Dataset · Yuanshi
200k paired images with subject preservation
Yuanshi/Subjects200K
Paired pictures of the same object in different environments for controllable generation tasks.
What's inside
The dataset contains 200,000 pairs of images with textual data and was introduced as part of the OminiControl project. Each pair maintains consistency of the main object while changing the context and background scene. It includes subdivisions into collections, such as collection_2 for filtering high-quality pairs.
What you can do with it
- Training LoRA and ControlNet adapters for transferring characters or objects into new scenes
- Fine-tuning diffusion models to preserve object identity during generation
- Size
- 100K–1M rows
- Downloads this month
- 637
- Likes
- 49
How to load it
from datasets import load_dataset
ds = load_dataset("Yuanshi/Subjects200K")