Skip to content

Dataset · UCSC-VLAA

One million triplets for image editing

UCSC-VLAA/GPT-Image-Edit-1.5M

Source image, instruction, and edited image pairs for training and evaluating image-to-image models.

What's inside

The dataset contains image editing triplets: an input image, a text instruction, and an editing result. Edits are generated using GPT-Image-1, and a reproducible quality control pipeline is applied to filter usable pairs. The instruction language is English; the scale is reported as approximately one million examples with a 200-example audit split.

What you can do with it

  • Train image-to-image models or LoRA using text instructions
  • Evaluate quality and instruction-following in image editing
  • Tune quality filters for image-instruction pairs
Tasks
Image → image
Languages
English
Size
1M–10M rows
Downloads this month
16.5K
Likes
97

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("UCSC-VLAA/GPT-Image-Edit-1.5M")