Submit
The source packs are not distributed. Fill in the form and the maintainers will get in touch about how they will run your agent on the benchmark internally.
What a run covers
| Collection | Inputs | Also applies |
|---|---|---|
| EditStock11 tasks | 1,707 files99.7 GB | EditStock’s credit applies: any still from these projects in a paper names the project title, the filmmaker and EditStock. |
| Cinestudy15 tasks | 407 files155.6 GB | Cinestudy’s credits and hashtags apply to anything you show from these projects. |
| UGC15 tasks | 1,080 files163.1 GB | Presenter-led ads with identifiable people: do not use them to identify, profile or imitate anyone. |
| Commercial15 tasks | 1,809 files254.3 GB | Some packs show identifiable people: do not use them to identify, profile or imitate anyone. |
| All56 tasks | 5,003 files672.7 GB | Some music and sound effects come from free music libraries such as Pixabay, whose terms continue to apply. |
Access terms (summary)
- Files are not distributedThe source packs stay with the maintainers. Filling in this form does not grant a download, a URL or a token for the benchmark files.
- We run your agentAfter you submit the form, the maintainers get in touch about how they will run your agent on the benchmark internally.
- Research and evaluation onlyRuns are for evaluating agents on TimelineBench tasks, scoring outputs with the verifier, and reporting aggregate and per-task results.
- No trainingDo not train generative models on the inputs or on anything derived from them (features, transcripts, embeddings), and do not train or tune on the benchmark.
- Cite the benchmarkCite TimelineBench in any work that reports results from a run.
- Respect the people shownMany packs show identifiable people. Do not use the media to identify, profile or imitate them.
- Results stay privateOutput videos stay with the maintainers. They are not published in a pull request, issue or public bucket.
- Never distributedThe professional cuts and the paper’s agent outputs are never distributed, and the rubrics’ paperwork quotes for EditStock and Cinestudy (111 items) stay withheld.
A plain-language summary, not legal advice; the repository's MEDIA_TERMS.md still governs any media the maintainers hold.
How it works
- 01 Fill in the form: who you are, the agent you want evaluated, and which collections the run should cover. Accept the terms.
- 02 The maintainers review the request and get in touch about how they will run your agent on the benchmark internally.
- 03 The source packs are not sent to you. The run happens on the maintainers' side, and every file is checked by size and sha256 before an agent starts. See Inputs & media access.
A problem with a request, or a listed model you want removed? Use the contact form and quote your reference ID.