The latest release, Animagine XL 3.1, is an open-source anime-themed text-to-image model that has enhanced its understanding of a wide range of anime works and styles through upgrades and optimizations. This new version addresses the overexposure issue, introduces aesthetic tags, and updates quality and year tags, resulting in images that more closely meet user needs. By adopting a tag sorting method, the accuracy of the generated results has been improved. The training process utilized 2x A100 80GB GPUs for approximately 350 hours. The pre-training phase involved a dataset of 870,000 ordered and labeled images, providing the model with a robust knowledge base. This model focuses on generating anime-style images, optimizing hand anatomy, image detail quality, and the ability to parse prompts and understand concepts in the output results.