danielrosehill/Hebrew-Image-Eval-111225

该数据集名为Hebrew Image Generation Evaluation，旨在评估主要文本到图像生成模型在准确渲染希伯来语文本方面的能力。测试了12个模型在两个希伯来语单词（שלום和פירגון）上的表现，包括单词的准确文本、标准希伯来语脚本和从右到左的渲染。结果显示，大多数模型在希伯来语文本渲染方面表现不佳，只有Gemini 3 Pro和Nano Banana Pro能够可靠地处理希伯来语排版。数据集还包含详细的评估标准、测试模型列表、样本图像和仓库结构等信息。

The dataset named Hebrew Image Generation Evaluation evaluates the ability of major text-to-image models to accurately render Hebrew text. It tests 12 models on two Hebrew words (שלום and פירגון), assessing accurate text, standard Hebrew script, and right-to-left rendering. Results show that most models struggle significantly with Hebrew text rendering, with only Gemini 3 Pro and Nano Banana Pro demonstrating reliable Hebrew typography capabilities. The dataset includes detailed evaluation criteria, a list of tested models, sample images, and repository structure.