Skip to content

Dataset · OmniGen2

Multimodal Image Generation and Editing Dataset

OmniGen2/X2I2

Millions of English and Chinese examples for training image editing and generation models.

What's inside

A dataset ranging from 1 to 10 million records for text-to-image, image-to-image, and any-to-any tasks, including video editing and inpainting. Metadata is stored in jsonl format in English and Chinese, and images are packaged in multi-volume tar.gz archives.

What you can do with it

  • Training universal instruction-based image generation and editing models
  • Training LoRA for local editing of images and video frames
Tasks
Image → image, Text → image, Multimodal
Languages
English, Chinese
Size
1M–10M rows
Downloads this month
5.7K
Likes
41

How to load it

from datasets import load_dataset

ds = load_dataset("OmniGen2/X2I2")