Bananamark: Human Preferences for Banana Image Generation Across 11 Models

Rapidata's Bananamark dataset contains over 20,000 human preference annotations comparing banana images generated by 11 different AI models. The data was collected via the Rapidata Python API, where annotators chose which image had better bananas for various banana-themed prompts. The dataset was last updated on December 12, 2025.

Use Cases

• Benchmarking image generation models based on human preference data mentioned in the description
• Training or fine-tuning reward models for generative AI based on comparative annotations
• Analyzing the performance of specific models like Nano Banana, FLUX 2, or Seedream 4 on object-specific generation tasks
• Studying human perception and preference in AI-generated visual content

Strengths

• Over 20,000 human preference annotations provide a substantial basis for evaluation
• Compares outputs from 11 distinct image generation models
• Focuses on a specific, well-defined object (bananas) for controlled comparison

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment
• Data may reflect bias inherent to the specific prompts and annotator pool used

Provenance

Source

Rapidata

Collection Method

Human preferences collected via the Rapidata Python API for images generated from banana-themed prompts

Freshness

Last updated 2025-12-12 10:05:23; freshness should be verified