Datasetpublic

cuteLyra/minmaxh3-base

MiniMax H3 base video-edit results This dataset contains one canonical MiniMax H3 video-edit result for each CoVEBench task task_0001 through task_0626 (626 videos total). Layout videos/task_XXXX.mp4: canonical result video for task XXXX. manifest.jsonl: task metadata, preprocessing contract, SHA256, and result provenance for all 626 videos. manifest_online_353.jsonl: manifest downloaded from the online reverse worker before merging. merge_report.json: merge… See the full description on the dataset page: https://huggingface.co/datasets/cuteLyra/minmaxh3-base.

sourceHugging Faceupdated 24d agoView on Hugging Face

0likes479downloads

Dataset Card

MiniMax H3 base video-edit results

This dataset contains one canonical MiniMax H3 video-edit result for each CoVEBench task task_0001 through task_0626 (626 videos total).

Layout

—videos/task_XXXX.mp4: canonical result video for task XXXX.
—manifest.jsonl: task metadata, preprocessing contract, SHA256, and result provenance for all 626 videos.
—manifest_online_353.jsonl: manifest downloaded from the online reverse worker before merging.
—merge_report.json: merge counts and hashes for the 45 overlapping local/online results.

Merge policy

The online Hugging Face result is canonical when the task exists there. Local H3 outputs are used to fill task IDs missing from the online snapshot. The final set has no missing task IDs.

All outputs use the current H3 preprocessing contract: 107 frames, 24 FPS, a 1344x768 letterboxed canvas with black padding, followed by output cropping.