Skip to content

Dataset · QingyanBai

Synthetic dataset for text-based video editing

QingyanBai/Ditto-1M

Large dataset of synthetic video-instruction pairs in English for training video editing models.

What's inside

A large-scale dataset (over 1 trillion units in metadata, million-scale by name) with synthetic examples of video editing using text instructions in English. Data created using an automatic generation pipeline ensuring high quality and diversity of video modification scenarios.

What you can do with it

  • Training video-to-video models for editing clips via text commands
  • Fine-tuning video diffusion models on complex and diverse edits
  • Researching methods for synthetic generation of training video data
Tasks
Video → video
Languages
English
Size
over 1T rows
Downloads this month
9.5K
Likes
54

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("QingyanBai/Ditto-1M")