A massive dataset of approximately 28 trillion pixels, known as the Giant Permissive Image Corpus (GPIC), has been introduced to facilitate scalable visual generative modeling. This corpus consists of a diverse range of internet images, each captioned by a state-of-the-art vision-language model, and is divided into 100 million training examples, 200,000 validation examples, and 1 million test examples. Notably, all images in the GPIC are permissively licensed, allowing for unrestricted use. The creation of GPIC addresses the need for large, accessible, and stable datasets in the field of visual generative modeling1. The availability of such a vast dataset is expected to significantly impact the development of visual generation models, enabling more accurate and efficient training. This matters to practitioners because access to large, permissively licensed datasets like GPIC can accelerate advancements in visual generative modeling, ultimately driving innovation in various applications.