Pith. sign in

REVIEW 2 major objections 6 minor 66 references

A training-free detector finds AI-generated and partially edited videos by measuring rough patch trajectories and incoherent local motion in frozen vision features, at 0.1% false-positive rates on real 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

2026-07-12 06:21 UTC pith:SJ5G4C2O

load-bearing objection Training-free patch-level detector that actually works at 0.1% FPR on full and partial fakes, with a service-aligned protocol that is more useful than AUROC. the 2 major comments →

arxiv 2607.02886 v1 pith:SJ5G4C2O submitted 2026-07-03 cs.CV cs.AIcs.LG

SPLIT: Training-Free AI-Generated and Partially Edited Video Detection via Spatial Patch-Level Incoherence and Temporal Roughness

Jongyeop Hyun , Hyounghun Kim This is my paper
classification cs.CV cs.AIcs.LG
keywords AI-generated video detectiontraining-free detectionpartial video manipulationpatch tokenstemporal roughnessmotion incoherenceultra-low FPRreal-only calibration
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.

Real-world video services cannot afford to reject authentic user uploads, so detectors must keep the false-positive rate on real video extremely low (around 0.1%). Standard scores such as AUROC hide how poorly many methods actually operate at that threshold. This paper introduces SPLIT, a training-free method that never sees a single synthetic example. It simply extracts patch tokens from a frozen vision encoder and looks for two statistical signatures of manipulation: trajectories that are rough rather than smooth (one-step path length exceeds the normalized two-step chord) and motion fields whose neighboring patches disagree. The two signals are multiplied with a fixed gamma correction and thresholded using only real video. On three complementary benchmarks covering partial edits, earlier generators, and the newest 2024–2025 systems, the method records the highest Fake Recall at the strictest operating point while remaining robust to blur, compression, and flips and adding almost no cost beyond the encoder forward pass. The authors also supply a service-aligned evaluation protocol—Fake Recall at fixed FPR with real-only calibration and cross-real threshold transfer—so that future detectors can be judged under the constraints that actually matter for deployment.

Core claim

Localized spatiotemporal incoherence measured on patch tokens of a frozen vision encoder—specifically the log-ratio of one-step versus two-step feature path lengths (TTR) and the average magnitude of spatial gradients of the feature-space motion field (LSMI)—yields a single score that separates both fully generated and sparsely edited video from real video at ultra-low false-positive rates after real-only threshold calibration, without any task-specific training.

What carries the argument

Two-step Temporal Roughness (TTR) together with Local Spatial Motion Incoherence (LSMI), fused as Score = TTR^γ · LSMI (γ = 8). TTR compares accumulated consecutive-step displacements of each patch token against a normalized two-step chord; LSMI measures how much neighboring patches disagree in their temporal feature changes. The product, thresholded on real video alone, is the detector.

Load-bearing premise

That the excess of one-step over two-step patch roughness and of local motion-field gradients remains a stable, generator-agnostic cue once a single fixed gamma and a real-only threshold are chosen.

What would settle it

A new video generator (or a post-processing pipeline) that produces patch-token trajectories whose one-step and two-step lengths stay as well matched as real video, and whose feature-space motion fields remain spatially coherent, would drive Fake Recall at 0.1% FPR toward the levels of existing baselines and falsify the claim.

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

If this is right

  • Service operators can set a single real-only threshold and obtain high recall on both fully synthetic and partially edited uploads without collecting or labeling generator-specific training data.
  • Partial-edit benchmarks such as FakeParts become practical stress tests rather than near-failure modes, because the detector never collapses the frame into a global embedding.
  • Cross-real threshold transfer can replace ad-hoc AUROC reporting as the primary metric for video-manipulation detectors that must operate under fixed false-positive budgets.
  • Any frozen vision encoder that yields patch tokens can be swapped in; the scoring logic itself adds negligible overhead and requires no fine-tuning.

Where Pith is reading between the lines

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

  • The same patch-level roughness statistics may transfer to still-image partial edits if the two-step idea is replaced by multi-scale spatial path-length comparisons.
  • If future generators begin to explicitly regularize feature-space trajectories for smoothness, the method will force a measurable quality–detectability trade-off that can be tracked over time.
  • The real-only calibration protocol itself is a reusable template for any high-stakes media detector (audio, image, video) that must guarantee ultra-low false-positive rates without target-domain fakes.

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 / 6 minor

Summary. The paper proposes SPLIT, a training-free detector for fully AI-generated and partially edited videos that scores frozen vision-encoder patch tokens with two complementary signals: Two-step Temporal Roughness (TTR; Eqs. 2–5), a log-ratio of accumulated one-step versus normalized two-step feature path lengths, and Local Spatial Motion Incoherence (LSMI; Eqs. 6–9), the mean magnitude of spatial gradients of a feature-space motion field. The signals are fused as TTR^γ · LSMI with a single fixed γ=8 (Eq. 10) and thresholded under a service-aligned protocol: Fake Recall at FPR ∈ {0.1%, 1%, 5%} with real-only calibration and cross-real threshold transfer. On FakeParts, GenVideo, and ViF-Bench, SPLIT reports the highest Fake Recall at FPR=0.1% (74.45%, 85.16%, 84.52% overall under cross-real transfer), outperforming supervised baselines and training-free D3, with supporting ablations (components, γ, encoders), post-processing robustness, bootstrap CIs, and AUROC tables.

Significance. If the results hold, the work is significant on two fronts. First, it shows that localized spatiotemporal incoherence on frozen patch tokens can detect both full synthesis and sparse partial edits without task-specific training, with large gains at the ultra-low-FPR regime that matters for deployment. Second, the evaluation protocol itself—Fake Recall at fixed FPR with real-only calibration and cross-real transfer—is a useful corrective to AUROC-centric reporting in this area. Strengths include public code, multi-benchmark generator- and category-stratified tables, encoder-agnostic gains (Tab. 7), fixed-γ saturation (Fig. 4, App. E), cross-real FPR stability (Tab. 4), and bootstrap CIs that separate SPLIT from D3. The main practical caveat is weak inpainting recall, which the authors disclose.

major comments (2)
  1. §4.2 and Tab. 1: Inpainting Fake Recall at FPR=0.1% is only 2.48% (second-best among compared methods). Because FakeParts is a central claim for partial-edit detection, the manuscript should more explicitly bound when patch-level TTR/LSMI is expected to fail (e.g., very small spatial support) and, if possible, quantify recall as a function of edited-area fraction or provide a simple spatial-max/top-k pooling variant so readers can see whether the failure is fundamental or aggregation-related.
  2. §3.5 and §5.2: γ=8 is chosen once on a small held-out split and then frozen. The paper already shows saturation on FakeParts/GenVideo/ViF-Bench, but the held-out split is not named or sized. A one-sentence statement of which data were used for that single choice (and confirmation that no test-set FPR targets were used) would close the only remaining free-parameter concern for reproducibility.
minor comments (6)
  1. Fig. 2 fusion formula uses nonstandard notation (TTR! × LSMI, γ=8); align it with Eq. (10) for consistency.
  2. Eq. (3): the normalization factor (T−1)/(T−2) is clear for T≥3, but a brief note on minimum clip length after 8 fps sampling would help implementers.
  3. Tab. 3 header says FPR=0.1% only; a pointer in the caption to App. Tabs. 8–9 for 1%/5% would improve navigation.
  4. §4.1 Implementation Details: crop 10% from the longer side is underspecified (center vs. random); state the exact policy used for all reported numbers.
  5. Related Work: a short explicit contrast with ReStraV’s frame-level straightening vs. SPLIT’s patch-level 1-step/2-step ratio would sharpen positioning.
  6. Appendix C.1: the multi-step slope fit for K>2 is interesting; a one-line intuition why longer steps dilute the local inconsistency cue would help readers.

Circularity Check

0 steps flagged

No significant circularity: training-free patch signals, fixed gamma, and real-only thresholds do not reduce Fake Recall to evaluation inputs by construction.

full rationale

SPLIT is a training-free detector whose scores are defined from frozen encoder patch tokens via explicit formulas (TTR as the log-ratio of accumulated 1-step vs. normalized 2-step path lengths, LSMI as mean spatial gradients of the feature-space motion field, fused as TTR^γ · LSMI with a single fixed γ=8 chosen once on a held-out split). Thresholds are calibrated solely on real videos to meet a target FPR and then transferred; Fake Recall is measured on held-out manipulated sets (FakeParts, GenVideo, ViF-Bench). No equation equates the reported Fake Recall to a quantity fitted on the same fake data, no uniqueness theorem is imported from the authors, and self-citations are limited to standard baselines and prior detectors (D3, ReStraV, etc.) that are independently evaluated rather than load-bearing premises. Ablations (patch-level vs. global, γ sweep, encoder sweep, post-processing) further show the gains are not forced by construction. The service-aligned protocol itself is an evaluation contribution, not a circular derivation. Score 0 is therefore appropriate.

Axiom & Free-Parameter Ledger

3 free parameters · 3 axioms · 2 invented entities

The method rests on standard frozen-encoder assumptions plus two hand-designed statistics and one fixed sharpening constant. No new physical entities are postulated; the free parameters are ordinary hyper-parameters of a training-free detector.

free parameters (3)
  • gamma (fusion exponent) = 8
    Fixed at 8 after a single held-out validation sweep; amplifies TTR differences. Ablated in Fig. 4 and appendix; performance saturates beyond 8.
  • epsilon (numerical stability) = small positive constant
    Small additive constant inside log-ratio of TTR; standard numerical safeguard, not fitted to maximize recall.
  • frame sampling / clip length = 8 fps, ≤2 s
    8 fps, up to 2 s, 10% crop, 224×224 resize—chosen to match D3 protocol rather than optimized on test sets.
axioms (3)
  • domain assumption Frozen vision-encoder patch tokens preserve localized generation and editing artifacts as measurable deviations from smooth temporal evolution and spatially coherent motion.
    Stated in §3.1–3.4 and used throughout; empirically supported on the three benchmarks but not proved for arbitrary future generators.
  • domain assumption Real-video score tails under TTR×LSMI are stable enough that a threshold calibrated on one real corpus (ROVI or MSR-VTT) approximately preserves the target FPR on another.
    Formalized by the cross-real transfer protocol of §4.1 and Tab. 4; central to the service-aligned claim.
  • standard math Standard arithmetic and log operations on feature differences yield a well-defined roughness ratio (Eqs. 2–5).
    Elementary; no novel mathematics.
invented entities (2)
  • Two-step Temporal Roughness (TTR) no independent evidence
    purpose: Scalar per-patch and clip-level statistic that contrasts accumulated 1-step versus normalized 2-step feature path lengths.
    Defined in Eqs. 2–5; no independent physical existence outside the detector.
  • Local Spatial Motion Incoherence (LSMI) no independent evidence
    purpose: Mean magnitude of spatial gradients of the feature-space motion field.
    Defined in Eqs. 6–9; engineered cue rather than a previously measured physical quantity.

pith-pipeline@v1.1.0-grok45 · 48896 in / 2694 out tokens · 28359 ms · 2026-07-12T06:21:53.653708+00:00 · methodology

0 comments
read the original abstract

Deploying AI-generated video detectors in real-world services demands an ultra-low false positive rate (FPR) on real videos to avoid falsely rejecting authentic content, a regime where standard metrics such as AUROC fail to reflect actual operating behavior. We introduce Spatial Patch-Level Incoherence and Temporal Roughness (SPLIT), a training-free detector that operates on patch tokens from a frozen vision encoder to detect both fully generated and partially edited videos. SPLIT computes two complementary signals: Two-step Temporal Roughness (TTR), capturing non-smooth patch trajectories via one-step and two-step feature variation contrast, and Local Spatial Motion Incoherence (LSMI), measuring spatially inconsistent temporal changes through gradients of a feature-space motion field. The two are fused multiplicatively with gamma correction to sharpen real-fake separation at strict thresholds. We further propose a service-aligned evaluation protocol based on Fake Recall at fixed FPR with real-only threshold calibration and cross-real threshold transfer. Across three benchmarks (FakeParts, GenVideo, and ViF-Bench), SPLIT achieves the highest Fake Recall at FPR = $0.1\%$, substantially outperforming supervised and training-free baselines while remaining robust to post-processing with negligible overhead. The code is publicly available at https://github.com/mldljyh/SPLIT .

Figures

Figures reproduced from arXiv: 2607.02886 by Hyounghun Kim, Jongyeop Hyun.

Figure 1
Figure 1. Figure 1: Fake Recall (%) at FPR = 0.1% for FID, AIGVDet, ReStraV, D3, and SPLIT across FakeParts, GenVideo, and ViF-Bench. SPLIT (red) consistently achieves the highest and most uniform recall across all manipulation types and generators. rare (e.g., FPR on real videos ≤ 0.1%), yet recall on manipulated content must stay high [55]. In this regime, aggregate metrics such as AUROC can be mis￾leading because they do n… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of SPLIT. A frozen vision encoder produces patch tokens P ∈ R T ×N×D, from which two signals are computed: Two-step Temporal Roughness (TTR), comparing 1-step and 2-step feature variations, and Local Spatial Motion Incoherence (LSMI), measuring spatial gradients of a feature-space motion field. These are com￾bined via gamma-corrected multiplicative fusion and thresholded using real-only cali￾brati… view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of FakeParts Overall Fake Recall at FPR = 0.1% (left) and end￾to-end inference time on 1,000 FakeParts T2V samples and batch size of 1 (right, log scale). Inference time for AIGVDet includes RAFT [47]-based optical flow preprocess￾ing. Analysis. The performance gap is most pronounced at FPR = 0.1%, precisely the regime relevant to deployment. This supports our central hypothesis: local￾ized spat… view at source ↗
Figure 4
Figure 4. Figure 4: Effect of the gamma exponent γ on FakeParts Overall Fake Recall (%). 5 Ablation Studies 5.1 Contribution of Each Component We isolate the contribution of each SPLIT component on FakeParts Overall under cross-real threshold transfer, as summarized in Tab. 6. Replacing global [CLS] embeddings with patch-level tokens for TTR yields the single largest im￾provement, boosting Fake Recall by +14.08 points at FPR … view at source ↗
Figure 5
Figure 5. Figure 5: For each patch index n, we plot the mean log2 (L1(n) + ϵ) and mean log2 (L2(n)+ϵ) separately for real and fake videos, with shaded regions indicating standard deviation across videos. Since \mathrm {TTR}(n)=\log _2(L_1(n)+\epsilon )-\log _2(L_2(n)+\epsilon ), the vertical gap between the two curves at a given patch index directly reflects the average per-patch roughness signal [PITH_FULL_IMAGE:figures/ful… view at source ↗
Figure 5
Figure 5. Figure 5: Patch-wise visualization of the two TTR components on FakeParts (left: real videos from ROVI; right: fake videos from Style Change). For each patch index n, we plot the mean log2 (L1(n) + ϵ) and mean log2 (L2(n) + ϵ), with shaded bands denoting one standard deviation across videos. Two observations are clear. First, for real videos, the two curves remain relatively close across most patch indices. This ind… view at source ↗
Figure 6
Figure 6. Figure 6: Additional sensitivity to the gamma exponent on GenVideo and ViF-Bench. then saturates, with little additional gain at γ = 16. These results match the FakeParts trend and support using a single fixed value, γ = 8, for all datasets and FPR targets without per-benchmark tuning. F AUROC Results and Analysis For completeness, we report conventional cross-real AUROC (%, harmonic mean across the two real domains… 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

66 extracted references · 7 canonical work pages

  1. [1]

    In: Lin, Z., Cheng, M.M., He, R., Ubul, K., Silamu, W., Zha, H., Zhou, J., Liu, C.L

    Bai, J., Lin, M., Cao, G., Lou, Z.: AI-Generated Video Detection via Spatial- Temporal Anomaly Learning. In: Lin, Z., Cheng, M.M., He, R., Ubul, K., Silamu, W., Zha, H., Zhou, J., Liu, C.L. (eds.) Pattern Recognition and Computer Vision. pp. 460–470. Springer Nature (2025).https://doi.org/10.1007/978-981-97- 8792-0_32

    doi:10.1007/978-981-97- 2025

  2. [2]

    Blattmann, A., Dockhorn, T., Kulal, S., Mendelevitch, D., Kilian, M., Lorenz, D., Levi, Y., English, Z., Voleti, V., Letts, A., Jampani, V., Rombach, R.: Stable Video Diffusion: Scaling Latent Video Diffusion Models to Large Datasets (2023).https: //doi.org/10.48550/arXiv.2311.15127,http://arxiv.org/abs/2311.15127

    Pith/arXiv arXiv doi:10.48550/arxiv.2311.15127 2023

  3. [3]

    OpenAI Blog1(8), 1 (2024)

    Brooks, T., Peebles, B., Holmes, C., DePue, W., Guo, Y., Jing, L., Schnurr, D., Taylor, J., Luhman, T., Luhman, E., et al.: Video generation models as world simulators. OpenAI Blog1(8), 1 (2024)

    2024

  4. [4]

    Hyun and H

    Chen, G., Lin, D., Yang, J., Lin, C., Zhu, J., Fan, M., Zhang, H., Chen, S., Chen, Z., Ma, C., Xiong, W., Wang, W., Pang, N., Kang, K., Xu, Z., Jin, Y., Liang, Y., Song, Y., Zhao, P., Xu, B., Qiu, D., Li, D., Fei, Z., Li, Y., Zhou, Y.: Skyreels-v2: Infinite-length film generative model (2025),https://arxiv.org/abs/2504.13074 16 J. Hyun and H. Kim

    Pith/arXiv arXiv 2025

  5. [5]

    Chen, H., Xia, M., He, Y., Zhang, Y., Cun, X., Yang, S., Xing, J., Liu, Y., Chen, Q., Wang, X., Weng, C., Shan, Y.: Videocrafter1: Open diffusion models for high- quality video generation (2023),https://arxiv.org/abs/2310.19512

    Pith/arXiv arXiv 2023

  6. [6]

    In: 2024 IEEE/CVF Conference on Computer Vision and Pattern Recog- nition (CVPR)

    Chen, H., Zhang, Y., Cun, X., Xia, M., Wang, X., Weng, C., Shan, Y.: VideoCrafter2: Overcoming Data Limitations for High-Quality Video Diffusion Models. In: 2024 IEEE/CVF Conference on Computer Vision and Pattern Recog- nition (CVPR). pp. 7310–7320 (2024).https://doi.org/10.1109/CVPR52733. 2024.00698

    doi:10.1109/cvpr52733 2024

  7. [7]

    Chen, H., Hong, Y., Huang, Z., Xu, Z., Gu, Z., Li, Y., Lan, J., Zhu, H., Zhang, J., Wang, W., Li, H.: DeMamba: AI-Generated Video Detection on Million-Scale GenVideo Benchmark (2024).https://doi.org/10.48550/arXiv.2405.19707

    Pith/arXiv arXiv doi:10.48550/arxiv.2405.19707 2024

  8. [8]

    IEEE Transactions on Information Forensics and Security19, 6084–6096 (2024)

    Coccomini, D.A., Zilos, G.K., Amato, G., Caldelli, R., Falchi, F., Papadopoulos, S., Gennaro, C.: MINTIME: Multi-Identity Size-Invariant Video Deepfake Detection. IEEE Transactions on Information Forensics and Security19, 6084–6096 (2024). https://doi.org/10.1109/TIFS.2024.3409054

    doi:10.1109/tifs.2024.3409054 2024

  9. [9]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (June 2019)

    Deng, J., Guo, J., Xue, N., Zafeiriou, S.: Arcface: Additive angular margin loss for deep face recognition. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (June 2019)

    2019

  10. [10]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)

    Esser, P., Chiu, J., Atighehchian, P., Granskog, J., Germanidis, A.: Structure and content-guided video synthesis with diffusion models. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). pp. 7346–7356 (October 2023)

    2023

  11. [11]

    In: First Conference on Language Modeling (2024)

    Gu, A., Dao, T.: Mamba: Linear-Time Sequence Modeling with Selective State Spaces. In: First Conference on Language Modeling (2024)

    2024

  12. [12]

    In: Proceedings of the 29th ACM International Conference on Multimedia

    Gu, Z., Chen, Y., Yao, T., Ding, S., Li, J., Huang, F., Ma, L.: Spatiotempo- ral Inconsistency Learning for DeepFake Video Detection. In: Proceedings of the 29th ACM International Conference on Multimedia. pp. 3473–3481. ACM (2021). https://doi.org/10.1145/3474085.3475508

    work page doi:10.1145/3474085.3475508 2021

  13. [13]

    In: 2018 15th IEEE International Conference on Advanced Video and Sig- nal Based Surveillance (AVSS)

    Güera, D., Delp, E.J.: Deepfake Video Detection Using Recurrent Neural Net- works. In: 2018 15th IEEE International Conference on Advanced Video and Sig- nal Based Surveillance (AVSS). pp. 1–6 (2018).https://doi.org/10.1109/AVSS. 2018.8639163

    work page doi:10.1109/avss 2018

  14. [14]

    HaCohen, Y., Chiprut, N., Brazowski, B., Shalem, D., Moshe, D., Richardson, E., Levin, E., Shiran, G., Zabari, N., Gordon, O., Panet, P., Weissbuch, S., Kulikov, V., Bitterman, Y., Melumian, Z., Bibi, O.: Ltx-video: Realtime video latent diffusion (2024),https://arxiv.org/abs/2501.00103

    Pith/arXiv arXiv 2024

  15. [15]

    Machine Learning77(1), 103–123 (Oct 2009).https:// doi.org/10.1007/s10994-009-5119-5

    Hand, D.J.: Measuring classifier performance: A coherent alternative to the area under the ROC curve. Machine Learning77(1), 103–123 (Oct 2009).https:// doi.org/10.1007/s10994-009-5119-5

    doi:10.1007/s10994-009-5119-5 2009

  16. [16]

    In: CVPR

    He, K., Zhang, X., Ren, S., Sun, J.: Deep Residual Learning for Image Recognition. In: CVPR. pp. 770–778 (2016)

    2016

  17. [17]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Howard, A., Sandler, M., Chu, G., Chen, L.C., Chen, B., Tan, M., Wang, W., Zhu, Y., Pang, R., Vasudevan, V., Le, Q.V., Adam, H.: Searching for MobileNetV3. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 1314–1324 (2019)

    2019

  18. [18]

    In: Belgrave, D., Zhang, C., Lin, H., Pascanu, R., Koniusz, P., Ghassemi, M., Chen, N

    Internò, C., Geirhos, R., Olhofer, M., Liu, S., Hammer, B., Klindt, D.: Ai-generated video detection via perceptual straightening. In: Belgrave, D., Zhang, C., Lin, H., Pascanu, R., Koniusz, P., Ghassemi, M., Chen, N. (eds.) Advances in Neu- ral Information Processing Systems. vol. 38, pp. 20672–20705. Curran Associates, Inc. (2025),https://proceedings.ne...

    2025

  19. [19]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)

    Jiang, Z., Han, Z., Mao, C., Zhang, J., Pan, Y., Liu, Y.: Vace: All-in-one video creation and editing. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). pp. 17191–17202 (October 2025)

    2025

  20. [20]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Kara, O., Kurtkaya, B., Yesiltepe, H., Rehg, J.M., Yanardag, P.: Rave: Random- ized noise shuffling for fast and consistent video editing with diffusion models. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 6507–6516 (2024)

    2024

  21. [21]

    com/global/

    Kling: Kling AI: Next-Generation AI Creative Studio (2025),https://klingai. com/global/

    2025

  22. [22]

    Kong, W., Tian, Q., Zhang, Z., Min, R., Dai, Z., Zhou, J., Xiong, J., Li, X., Wu, B., Zhang, J., Wu, K., Lin, Q., Yuan, J., Long, Y., Wang, A., Wang, A., Li, C., Huang, D., Yang, F., Tan, H., Wang, H., Song, J., Bai, J., Wu, J., Xue, J., Wang, J., Wang, K., Liu, M., Li, P., Li, S., Wang, W., Yu, W., Deng, X., Li, Y., Chen, Y., Cui, Y., Peng, Y., Yu, Z., H...

    Pith/arXiv arXiv 2025

  23. [23]

    Transactions on Machine Learning Research (2024),https://openreview.net/forum?id=RFrJCkw2oa, reproducibility Certifi- cation

    Ku, M., Wei, C., Ren, W., Yang, H., Chen, W.: Anyv2v: A tuning-free framework for any video-to-video editing tasks. Transactions on Machine Learning Research (2024),https://openreview.net/forum?id=RFrJCkw2oa, reproducibility Certifi- cation

    2024

  24. [24]

    Li, X., Xue, H., Ren, P., Bo, L.: Diffueraser: A diffusion model for video inpainting (2025),https://arxiv.org/abs/2501.10018

    Pith/arXiv arXiv 2025

  25. [25]

    https://doi.org/10.48550/arXiv.2512.15693

    Li, Y., Zheng, W., Zhang, Y., Sun, R., Zheng, Y., Chen, L., Zhou, J., Lu, J.: Skyra: AI-Generated Video Detection via Grounded Artifact Reasoning (2025). https://doi.org/10.48550/arXiv.2512.15693

    Pith/arXiv arXiv doi:10.48550/arxiv.2512.15693 2025

  26. [26]

    In: CVPRW

    Li, Y., Lyu, S.: Exposing DeepFake Videos By Detecting Face Warping Artifacts. In: CVPRW. pp. 46–52 (2019)

    2019

  27. [27]

    In: CVPR

    Liu, Y., Cun, X., Liu, X., Wang, X., Zhang, Y., Chen, H., Liu, Y., Zeng, T., Chan, R., Shan, Y.: EvalCrafter: Benchmarking and Evaluating Large Video Generation Models. In: CVPR. pp. 22139–22149 (2024)

    2024

  28. [28]

    https://doi.org/10.48550/arXiv.2508.21052

    Liu, Z., Gabetni, F., Sani, A.H., Wang, X., Daiboo, S., Brison, G., Franchi, G., Kalogeiton, V.: FakeParts: A New Family of AI-Generated DeepFakes (2025). https://doi.org/10.48550/arXiv.2508.21052

    work page doi:10.48550/arxiv.2508.21052 2025

  29. [29]

    https://hailuo- 02.com/ (2025)

    MiniMax: Hailuo 02 - #2 Global AI Video Generation Model. https://hailuo- 02.com/ (2025)

    2025

  30. [30]

    https://www.moonvalley.com/ (2022)

    Moonvalley: Moonvalley - the imagination research company. https://www.moonvalley.com/ (2022)

    2022

  31. [31]

    Morph Studio: Morph Studio - AI Image & Video Generator: Free Art Creator & Editor (2023),https://www.morphstudio.com/

    2023

  32. [32]

    com/hotshotco/hotshot-xl

    Mullan, J., Crawbuck, D., Sastry, A.: Hotshot-XL (Oct 2023),https://github. com/hotshotco/hotshot-xl

    2023

  33. [33]

    In: Avidan, S., Brostow, G., Cissé, M., Farinella, G.M., Hassner, T

    Ni, B., Peng, H., Chen, M., Zhang, S., Meng, G., Fu, J., Xiang, S., Ling, H.: Expanding Language-Image Pretrained Models for General Video Recognition. In: Avidan, S., Brostow, G., Cissé, M., Farinella, G.M., Hassner, T. (eds.) Computer Vision – ECCV 2022. pp. 1–18. Springer Nature Switzerland (2022).https://doi. org/10.1007/978-3-031-19772-7_1

    work page doi:10.1007/978-3-031-19772-7_1 2022

  34. [34]

    https://doi.org/10.48550/arXiv.2501.11340 18 J

    Ni, Z., Yan, Q., Huang, M., Yuan, T., Tang, Y., Hu, H., Chen, X., Wang, Y.: GenVidBench: A 6-Million Benchmark for AI-Generated Video Detection (2025). https://doi.org/10.48550/arXiv.2501.11340 18 J. Hyun and H. Kim

    doi:10.48550/arxiv.2501.11340 2025

  35. [35]

    NVIDIA, :, Agarwal, N., Ali, A., Bala, M., Balaji, Y., Barker, E., Cai, T., Chat- topadhyay, P., Chen, Y., Cui, Y., Ding, Y., Dworakowski, D., Fan, J., Fenzi, M., Ferroni, F., Fidler, S., Fox, D., Ge, S., Ge, Y., Gu, J., Gururani, S., He, E., Huang, J., Huffman, J., Jannaty, P., Jin, J., Kim, S.W., Klár, G., Lam, G., Lan, S., Leal- Taixe, L., Li, A., Li, ...

    Pith/arXiv arXiv 2025

  36. [36]

    van den Oord, A., Roman, E.: State-of-the-art video and image generation with veo 2 and imagen 3.https://blog.google/innovation-and-ai/models- and-research/google-labs/video-image-generation-update-december-2024/ (Dec 2024)

    2024

  37. [37]

    OpenAI: Sora 2 is here (2025),https://openai.com/index/sora-2/

    2025

  38. [38]

    TMLR (2023),https:// openreview.net/forum?id=a68SUt6zFt

    Oquab, M., Darcet, T., Moutakanni, T., Vo, H.V., Szafraniec, M., Khalidov, V., Fernandez, P., Haziza, D., Massa, F., El-Nouby, A., Assran, M., Ballas, N., Galuba, W.,Howes,R.,Huang,P.Y.,Li,S.W.,Misra,I.,Rabbat,M.,Sharma,V.,Synnaeve, G., Xu, H., Jegou, H., Mairal, J., Labatut, P., Joulin, A., Bojanowski, P.: DINOv2: Learning Robust Visual Features without ...

    2023

  39. [39]

    https://pika.art/ (2025)

    Pika Art: Pika art. https://pika.art/ (2025)

    2025

  40. [40]

    https://app.pixverse.ai (2025)

    PixVerse AI: PixVerse | Create Amazing AI Videos from Text & Photos with AI Video Generator. https://app.pixverse.ai (2025)

    2025

  41. [41]

    In: Proceedings of the 38th International Conference on Machine Learning

    Radford, A., Kim, J.W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., Krueger, G., Sutskever, I.: Learning Transferable Visual Models From Natural Language Supervision. In: Proceedings of the 38th International Conference on Machine Learning. pp. 8748–8763. PMLR (2021)

    2021

  42. [42]

    https://runwayml.com/research/introducing-runway-gen-4 (2025)

    Runway: Runway Research | Introducing Runway Gen-4. https://runwayml.com/research/introducing-runway-gen-4 (2025)

    2025

  43. [43]

    Simonyan, K., Zisserman, A.: Very Deep Convolutional Networks for Large-Scale Image Recognition (2015).https://doi.org/10.48550/arXiv.1409.1556,http: //arxiv.org/abs/1409.1556

    Pith/arXiv arXiv doi:10.48550/arxiv.1409.1556 2015

  44. [44]

    In: 2024 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR)

    Tan, C., Liu, H., Zhao, Y., Wei, S., Gu, G., Liu, P., Wei, Y.: Rethinking the Up- Sampling Operations in CNN-Based Generative Network for Generalizable Deep- fake Detection. In: 2024 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR). pp. 28130–28139 (2024).https://doi.org/10.1109/ CVPR52733.2024.02657

    arXiv 2024

  45. [45]

    In: Proceedings of the 36th International Conference on Machine Learning

    Tan, M., Le, Q.: EfficientNet: Rethinking Model Scaling for Convolutional Neu- ral Networks. In: Proceedings of the 36th International Conference on Machine Learning. pp. 6105–6114. PMLR (2019)

    2019

  46. [46]

    Team, G.: Mochi 1.https://github.com/genmoai/models(2024)

    2024

  47. [47]

    In: Vedaldi, A., Bischof, H., Brox, T., Frahm, J.M

    Teed, Z., Deng, J.: RAFT: Recurrent All-Pairs Field Transforms for Optical Flow. In: Vedaldi, A., Bischof, H., Brox, T., Frahm, J.M. (eds.) Computer Vision – ECCV

  48. [48]

    pp. 402–419. Springer International Publishing, Cham (2020).https://doi. org/10.1007/978-3-030-58536-5_24

    doi:10.1007/978-3-030-58536-5_24 2020

  49. [49]

    Wan, T., Wang, A., Ai, B., Wen, B., Mao, C., Xie, C.W., Chen, D., Yu, F., Zhao, H., Yang, J., Zeng, J., Wang, J., Zhang, J., Zhou, J., Wang, J., Chen, J., Zhu, K., SPLIT: Training-Free AI-Generated and Edited Video Detection 19 Zhao, K., Yan, K., Huang, L., Feng, M., Zhang, N., Li, P., Wu, P., Chu, R., Feng, R., Zhang, S., Sun, S., Fang, T., Wang, T., Gui...

    Pith/arXiv arXiv doi:10.48550/arxiv.2503.20314 2025

  50. [50]

    Wang, J., Yuan, H., Chen, D., Zhang, Y., Wang, X., Zhang, S.: Modelscope text- to-video technical report (2023),https://arxiv.org/abs/2308.06571

    Pith/arXiv arXiv 2023

  51. [51]

    In: The Thirteenth In- ternational Conference on Learning Representations (2025),https://openreview

    Wang, W., Wang, Q., Zheng, K., OUYANG, H., Chen, Z., Gong, B., Chen, H., Shen, Y., Shen, C.: Framer: Interactive frame interpolation. In: The Thirteenth In- ternational Conference on Learning Representations (2025),https://openreview. net/forum?id=Lp40Z40N07

    2025

  52. [52]

    NeurIPS37, 65618–65642 (2024).https://doi

    Wang, W., Yang, Y.: VidProM: A Million-scale Real Prompt-Gallery Dataset for Text-to-Video Diffusion Models. NeurIPS37, 65618–65642 (2024).https://doi. org/10.52202/079017-2096

    doi:10.52202/079017-2096 2024

  53. [53]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Wang, X., Courant, R., Christie, M., Kalogeiton, V.: Akira: Augmentation kit on rays for optical video generation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 2609–2619 (June 2025)

    2025

  54. [54]

    International Journal of Computer Vision133(5), 3059– 3078 (May 2025).https://doi.org/10.1007/s11263-024-02295-1

    Wang, Y., Chen, X., Ma, X., Zhou, S., Huang, Z., Wang, Y., Yang, C., He, Y., Yu, J., Yang, P., Guo, Y., Wu, T., Si, C., Jiang, Y., Chen, C., Loy, C.C., Dai, B., Lin, D., Qiao, Y., Liu, Z.: LaVie: High-Quality Video Generation with Cascaded Latent Diffusion Models. International Journal of Computer Vision133(5), 3059– 3078 (May 2025).https://doi.org/10.100...

    work page doi:10.1007/s11263-024-02295-1 2025

  55. [55]

    In: CVPR

    Wu, J., Li, X., Si, C., Zhou, S., Yang, J., Zhang, J., Li, Y., Chen, K., Tong, Y., Liu, Z., Loy, C.C.: Towards Language-Driven Video Inpainting via Multimodal Large Language Models. In: CVPR. pp. 12501–12511 (2024)

    2024

  56. [56]

    In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)

    Xiong, X., Patel, P., Fan, Q., Wadhwa, A., Selvam, S., Guo, X., Qi, L., Liu, X., Sengupta, R.: Talkingheadbench: A multi-modal benchmark & analysis of talking- head deepfake detection. In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV). pp. 4139–4149 (March 2026)

    2026

  57. [57]

    In: CVPR

    Xu, J., Mei, T., Yao, T., Rui, Y.: MSR-VTT: A Large Video Description Dataset for Bridging Video and Language. In: CVPR. pp. 5288–5296 (2016)

    2016

  58. [58]

    In: 2023 IEEE/CVF International Conference on Computer Vision (ICCV)

    Xu, Y., Liang, J., Jia, G., Yang, Z., Zhang, Y., He, R.: TALL: Thumbnail Layout for Deepfake Video Detection. In: 2023 IEEE/CVF International Conference on Computer Vision (ICCV). pp. 22601–22611. IEEE (2023).https://doi.org/10. 1109/ICCV51070.2023.02071

    arXiv 2023

  59. [59]

    In: The Thirteenth International Conference on Learning Represen- tations (2025),https://openreview.net/forum?id=LQzN6TRFg9

    Yang, Z., Teng, J., Zheng, W., Ding, M., Huang, S., Xu, J., Yang, Y., Hong, W., Zhang, X., Feng, G., Yin, D., Yuxuan.Zhang, Wang, W., Cheng, Y., Xu, B., Gu, X., Dong, Y., Tang, J.: Cogvideox: Text-to-video diffusion models with an expert transformer. In: The Thirteenth International Conference on Learning Represen- tations (2025),https://openreview.net/fo...

    2025

  60. [60]

    International Journal of Computer Vision133(4), 1879–1893 (Apr 2025).https: //doi.org/10.1007/s11263-024-02271-9

    Zhang, D.J., Wu, J.Z., Liu, J.W., Zhao, R., Ran, L., Gu, Y., Gao, D., Shou, M.Z.: Show-1:MarryingPixelandLatentDiffusionModelsforText-to-VideoGeneration. International Journal of Computer Vision133(4), 1879–1893 (Apr 2025).https: //doi.org/10.1007/s11263-024-02271-9

    work page doi:10.1007/s11263-024-02271-9 2025

  61. [61]

    NeurIPS37, 59570–59596 (2024).https://doi.org/10.52202/079017-1903 20 J

    Zheng, C., Lin, C., Zhao, Z., Wang, H., Guo, X., Liu, S., Shen, C.: Breaking Semantic Artifacts for Generalized AI-generated Image Detection. NeurIPS37, 59570–59596 (2024).https://doi.org/10.52202/079017-1903 20 J. Hyun and H. Kim

    work page doi:10.52202/079017-1903 2024

  62. [62]

    In: ICCV

    Zheng, C., Suo, R., Lin, C., Zhao, Z., Yang, L., Liu, S., Yang, M., Wang, C., Shen, C.: D3: Training-Free AI-Generated Video Detection Using Second-Order Features. In: ICCV. pp. 12852–12862 (2025)

    2025

  63. [63]

    In: 2021 IEEE/CVF International Conference on Computer Vision (ICCV)

    Zheng, Y., Bao, J., Chen, D., Zeng, M., Wen, F.: Exploring Temporal Coherence for More General Video Face Forgery Detection. In: 2021 IEEE/CVF International Conference on Computer Vision (ICCV). pp. 15024–15034 (2021).https://doi. org/10.1109/ICCV48922.2021.01477

    doi:10.1109/iccv48922.2021.01477 2021

  64. [64]

    Zheng, Z., Peng, X., Yang, T., Shen, C., Li, S., Liu, H., Zhou, Y., Li, T., You, Y.: Open-sora: Democratizing efficient video production for all (2024),https: //arxiv.org/abs/2412.20404

    Pith/arXiv arXiv 2024

  65. [65]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)

    Zhou, S., Li, C., Chan, K.C., Loy, C.C.: Propainter: Improving propagation and transformer for video inpainting. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). pp. 10477–10486 (October 2023)

    2023

  66. [66]

    Zhou, Y., Wang, Q., Cai, Y., Yang, H.: Allegro: Open the black box of commercial- level video generation model (2024),https://arxiv.org/abs/2410.15458 SPLIT: Training-Free AI-Generated and Edited Video Detection 21 Appendix A Benchmark Statistics This section provides detailed category and generator breakdowns for each test benchmark used in our experimen...

    Pith/arXiv arXiv 2024