该lora使用了纳米香蕉蒸馏出来的3d图片进行训练，同时加入了自然语言和tag混合训练，得到更好的画面遵循度。
This LoRa algorithm was trained using 3D images distilled from nano-bananas, and also incorporated natural language processing and tag-based training to achieve better image conformance.