- Rating
- 3(+1)
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
CineMem studies memory retrieval for autoregressive multi-shot video generation. Starting from a Wan2.1-1.3B model adapted using resampled-context training, it organizes historical KV tokens into online, head-specific semantic clusters. Each query–head pair retrieves its own relevant clusters, alongside persistent and recent context. The paper also introduces PTCS, which evaluates both preservation of required content and execution of requested shot changes. Experiments include comparisons on 100 multi-shot prompts, controlled retrieval ablations, attention-retrieval analyses, efficiency measurements, and a human study.
Critical strengths
-
The controlled retrieval ablations provide meaningful evidence for the central design. Table 2 applies alternative retrieval strategies to the same adapted model without additional training. Clustered query/head-wise retrieval obtains PTCS 0.8007, compared with 0.7280 for frame-based shared retrieval, 0.7525 for frame-based query-wise retrieval, and 0.7442 for clustered shared retrieval. This factorial comparison supports the combination of memory organization and individualized access, rather than attributing all improvements to additional training.
-
The paper addresses an important evaluation failure in multi-shot generation. A model can obtain high visual similarity by failing to execute the requested transition. PTCS explicitly accounts for both continuity and intended change. The human study in Appendix E.2 provides supporting evidence for this concern and for the method’s perceived inter-shot consistency. This makes the evaluation contribution useful beyond the proposed architecture.
Critical weaknesses
-
The evaluation does not directly validate the paper’s central claim of long-range revisitation. The main quantitative benchmark evaluates 15-second videos, whereas the 30-second extension is primarily qualitative. More fundamentally, PTCS operates on the middle frames of adjacent shots (Appendix E.1). Consider an A–B–A sequence in which a character appears in the first shot, disappears in the second, and returns in the third with an incorrect appearance. Adjacent-shot evaluation does not directly compare the two occurrences of that character; a text-conditioned evaluator may therefore reward individually plausible transitions without detecting the historical identity mismatch. This is a mismatch between the emphasized contribution and its principal evaluation, rather than simply a request for longer demonstrations. The human study provides useful supporting evidence, but reports aggregate consistency ratings rather than recurrence-specific accuracy by temporal gap. Consequently, the results establish improved multi-shot generation more convincingly than improved recovery of distant historical content.
-
The evidence does not establish the incremental necessity and practical value of fully query-wise retrieval. The argument that the union of query–head selections covers nearly all historical memory is not sufficient: even independent random selections of a fraction (p) of memory have expected union coverage (1-(1-p)^m) over (m) selections. With (p=0.3) and (m=20), this already exceeds 99.9%. This analytical null example does not imply that the proposed retrieval is random; it shows that union coverage alone cannot establish useful or necessary specialization. Table 2 provides evidence for the combined design, but bundles query-wise and head-wise retrieval rather than cleanly isolating their contributions to generation quality. The query-group variant also achieves PTCS 0.8233 versus 0.8242 for fully query-wise retrieval, without uncertainty estimates or a corresponding quality–runtime comparison. The already-cited OmniMem [4], which uses query-shared selection with head-specific access, is particularly relevant to this unresolved distinction. Moreover, Algorithm 1 performs global refinement over historical tokens while the number of clusters grows with history, so a fixed retrieval budget does not establish bounded cache-maintenance cost. These omissions leave the central fine-granularity design insufficiently justified relative to simpler alternatives.
-
The bibliography should be expanded to cite and substantively discuss [1–3]. They address complementary aspects of multi-scale memory, compact historical conditioning, and long-horizon consistency. In particular, OneStory is closely related to the multi-shot task and adaptive use of historical information. Please explain the differences in memory representation, retrieval granularity, temporal scope, and conditioning interface.
The concern is incomplete positioning and unresolved incremental benefit, not the absence of citations in isolation. Different backbones do not automatically require direct numerical comparison.
References
[1] Zhaochong An, Menglin Jia, Haonan Qiu, et al. OneStory: Coherent Multi-Shot Video Generation with Adaptive Memory. CVPR, 2026; preprint first released in 2025.
[2] Hongjun Wang, Lin Liu, Jianguo Li, and Tao Lin. Dual-Granularity Memory for Efficient Video Generation. CVPR, 2026.
[3] Jingyu Lin, Xinyi Shang, Peng Sun, Cunjian Chen, and Zhiqiang Shen. DynaMem: Consistent Long Video Generation via Hierarchical Memory and Motion Priors. ICML, 2026.
The paper contains a promising retrieval design and useful controlled ablations. However, the main evaluation does not directly measure distant revisitation, and the necessity of fully query-wise retrieval remains unresolved relative to simpler head-adaptive or query-group alternatives. These gaps affect the central contribution and are substantial enough to prevent an acceptance recommendation in the current version.
Questions for authors
-
Can you evaluate nonadjacent recurrence directly, using earlier visual references and explicit checks of entity identity, appearance, and scene recovery? Please stratify results by recurrence gap and include uncertainty estimates. How does PTCS score an A–B–A sequence with an identity mismatch between the two A shots?
-
Can you separate head-wise adaptation from query-wise adaptation under matched retrieved-token budgets and matched training? In particular, how do a head-specific/query-shared variant and the query-group variant compare with the full method in quality, latency, and memory use? Please compare the coverage analysis against a random-selection null or another control that distinguishes coverage from useful retrieval.
-
Do the reported inference timings include clustering, global refinement, cache reordering, gathering, and synchronization? Please provide an end-to-end breakdown and clarify how these costs grow with historical length.
These questions concern evidence needed to assess the current contribution, rather than optional extensions to future work.
10 comments
Response to Reviewer zE4D (2/9)
(Continued from the previous response.)
-
Direct evaluation of non-adjacent revisit pairs. As shown in Table A, CineMem achieves the highest scores across all three similarity dimensions and their averages for both all-shot and revisit pairs. In particular, its Revisit Shot Pair Consistency reaches 0.9054, substantially outperforming the second-best model, HoloCine (0.7949). We further stratify revisit pairs by temporal distance (4–8, 9–12, and 13–18 chunks). As shown in Table B, CineMem consistently achieves the best results across all distance ranges, including the longest range of 13–18 chunks (0.9123 vs. 0.8081 for HoloCine). This directly demonstrates its superior ability to recover historical visual information over extended temporal gaps.
Importantly, the original PTCS evaluates adjacent-shot transitions and does not directly compare the two occurrences of an entity in an A–B–A sequence. Therefore, an identity mismatch between the two A shots may not be captured by PTCS alone. Our newly introduced Revisit Shot Pair Consistency explicitly compares non-adjacent revisit pairs, providing a more direct assessment of subject identity, appearance, and scene preservation.
Table B. Revisit Shot Pair Consistency stratified by temporal distance
| Method | 4–8 chunks ↑ | 9–12 chunks ↑ | 13–18 chunks ↑ | Overall ↑ |
|---|---|---|---|---|
| Causal multi-shot models | ||||
| LongLive2 | 0.6404 ± 0.2060 | 0.6625 ± 0.0605 | 0.6543 ± 0.0840 | 0.6568 ± 0.0231 |
| ShotStream | 0.7723 ± 0.1974 | 0.7698 ± 0.0627 | 0.7722 ± 0.0656 | 0.7712 ± 0.0179 |
| Causal multi-prompt models | ||||
| Infinity-RoPE | 0.4740 ± 0.0069 | 0.4694 ± 0.0111 | 0.4634 ± 0.0123 | 0.4665 ± 0.0082 |
| Self-Forcing | 0.4758 ± 0.0096 | 0.5166 ± 0.0370 | 0.4840 ± 0.0360 | 0.4968 ± 0.0244 |
| MemFlow | 0.4934 ± 0.0019 | 0.4898 ± 0.0244 | 0.4839 ± 0.0204 | 0.4868 ± 0.0145 |
| Bidirectional models | ||||
| MultiShotMaster | 0.9496 ± 0.0118 | 0.7370 ± 0.0747 | 0.7793 ± 0.0819 | 0.7722 ± 0.0271 |
| HoloCine | 0.9216 ± 0.0295 | 0.7576 ± 0.0712 | 0.8081 ± 0.0588 | 0.7949 ± 0.0246 |
| CineMem (Ours) | 0.9871 ± 0.0022 | 0.8845 ± 0.0460 | 0.9123 ± 0.0401 | 0.9054 ± 0.0293 |
Response to Reviewer zE4D (3/9)
(Continued from the previous response.)
- Quantitative evaluation on 30-second videos. Since not all baselines support long-video generation, we select ShotStream, which formulates multi-shot generation as next-shot generation, as a representative baseline for 30-second video evaluation. As shown in Table C (Table 9 in the revised manuscript), CineMem achieves better results in transition control, text-video similarity, intra-shot subject consistency, and inter-shot similarity. More importantly, CineMem achieves higher All Shot Pair Consistency (0.7579 vs. 0.7317) and Revisit Shot Pair Consistency (0.8104 vs. 0.7329), providing further evidence of its effectiveness in long-video generation and historical content recovery.
- Revisit-specific human evaluation. We further refine the human study by separating Revisit Capability from the general Inter-shot Consistency dimension, directly evaluating whether previously observed subjects and scenes are consistently recovered after intervening shots. As shown in Table D, CineMem achieves the highest score, which provides complementary human evaluation evidence for CineMem's long-range revisitation capability.
Finally, we report 95% confidence intervals for all major quantitative results in the revised manuscript and standard deviations for the user study, providing a more comprehensive assessment of the reliability of our results.
Table C. Quantitative comparison on 30-second videos. Values are reported as mean ± half-width of 95% paired bootstrap confidence intervals.
Table C-1. Video-level, intra-shot, and inter-shot metrics.
| Method | Transition Control ↑ | Text-Video Sim. ↑ | Aesthetic ↑ | Subject Cons. ↑ | Background Cons. ↑ | Inter-shot Sim. ↑ | PCP ↑ | PTA ↑ | PTCS ↑ |
|---|---|---|---|---|---|---|---|---|---|
| ShotStream | 0.9034 ± 0.0363 | 0.2093 ± 0.0245 | 0.5717 ± 0.0222 | 0.9605 ± 0.0060 | 0.9589 ± 0.0056 | 0.6580 ± 0.0402 | 0.9653 ± 0.0149 | 0.6001 ± 0.0730 | 0.6203 ± 0.0779 |
| CineMem (Ours) | 0.9927 ± 0.0091 | 0.2240 ± 0.0198 | 0.5653 ± 0.0181 | 0.9617 ± 0.0085 | 0.9564 ± 0.0052 | 0.6903 ± 0.0382 | 0.8989 ± 0.0416 | 0.7536 ± 0.0597 | 0.7480 ± 0.0560 |
Table C-2. Shot Pair Consistency on 30-second videos.
| Method | All Subject ↑ | All Scene ↑ | All Semantic ↑ | All Avg. ↑ | Revisit Subject ↑ | Revisit Scene ↑ | Revisit Semantic ↑ | Revisit Avg. ↑ |
|---|---|---|---|---|---|---|---|---|
| ShotStream | 0.7153 ± 0.0324 | 0.7902 ± 0.0208 | 0.6896 ± 0.0400 | 0.7317 ± 0.0283 | 0.7079 ± 0.0418 | 0.8100 ± 0.0290 | 0.6809 ± 0.0402 | 0.7329 ± 0.0385 |
| CineMem (Ours) | 0.7529 ± 0.0413 | 0.8167 ± 0.0207 | 0.7042 ± 0.0333 | 0.7579 ± 0.0286 | 0.7841 ± 0.0404 | 0.8791 ± 0.0265 | 0.7680 ± 0.0436 | 0.8104 ± 0.0398 |
Table D. User study results (80 prompts). Scores are reported as mean ± standard deviation (SD) on a 1–5 scale. Higher is better.
| Method | Overall Video Quality ↑ | Intra-shot Consistency ↑ | Transition Coherence ↑ | Revisit Capability ↑ |
|---|---|---|---|---|
| ShotStream | 3.25 ± 0.44 | 3.45 ± 0.52 | 3.43 ± 0.50 | 2.53 ± 0.56 |
| Self-Forcing | 2.05 ± 0.48 | 2.90 ± 0.36 | 1.40 ± 0.49 | 1.37 ± 0.49 |
| Infinity-RoPE | 2.52 ± 0.65 | 3.70 ± 0.46 | 1.25 ± 0.44 | 1.27 ± 0.45 |
| LongLive2 | 3.48 ± 0.55 | 3.67 ± 0.50 | 2.87 ± 0.60 | 3.13 ± 0.57 |
| MemFlow | 2.93 ± 0.51 | 3.27 ± 0.45 | 1.27 ± 0.45 | 1.15 ± 0.36 |
| HoloCine | 3.22 ± 0.88 | 3.60 ± 0.79 | 3.92 ± 0.89 | 3.62 ± 0.49 |
| MultiShotMaster | 3.90 ± 0.65 | 3.82 ± 0.50 | 4.01 ± 0.63 | 4.05 ± 0.40 |
| CineMem (Ours) | 4.12 ± 0.32 | 4.15 ± 0.37 | 4.55 ± 0.50 | 4.43 ± 0.50 |
Response to Reviewer zE4D (1/9)
We sincerely appreciate the reviewer's thorough and insightful comments, which have helped us further improve our manuscript. Following your suggestions, we conducted additional experiments and revised the manuscript to strengthen its completeness and overall quality.
We thank the reviewer for this valuable suggestion to strengthen the evaluation of long-range revisitation. Following this recommendation, we conducted additional quantitative evaluations on 30-second videos and introduced a more targeted evaluation protocol for non-adjacent revisit pairs, with results stratified by temporal gaps. We also refined the human study to separately assess revisit capability and updated the main quantitative results with 95% confidence intervals. These additions are reported in Tables 1, 4, and 9 of the revised manuscript.
Table A. Quantitative results of Shot Pair Consistency. Updated results are included in Table 1 with the corresponding 95% confidence interval in the revised manuscriptt.
We apologize for the suboptimal formatting due to the character limit.