Insight for: 脚本上没有提供prompt.json的参数，也没有设置存放的位置，只有example文件夹，那怎么去推理

This issue exposes a critical gap in the 'Lance' model's usability: the absence of a clear mechanism for providing `prompt.json` during inference. Despite providing an `inference_lance.sh` script with various parameters for video generation, the fundamental input method for prompts is missing or undocumented. This directly impedes users from performing core text-to-video tasks, rendering the model's advanced capabilities inaccessible. For a 3B-parameter multimodal model, a functional and intuitive inference pipeline is non-negotiable for adoption and practical application. This indicates a significant oversight in developer experience and product completeness, hindering market utility.