本仓库是基于huggingface的示例diffusers/examples/textual_inversion改编而成的Textual Inversion方法，你可以执行MTI_train脚本进行训练。主要改动如下：

• 可自定义提示词，你可以将提示词存储到.json文件，并放入anno文件夹，具体参考anno文件夹的示例。如果不指定提示词，则默认使用原本的模板构建提示词。
• 可自定义嵌入的数量，修改嵌入数量的方法可参考MTI_train.sh文件中的embeddings_number设定。

This repository is an adaptation of Textual Inversion based on the example of huggingface diffusers/examples/textual_inversion. You can execute MTI_train for training. The main changes are as follows:

• Customize the prompt words. You can store prompts in a .json file and put them in the anno folder. For details, see the example in anno folder. If prompts are not specified, the original template will be used to build the prompts by default.
• Customize embeddings number. The method of modifying the number of embeddings can refer to the embeddings_number setting in the MTI_train.sh file.