Datasets:
Formal Protest Regarding Serious Uncredited Overlap with CoVEBench
We are the authors of CoVEBench: Can Video Editing Models Handle Complex Instructions?
CoVEBench: https://arxiv.org/abs/2606.08415
First released: June 7, 2026
CoinVE-200K: https://arxiv.org/abs/2608.17566
Released: August 2026
We raise a formal and serious concern regarding the substantial conceptual overlap between our prior work, CoVEBench, and the subsequently released CoinVE-200K / CoinVE-Bench, without appropriate acknowledgment of CoVEBench.
CoVEBench had already explicitly introduced and systematically studied:
Compositional instruction-guided video editing as a distinct research problem;
Complex instructions composed of multiple atomic editing operations;
The need to execute multiple edits jointly while preserving unrelated or irrelevant content;
A dedicated benchmark for evaluating compositional video editing;
Fine-grained, checklist-based evaluation of instruction following, editing realism, content preservation, and video quality.
CoinVE-200K was released approximately two months later and adopts strikingly similar:
problem framing and terminology;
motivation contrasting isolated edits with real-world compositional requests;
decomposition of instructions into atomic editing operations;
emphasis on multi-intent execution and preservation of irrelevant content;
benchmark objectives and evaluation dimensions;
even the highly similar naming pattern CoVE / CoinVE.
Nevertheless, the current CoinVE-200K manuscript appears not to cite or discuss CoVEBench. It instead describes this direction as “under-explored” and presents CoinVE-Bench without clearly acknowledging the directly relevant prior benchmark.
This is not a minor bibliographic omission. Failure to acknowledge directly preceding work can erase academic priority, mislead readers about the novelty of a contribution, and distort the scholarly record.
We therefore formally request that the CoinVE-200K authors:
Publicly acknowledge CoVEBench as directly relevant prior work;
Add a clear and prominent citation and comparison in the abstract, introduction, related work, and benchmark sections;
Correct or qualify novelty and priority claims concerning compositional instruction-guided video editing and its benchmarking;
Explain the independent development timeline of CoinVE-Bench and its relationship to CoVEBench;
Release sufficient benchmark construction and evaluation details to permit an objective comparison.
Until these concerns are transparently addressed, we ask the research community, repositories, reviewers, and platform maintainers to withhold endorsement and promotion of the disputed novelty claims and to treat those claims with appropriate scrutiny.
We welcome an evidence-based public response. We do not encourage personal attacks or harassment. Our objection concerns academic attribution, research integrity, and the accurate preservation of scholarly priority.
Research progress depends on building upon prior work openly—not erasing it.
— The CoVEBench authors
Thank you very much for bringing this concern to our attention and for providing the detailed comparison between CoinVE-Bench and CoVEBench.
First of all, we sincerely acknowledge the importance and value of CoVEBench: Can Video Editing Models Follow Compositional Instructions? We recognize that CoVEBench has made an important contribution to the community by introducing a systematic benchmark for evaluating whether video editing models can follow compositional instructions. In particular, its checklist-driven evaluation framework, structured annotation protocol, and fine-grained diagnostic evaluation perspective have provided valuable insights for subsequent research on compositional video editing evaluation.
We would also like to clarify that our work was indeed inspired by CoVEBench. The omission of the citation and discussion of CoVEBench in the current version was an oversight on our part. We sincerely apologize for this mistake. We fully agree that prior related work should be properly acknowledged, especially when it has influenced the design and positioning of a new benchmark.
To address this issue properly, we will take the following actions as soon as possible:
- Add the citation to CoVEBench in the related work, benchmark design, and evaluation sections where appropriate.
- Explicitly acknowledge the influence of CoVEBench on our benchmark design, especially regarding the checklist-based evaluation framework and the compositional instruction evaluation setting.
- Clarify the relationship between CoVEBench and CoinVE-Bench, including both the shared motivations and the differences or extensions introduced in our work.
- Revise the paper accordingly in the next version, including the main text, references, and any relevant appendix materials.
- Update the project page, benchmark documentation, and released materials to include the proper citation and discussion.
- Review our checklist files and annotation descriptions to ensure that the provenance and design rationale are presented clearly and transparently.
- Make sure future submissions and public versions of CoinVE-Bench properly cite and discuss CoVEBench.
Again, we sincerely apologize for the missing citation and any confusion this may have caused. We greatly appreciate your careful reading and constructive feedback, and we have great respect for the contribution that CoVEBench has made to the community.
Please feel free to let us know if there is a preferred citation format or any specific point you would like us to clarify in the revised version. We will make the corresponding revisions promptly and carefully.
Update: We have now updated the README files of CoinVE-Bench on both GitHub and Hugging Face to include the citation and acknowledgment of CoVEBench.
We would like to again express our sincere appreciation for the contribution of CoVEBench to the compositional video editing evaluation community. Its benchmark design and checklist-based evaluation methodology have provided important inspiration for our work.
Due to the recent slowdown of our project progress, the corresponding arXiv revision has not yet been completed. We plan to update the arXiv version by next week, where we will further clarify the relationship between CoinVE-Bench and CoVEBench, explicitly acknowledge the influence of CoVEBench, and provide a more detailed discussion of the similarities, differences, and extensions in our benchmark design.
We will keep this discussion updated once the revised arXiv version is available.
Update — August 25, 2026
We would like to share a follow-up update: we have explicitly acknowledged the contribution of CoVEBench in the revised arXiv version of our paper, particularly in the Introduction and Benchmark Design sections.
The revised manuscript clarifies CoVEBench as important prior work for compositional instruction-following evaluation in video editing, acknowledges its influence on the motivation and design of CoinVE-Bench, and further discusses the relationship between the two benchmarks, including their shared goals, differences, and extensions.
The updated arXiv version has already been submitted and is currently under moderation/review by arXiv. Once it becomes publicly available, we will update this discussion with the corresponding link.
We sincerely appreciate your feedback and again thank the authors of CoVEBench for their important contribution to the community.
The arXiv has been updated in http://arxiv.org/abs/2608.17566