Skip to content

Dataset · Rapidata

Comparison of 11 image generators on banana prompts

Rapidata/bananamark-dataset

A dataset of human pairwise evaluations (2,161 rows) for banana images from 11 generative models.

What's inside

Contains 2,161 rows in parquet format with images and English text prompts, alongside 20,088 individual human evaluations. The authors tested 11 image generation models (including Google's Nano Banana, FLUX 2, and Seedream 4) on banana-themed prompts and collected preferences via the Rapidata Python API.

What you can do with it

  • Training reward models and aligning image generators via RLHF
  • Comparing visual generation quality of different models on narrow prompts
  • Training aesthetics and image-text alignment classifiers
Tasks
Text → image, Image → text, Image classification, Reinforcement learning
Languages
English
Size
1K–10K rows
Downloads this month
216
Likes
35

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("Rapidata/bananamark-dataset")