- Rating
- 2
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 1
Read review
Paper summary
This paper addresses the robustness shortcomings of existing image-to-video (I2V) generation protection methods by proposing the RVDGuard defense framework. It achieves its objectives by applying input-side transformation and optimizing only the first step. Experiments validated the effectiveness of the framework on three mainstream I2V models: CogVideoX and the Wan series.
Critical strengths
-
Precise problem identification: Poor robustness of input perturbations and the fact that the output remains usable are two common issues in real-world deployments, making this research valuable.
-
The method is designed to be simple and efficient.
Critical weaknesses
-
This approach is more of an engineering combination than a methodological innovation: The EoT method proposed in 2018 has been widely adopted, while the “first-step velocity perturbation” method simply simplifies existing methods to the first step (similar to “Perturbing the early stages” mentioned in DiffusionGuard ICLR’25).
-
The paper only verifies the robustness of its own method and does not compare it with baseline methods, failing to demonstrate the motivation behind the proposed robustness.
Questions for authors
-
Theoretically, why is First-Step Velocity Deflection more effective than full optimization? Are there any model-specific limitations?
-
How much has robustness improved compared to baseline methods?
-
Since the goal is to improve robustness, how does the method perform under diffusion-based purification attacks?
-
Why do some of the results shown in the table 4 (transferability) outperform the non-transfer results?
-
Only showing one visual result isn't enough to illustrate the findings. More visualizations and videos are needed.
4 comments
Response to Reviewer J3zn (2/4)
W2&Q2: Missing robustness comparison against baseline protection methods.
A2: We add a robustness comparison with all five baselines and report the results in Table R1. The baselines fail to withstand common transformations. PhotoGuard, Anti-I2V, and I2VGuard, their protection is largely erased by simple post-processing. Immune2V is the most robust baseline and still lowers quality-related metrics such as Imaging Quality. RVDGuard, in contrast, keeps Subject Consistency between 0.694 and 0.802 under all transformations.
Table R1. Robustness results on Wan2.2-TI2V-5B.
| Setting | Method | Subject Consistency ↓ | I2V Subject ↓ | Aesthetic Quality ↓ | Imaging Quality ↓ |
|---|---|---|---|---|---|
| JPEG compression, | PhotoGuard | 0.940 | 0.952 | 0.590 | 0.666 |
| Immune2V | 0.904 | 0.878 | 0.506 | 0.610 | |
| Anti-I2V | 0.941 | 0.967 | 0.618 | 0.665 | |
| Anti-Prompt | 0.946 | 0.967 | 0.605 | 0.671 | |
| I2VGuard | 0.937 | 0.954 | 0.595 | 0.657 | |
| RVDGuard | 0.702 | 0.868 | 0.467 | 0.567 | |
| Gaussian noise, | PhotoGuard | 0.939 | 0.949 | 0.591 | 0.630 |
| Immune2V | 0.892 | 0.867 | 0.486 | 0.648 | |
| Anti-I2V | 0.941 | 0.966 | 0.603 | 0.640 | |
| Anti-Prompt | 0.938 | 0.963 | 0.599 | 0.638 | |
| I2VGuard | 0.935 | 0.953 | 0.598 | 0.614 | |
| RVDGuard | 0.717 | 0.872 | 0.492 | 0.585 | |
| Resizing, | PhotoGuard | 0.954 | 0.956 | 0.610 | 0.623 |
| Immune2V | 0.937 | 0.890 | 0.518 | 0.523 | |
| Anti-I2V | 0.952 | 0.971 | 0.622 | 0.636 | |
| Anti-Prompt | 0.956 | 0.970 | 0.622 | 0.632 | |
| I2VGuard | 0.945 | 0.959 | 0.610 | 0.610 | |
| RVDGuard | 0.765 | 0.889 | 0.504 | 0.601 | |
| Gaussian blur, | PhotoGuard | 0.953 | 0.958 | 0.612 | 0.600 |
| Immune2V | 0.936 | 0.890 | 0.518 | 0.493 | |
| Anti-I2V | 0.953 | 0.973 | 0.629 | 0.613 | |
| Anti-Prompt | 0.952 | 0.970 | 0.623 | 0.610 | |
| I2VGuard | 0.941 | 0.959 | 0.611 | 0.587 | |
| RVDGuard | 0.694 | 0.850 | 0.489 | 0.558 | |
| Center cropping, ratio 0.7 | PhotoGuard | 0.939 | 0.920 | 0.578 | 0.625 |
| Immune2V | 0.915 | 0.851 | 0.482 | 0.577 | |
| Anti-I2V | 0.943 | 0.933 | 0.599 | 0.662 | |
| Anti-Prompt | 0.940 | 0.932 | 0.585 | 0.639 | |
| I2VGuard | 0.937 | 0.913 | 0.586 | 0.627 | |
| RVDGuard | 0.802 | 0.938 | 0.534 | 0.588 |
Response to Reviewer J3zn (3/4)
Q1: Theoretically, why is First-Step Velocity Deflection more effective than full optimization? Are there any model-specific limitations?
A3: Theoretical support. We have added a formal proposition to Appendix C. For a flow-matching model integrated with a deterministic linear solver, such as Euler or the UniPC sampler of Wan, the final latent satisfies with , where is the initial noise scale. Since clean and protected generations share , writing gives and hence , where and . The leading term is exactly the shift in the model's initial clean-video estimate, and when the two velocities have comparable norms, its magnitude grows monotonically with the cosine dissimilarity we maximize. Measured with the official samplers across 100 images per model, stays strictly below 1 on every image, with a mean of 0.68 on Wan2.1-I2V-14B and 0.74 on Wan2.2-TI2V-5B, retaining 51% and 55% of the first-step deviation in the final latent . By contrast, a random first-step change of the same norm yields and retains only 4%–5%.
Comparison with full optimization. While the ideal goal is to maximize the final latent deviation , back-propagating through the unrolled multi-step sampler largely fails in practice (0.900 vs. 0.659 in Table 9), whereas joint optimization over ten steps splits the fixed budget across nearly orthogonal step-wise directions (Tables 7–8), performing stronger on some metrics and weaker on others at ten times the computational cost (Table 9). Our first-step objective directly maximizes the leading term with a single clean denoiser evaluation.
Model-specific limitations. The exact linear decomposition above holds for flow-matching models with deterministic linear solvers, namely Wan2.1 and Wan2.2. For CogVideoX, a -prediction model sampled with DDIM, an analogous weighted decomposition holds where averages 0.90 and 21% of the deviation is retained, still far above 3.5% for a random change, consistent with Table 6 where later-step injections also remain active. Finally, the method requires the model's first-step prediction to be conditioned on the input image.
Response to Reviewer J3zn (4/4)
Q3: Missing evaluation under diffusion-based purification attacks.
A4: We evaluate two representative diffusion-based purification attacks: DiffPure [1], and IMPRESS [2]. For DiffPure, we encode the image using Stable Diffusion 1.5 and add noise at step 60 of 1000 before reverse denoising. For IMPRESS, we follow its default setup with 1000 Adam optimization steps with a learning rate of , and an LPIPS loss weight of 0.01 with threshold 0.1. We evaluate the first 100 images of VBench-I2V with Wan2.2-TI2V-5B, and we will extend it to the full evaluation set in the revised version. As shown in Table R2, existing protection baselines are largely neutralized by both purification attacks, four of the five baselines return to at least 0.956, and Immune2V stays at 0.908 and 0.949. In contrast, RVDGuard maintains significant degradation under both purifications, keeping Subject Consistency down to 0.712 and 0.836, and Imaging Quality to 0.496 and 0.605. This demonstrates that deflecting the first-step velocity direction provides substantially stronger resilience against diffusion-based purification than existing baselines.
Table R2. Quantitative results against diffusion-based purification on Wan2.2-TI2V-5B.
| Setting | Method | Subject Consistency ↓ | I2V Subject ↓ | Aesthetic Quality ↓ | Imaging Quality ↓ |
|---|---|---|---|---|---|
| DiffPure [1] | Clean | 0.973 | 0.984 | 0.722 | 0.713 |
| PhotoGuard | 0.960 | 0.948 | 0.598 | 0.714 | |
| Anti-Prompt | 0.963 | 0.954 | 0.608 | 0.719 | |
| Immune2V | 0.908 | 0.870 | 0.585 | 0.705 | |
| Anti-I2V | 0.965 | 0.967 | 0.614 | 0.722 | |
| I2VGuard | 0.959 | 0.946 | 0.604 | 0.701 | |
| RVDGuard | 0.712 | 0.861 | 0.535 | 0.496 | |
| IMPRESS [2] | Clean | 0.972 | 0.988 | 0.707 | 0.704 |
| PhotoGuard | 0.958 | 0.968 | 0.607 | 0.688 | |
| Anti-Prompt | 0.956 | 0.971 | 0.607 | 0.692 | |
| Immune2V | 0.949 | 0.917 | 0.592 | 0.701 | |
| Anti-I2V | 0.968 | 0.978 | 0.608 | 0.707 | |
| I2VGuard | 0.959 | 0.965 | 0.604 | 0.669 | |
| RVDGuard | 0.836 | 0.916 | 0.587 | 0.605 |
Q4: Unclear reason why transfer results in Table 4 outperform non-transfer results.
A5: The transfer results do not outperform the non-transfer results overall, and only a subset of metrics are lower because different target models have different sensitivities to image-conditioning perturbations, making raw scores not directly comparable across architectures. Specifically, DynamiCrafter is more sensitive to conditioning perturbations than the three DiT models, which is evident when comparing Table 4 with Table 1: although all baselines are optimized on Wan2.2-TI2V-5B alone, their Motion Smoothness on DynamiCrafter is lower than on Wan2.2-TI2V-5B across every single baseline, and four of the five baselines also yield lower I2V Subject and I2V Background on DynamiCrafter. RVDGuard follows the same pattern: four of the seven metrics remain lower in the non-transfer settings on Wan2.1-I2V-14B and CogVideoX-5B-I2V, and Subject Consistency, Background Consistency, and Aesthetic Quality remain lower on Wan2.2-TI2V-5B than on DynamiCrafter.
Q5: Insufficient visual results and video demonstrations.
A6: We thank the reviewer for this suggestion and have added six additional multi-model qualitative comparisons in Appendix J of the revised manuscript, along with the corresponding generated video files in the Supplementary Material.
References
[1] Nie, Weili, Brandon Guo, Yujia Huang, Chaowei Xiao, Arash Vahdat, and Animashree Anandkumar. "Diffusion models for adversarial purification." International Conference on Machine Learning (ICML), PMLR, 2022.
[2] Cao, Bochuan, Changjiang Li, Ting Wang, Jinyuan Jia, Bo Li, and Jinghui Chen. "IMPRESS: Evaluating the resilience of imperceptible perturbations against unauthorized data usage in diffusion-based generative AI." Advances in Neural Information Processing Systems 36 (2023): 10657-10677.
Sincerely,
Authors
Response to Reviewer J3zn (1/4)
Dear Reviewer J3zn,
We appreciate your recognition of the precise problem identification and the simplicity of the method. We respond to your concerns on novelty positioning, baseline robustness, and purification below; W2 and Q2 are merged as they address the same issue.
W1: Limited methodological innovation and combination of existing techniques.
A1: We respectfully disagree that RVDGuard merely simplifies existing methods to the first step. We agree that EoT is widely used and that DiffusionGuard already optimizes at the first denoising step, so neither is our claimed novelty. Our contribution is what is optimized at that step and why it suffices for I2V.
DiffusionGuard maximizes the norm of the first-step predicted noise of the protected image alone, motivated by inpainting models copying unmasked details at the first step. We instead maximize the directional deviation between the first-step velocities of the protected and the original image under a shared noise. This distinction matters in I2V: injecting the induced velocity difference at step 0 alone degrades the video almost as much as injecting it at every step, whereas a random change of the same magnitude has almost no effect (Tables 5 and 6). Perturbations optimized for different steps are nearly orthogonal (Table 7), so multi-step optimization multiplies the cost for mixed gains (Table 9).
The two components are also not independent: optimizing step 0 alone needs one denoiser pass per transformation, which is what makes averaging over transformations at every iteration tractable.