Dataset · Hugging Face

songyiren/Ref10K

Name: Ref10K
Creator: songyiren
License: cc-by-nc-4.0
Keywords: task_categories:text-to-video, language:en, language:zh, license:cc-by-nc-4.0, size_categories:10K<n<100K, format:json, modality:image, modality:text, modality:video, library:datasets, library:pandas, library:polars, library:mlcroissant, region:us, video-generation

Ref10K Ref10K is a set of about 10K reference-to-audio-video (Ref2VA) inputs for MiniMax-H3: reference images, videos and audio, each paired with a rewritten instruction and a six-section H3 prompt. It was built to distill a few-step student (Anyflow2) from the H3 Ref2VA teacher without paired target videos. Every record is an input only; there is no ground-truth output. Split Records File train 9,987 train.jsonl validation 491 val.jsonl (split by source group and… See the full description on the dataset page: https://huggingface.co/datasets/songyiren/Ref10K.

View on Hugging FaceReport an issue

Author / organization	songyiren
License (as declared)	cc-by-nc-4.0
Downloads	158 (as reported by Hugging Face)
Likes	1
Created	12 October 2026
Last updated at source	12 October 2026

video-generationreference-to-videovideo-editingaudio-visualminimax-h3ref2va

Embed this entry

<iframe src="https://silytics.si/embed/item/881e75b6-77af-41bf-8f50-98788a18b293" width="420" height="190" style="border:0" loading="lazy" title="Silytics entry"></iframe>

Free, no tracking. More about feeds and embeds.

Popularity, last 2 days

158 now (+10 since 2026-10-16). Counts as reported by the source, snapshotted daily by Silytics.

More from songyiren

songyiren/visual-reasoning-benchmark-resultsDataset
songyiren/RewardDataset
songyiren/Render-testDataset

Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 17 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.