JR_t‖2 added in Eq. (18); the smallness is a fitted outcome, not a predicted property of the ODE. The GFlowNet connection is explicitly inspirational rather than derivational, so it is not scored as circular, though it is also unsubstantiated. Overall, the empirical core is independent, but the headline pfODE and temporal-correction claims reduce in part to renaming and to the loss's own regularizer, giving partial circularity.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central method rests on the standard Poisson deformation solve and on two ad hoc assumptions about the temporal model: that a neural ODE with attention features is a pfODE, and that an L2 score-difference is a GFlowNet detailed balance objective. Both are asserted without derivation, which is the main circularity burden. There are no invented physical entities.

free parameters (3)
  • λ (flow matching loss weight) = 15
    Tuned in Sec 4.4; the paper reports convergence and quality change across λ=1 to >15 and selects 15 for all results.
  • guidance scale s for classifier-free guidance = 50
    Set for SDS loss (Sec 4.1); no study reported.
  • number of Bézier control points = 8-11
    Chosen per input (Sec 4.1); not justified quantitatively.
assumptions (5)
  • standard math The Poisson deformation solve (Eq 1-3) with cotangent Laplacian and keypoint constraints yields the deformed mesh.
    Standard linear FEM/geometry processing result, used in AniClipart and prior work; not re-derived.
  • domain assumption A pre-trained video diffusion model's score function can guide clipart deformation despite domain shift.
    Invoked in Sec 3.4 via Video SDS loss; the paper acknowledges the statistical mismatch between natural video and clipart, so this is an assumption that the score still provides useful gradients.
  • ad hoc to paper The neural ODE (Eq 12-13) implements a probability flow ODE that reverses temporal noise in spatial Jacobians.
    Sec 3.3 states this reformulation without derivation; no score function or Fokker-Planck equation is used in Eq 12-13.
  • ad hoc to paper The loss in Eq 17 is a detailed balance / flow matching objective in the GFlowNet sense.
    Sec 3.4 asserts the connection to GFlowNet DB but Eq 17 is simply an L2 difference of score estimates, with no flow function or policies; the second term minimizes ||Jt - JP_t|| directly.
  • domain assumption Temporal noise in Jacobian fields is Gaussian and can be denoised by the video diffusion model's score.
    Assumed in Sec 2.2 and 3.3 to justify pfODE; no evidence that Jacobian temporal noise follows a known diffusion process.

how reviews work

0 comments
Cite this review

Pith. "Pith review of FlexiClip: Locality-Preserving Free-Form Character Animation." pith.science (2026). https://pith.science/paper/NFUBLEJ7

@misc{pith2026250108676,
  author       = {Pith},
  title        = {Pith review of: FlexiClip: Locality-Preserving Free-Form Character Animation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NFUBLEJ7}},
  note         = {Machine review of arXiv:2501.08676}
}
read the original abstract

Animating clipart images with seamless motion while maintaining visual fidelity and temporal coherence presents significant challenges. Existing methods, such as AniClipart, effectively model spatial deformations but often fail to ensure smooth temporal transitions, resulting in artifacts like abrupt motions and geometric distortions. Similarly, text-to-video (T2V) and image-to-video (I2V) models struggle to handle clipart due to the mismatch in statistical properties between natural video and clipart styles. This paper introduces FlexiClip, a novel approach designed to overcome these limitations by addressing the intertwined challenges of temporal consistency and geometric integrity. FlexiClip extends traditional B\'ezier curve-based trajectory modeling with key innovations: temporal Jacobians to correct motion dynamics incrementally, continuous-time modeling via probability flow ODEs (pfODEs) to mitigate temporal noise, and a flow matching loss inspired by GFlowNet principles to optimize smooth motion transitions. These enhancements ensure coherent animations across complex scenarios involving rapid movements and non-rigid deformations. Extensive experiments validate the effectiveness of FlexiClip in generating animations that are not only smooth and natural but also structurally consistent across diverse clipart types, including humans and animals. By integrating spatial and temporal modeling with pre-trained video diffusion models, FlexiClip sets a new standard for high-quality clipart animation, offering robust performance across a wide range of visual content. Project Page: https://creative-gen.github.io/flexiclip.github.io/

Figures

Figures reproduced from arXiv: 2501.08676 by the authors.

Figure 1
Figure 1. FlexiClip generates high-quality clipart animations based on text prompts, ensuring visual consistency and smooth temporal motion. Original clipart images are outlined with dashed boxes. duces a novel paradigm by learning temporal Jacobians that incrementally correct the spatial Jacobian over time. This framework enables precise control over temporal evolution and prevents drift in the motion dynamics, thereby main￾… view at source ↗
Figure 2
Figure 2. System Design of FlexiClip: A novel framework for generating temporally coherent and geometrically consistent an￾imated clipart. FlexiClip leverages pfODE for continuous-time modeling on top of spatial posing, and a flow matching loss for re￾ducing temporal noise, enabling smooth, natural animations across complex motion sequences. 3. FlexiClip In this section, we introduce our mesh deformation frame￾work, FlexiClip… view at source ↗
Figure 3
Figure 3. FlexiClip vs AniClipart: Four consecutive frames are shown for comparison. AniClipart distorts the objects (e.g., boy/girl jumping, woman dancing), lacks proper text conditioning (e.g., parrot), and shows poor temporal consistency (e.g., boy jumping). In contrast, FlexiClip preserves the visual identity, maintaining consistent shapes and producing high-quality, smooth, and text-aligned animations. Detailed animation… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: FlexiClip vs T2V/I2V Models: LTXVideo and PyramidFlow show moderate performance, with scores higher than methods like DynamiCrafter and AnimateLCM-I2V, but still lagging behind FlexiClip. While these methods manage basic identity preservation, they struggle with animat…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

20 extracted references · 9 canonical work pages

  1. [1]

    J., Tiwari, M., and Bengio, E

    Bengio, Y., Lahlou, S., Deleu, T., Hu, E. J., Tiwari, M., and Bengio, E. Gflownet foundations. The Journal of Machine Learning Research, 24 0 (1): 0 10006--10060, 2023

    2023

  2. [2]

    A cgal implementation of the straight skeleton of a simple 2d polygon with holes

    Cacciola, F. A cgal implementation of the straight skeleton of a simple 2d polygon with holes. In 2nd CGAL user workshop, volume 1. Citeseer, 2004

    work page 2004

  3. [3]

    Videocrafter1: Open diffusion models for high-quality video generation

    Chen, H., Xia, M., He, Y., Zhang, Y., Cun, X., Yang, S., Xing, J., Liu, Y., Chen, Q., Wang, X., et al. Videocrafter1: Open diffusion models for high-quality video generation. arXiv preprint arXiv:2310.19512, 2023

    arXiv 2023

  4. [4]

    Videocrafter2: Overcoming data limitations for high-quality video diffusion models

    Chen, H., Zhang, Y., Cun, X., Xia, M., Wang, X., Weng, C., and Shan, Y. Videocrafter2: Overcoming data limitations for high-quality video diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 7310--7320, 2024

    work page 2024

  5. [5]

    and Pearson, J

    de Albuquerque, D. and Pearson, J. Inflationary flows: Calibrated bayesian inference with diffusion-based models. In Globerson, A., Mackey, L., Belgrave, D., Fan, A., Paquet, U., Tomczak, J., and Zhang, C. (eds.), Advances in Neural Information Processing Systems, volume 37, pp.\ 96847--96891. Curran Associates, Inc., 2024

    work page 2024

  6. [6]

    Breathing life into sketches using text-to-video priors

    Gal, R., Vinker, Y., Alaluf, Y., Bermano, A., Cohen-Or, D., Shamir, A., and Chechik, G. Breathing life into sketches using text-to-video priors. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 4325--4336, 2024

    2024

  7. [7]

    Ltx-video: Realtime video latent diffusion

    HaCohen, Y., Chiprut, N., Brazowski, B., Shalem, D., Moshe, D., Richardson, E., Levin, E., Shiran, G., Zabari, N., Gordon, O., et al. Ltx-video: Realtime video latent diffusion. arXiv preprint arXiv:2501.00103, 2024

    arXiv 2024

  8. [8]

    and Salimans, T

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

    arXiv 2022

Show all 20 references
  1. [9]

    Pyramidflow: High-resolution defect contrastive localization using pyramid normalizing flow

    Lei, J., Hu, X., Wang, Y., and Liu, D. Pyramidflow: High-resolution defect contrastive localization using pyramid normalizing flow. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp.\ 14143--14152, June 2023

    2023

  2. [10]

    Differentiable vector graphics rasterization for editing and learning

    Li, T.-M., Luk \'a c , M., Gharbi, M., and Ragan-Kelley, J. Differentiable vector graphics rasterization for editing and learning. ACM Transactions on Graphics (TOG), 39 0 (6): 0 1--15, 2020

    2020

  3. [11]

    Score-based generative modeling through stochastic evolution equations in hilbert spaces

    Lim, S., Yoon, E., Byun, T., Kang, T., Kim, S., Lee, K., and Choi, S. Score-based generative modeling through stochastic evolution equations in hilbert spaces. In Proceedings of the 37th International Conference on Neural Information Processing Systems, NIPS '23, Red Hook, NY,...

    2023

  4. [12]

    T., and Mildenhall, B

    Poole, B., Jain, A., Barron, J. T., and Mildenhall, B. Dreamfusion: Text-to-3d using 2d diffusion. arXiv preprint arXiv:2209.14988, 2022

    2022 arXiv

  5. [13]

    Animatelcm: Computation-efficient personalized style video generation without personalized video data

    Wang, F.-Y., Huang, Z., Bian, W., Shi, X., Sun, K., Song, G., Liu, Y., and Li, H. Animatelcm: Computation-efficient personalized style video generation without personalized video data. In SIGGRAPH Asia 2024 Technical Communications, pp.\ 1--5. 2024

    2024

  6. [14]

    Modelscope text-to-video technical report

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

    2023 arXiv

  7. [15]

    Aniclipart: Clipart animation with text-to-video priors

    Wu, R., Su, W., Ma, K., and Liao, J. Aniclipart: Clipart animation with text-to-video priors. International Journal of Computer Vision, pp.\ 1--17, 2024

    2024

  8. [16]

    Tooncrafter: Generative cartoon interpolation

    Xing, J., Liu, H., Xia, M., Zhang, Y., Wang, X., Shan, Y., and Wong, T.-T. Tooncrafter: Generative cartoon interpolation. ACM Transactions on Graphics (TOG), 43 0 (6): 0 1--11, 2024

    2024

  9. [17]

    Dynamicrafter: Animating open-domain images with video diffusion priors

    Xing, J., Xia, M., Zhang, Y., Chen, H., Yu, W., Liu, H., Liu, G., Wang, X., Shan, Y., and Wong, T.-T. Dynamicrafter: Animating open-domain images with video diffusion priors. In European Conference on Computer Vision, pp.\ 399--417. Springer, 2025

    2025

  10. [18]

    Unipose: Detecting any keypoints

    Yang, J., Zeng, A., Zhang, R., and Zhang, L. Unipose: Detecting any keypoints. arXiv preprint arXiv:2310.08530, 2023

    2023 arXiv

  11. [19]

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

    Zhang, S., Wang, J., Zhang, Y., Zhao, K., Yuan, H., Qin, Z., Wang, X., Zhao, D., and Zhou, J. I2vgen-xl: High-quality image-to-video synthesis via cascaded diffusion models. arXiv preprint arXiv:2311.04145, 2023

    2023 arXiv

  12. [20]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.