Skip to content

Dataset · Rapidata

Human evaluations for OpenAI 4o image generations

Rapidata/OpenAI-4o_t2i_human_preference

13k images from OpenAI 4o with over 200k real human ratings for training preference models.

What's inside

The dataset in parquet format includes 13,000 text-image pair rows in English and 211,946 ratings from approximately 45,000 raters. The data was collected via the Rapidata API and evaluates OpenAI 4o generations (versions from 03/26/2025) based on three criteria: overall preference, alignment, and prompt adherence.

What you can do with it

  • Training reward models to evaluate text-to-image generation quality
  • RL alignment of text-to-image generators with real user preferences
Tasks
Text → image, Image → text, Image classification, Reinforcement learning
Languages
English
Size
10K–100K rows
Downloads this month
965
Likes
35

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("Rapidata/OpenAI-4o_t2i_human_preference")