amazon-agi/Amazon-Nova-2-Omni-Image-Gen-Evals

---
language:
- en
license: cc-by-nc-4.0
task_categories:
- text-to-image
dataset_size: 1000
configs:
- config_name: prompts
data_files:
- split: latest
path: prompts.jsonl
---

# Nova 2.0 Omni Image Generation — Evaluation Prompt Set

This dataset contains the evaluation testing prompts used for benchmarking Nova 2.0 Omni Image Generation models. It includes 1,000 prompts organized around a focused taxonomy designed to capture the most relevant capability areas for evaluating our T2I capability.

## Overview

This testing set is organized around a focused taxonomy that captures the most relevant capability areas for evaluating modern text to image systems, including human depiction, typography, physics understanding, nature and landscape, as well as skillsets for testing compositional and spatial reasoning. These categories were selected for their practical diagnostic value and the range of visual reasoning skills they represent, with some prompts introducing subtle variations in environment, action, lighting, or style while keeping the core subject consistent. These controlled variations reflect the diversity of real user intent and gently probe whether models handle such nuances reliably. The set includes a mix of image styles such as photorealistic, cinematic, cartoon and other artistic formats to reflect common use cases and aesthetic preferences. The overall design aims to balance realistic image quality, which is central to industry-grade image generation, with coverage of instruction following and concept understanding for more abstract prompts.