LucasFang/FLUX-Reason-6M

FLUX-Reason-6M是一个庞大的文本到图像数据集，包含6百万张高质量的推理式图片，旨在提升生成模型的推理能力。该数据集包含2000万双语描述，并引入了生成思维链（GCoT）提示，详细描述了图像生成的过程。数据集围绕六个关键的推理特征进行组织，包括想象力、实体、文本渲染、风格、情感和构图。同时，还提供了PRISM-Bench基准测试，用于评估模型的推理能力。

FLUX-Reason-6M is a massive text-to-image dataset containing 6 million high-quality reasoning-focused images. It aims to enhance the reasoning capabilities of generative models. The dataset includes 20 million bilingual descriptions and pioneering Generation Chain-of-Thought (GCoT) prompts, which provide detailed step-by-step breakdowns of the image generation process. The dataset is organized around six key reasoning characteristics: Imagination, Entity, Text rendering, Style, Affection, and Composition. Additionally, the associated PRISM-Bench benchmark is introduced to evaluate the reasoning capabilities taught by the dataset.