我不会英文，以下译文来自chatGPT
I don't know English, the following translation is from chatGPT.
Preface
如果你是一个lora训练师，或者想尝试训练自己的模型，那么这篇文章可能会大大提升你的工作效率
If you are a LoRa trainer or want to try training your own model, this article may greatly improve your work efficiency.
如果你仅仅想生产一些图片，而不知道如何把想法通过prompt描述出来，甚至完全没有想法，这篇文章都会对你有所帮助。
If you only want to generate some images and don't know how to describe your ideas through prompts or even have no ideas at all, this article will be helpful to you.
在我尝试的项目中，它曾帮助我节省了非常多的时间和精力。甚至可以很骄傲的说，这可能是市面上你所能找到最简洁最有效率的prompt生成方法！当然，我分享我的技巧只是为了抛砖引玉，随着技术的发展一定会有更优秀的工具、方法出现，我们都在见证历史。
In my project attempts, it has helped me save a significant amount of time and effort. I can even proudly say that this may be the most concise and efficient prompt generation method you can find on the market! Of course, I'm sharing my tips just to spark ideas, and with the advancement of technology, there will surely be even better tools and methods emerging. We are all witnessing history in the making.
Keywords : chatGPT
是的，就是这款世界上最好的大语言模型，很多人用它来聊天、做咨询、读文档，随着GPT4可以上传WEB、PDF、IMG之后，我意识到，这将开启非常多的可能性，其中就包括为上传的图片生成prompt。
Yes, it's indeed one of the world's most advanced large language models. Many people use it for chatting, consulting, and document reading. With the introduction of GPT-4's capability to process web content, PDFs, and images, it opens up a wide range of possibilities, including generating prompts for uploaded images.
How did I do it ?
首先我上传了一张图片，让它读取其中的信息，同时给了它一些prompt的常用规则，比如标点符号，要求它使用中英文回复，这样更方便检查。
First, I uploaded an image, allowing it to read the information within. I also provided it with some common prompt rules, such as punctuation, and asked it to reply using both Chinese and English to make it easier to review.
The LoRa link
还不赖，但离我们的目标还有段距离，没关系，让我们来修正它。
这里我描述了我所想要的prompt的文本结构（个人习惯因人而异），同时希望它使用简短、清晰且客观的词语来描述
Not bad, but we're still a bit away from our goal. That's okay; let's work on refining it.
Here, I described the text structure I want for the prompt (personal preferences may vary), and I also hope it uses concise, clear, and objective language for description.
开始好一点了，这已经比大多数打标工具要好了，不是吗。但是有经验训练师不会止步于此
我要求他参考Danbooru/Gelbooru特定数据集的著名标签，且在意思相近的prompt中优先使用Danbooru/Gelbooru的著名标签来标注图像
It's getting better, and it's already better than most labeling tools, isn't it? But experienced trainers won't stop here.
I asked it to refer to the well-known tags from the Danbooru/Gelbooru specific dataset and to prioritize using these well-known tags in labeling images in similar prompts.
太棒了，这已经大大超出我所能撰写prompt的能力范围，当然，让我们再增加一点难度，我个人习惯是在文本的开头部分描述画面风格和画面主题，现在让它来实现这一点。
Great, this already goes far beyond my own ability to craft prompts. Of course, let's add a bit more complexity. My personal preference is to describe the visual style and theme of the scene in the beginning of the text. Now, let's have it do that.
完美！至少对我来说，我真的越来越喜欢GPT这个家伙啦！
现在我让它来总结上述的所有规则，并梳理出一套文本，以便日后使用
Perfect! At least for me, I'm really starting to like this GPT guy more and more!
Now, I'll have it summarize all the rules mentioned above and put together a set of text for future use.
看，这样以后再需要为图片打标时，只需要把这套文本给到GPT就可以啦！
See, in the future, when I need to label images, I can just provide this set of text to GPT, and it will do the job!
下图是使用最终打标结果在SD中生成的图像，能看出跟原图的差距并不大（除了粉色爆炸的烟雾，因为那是我的一个lora特效），这样你就完成了一次非常成功的打标
The image below is generated in SD using the final labeling result. It can be seen that the difference from the original image is not significant (except for the pink exploding smoke, which is one of my lora effects). This already constitutes a very successful labeling.
The LoRa link
Epilogue
在撰写这篇文章的时候，我得知GPTs的问世，未来还将有更多GPT的app出现，AI的发展实在太快啦就像龙卷风。我也在尝试利用它来制作一个自己的prompt robot，它现在还有些问题，我还在不断调试，当时机成熟时，也许可以把它分享出来，让更多的人从中受益。
好啦，这就是今天分享的全部内容，不得不说在调试GPT的过程中确实有很多问题，我简化了其中的一些步骤，但大体的思路是不变的，而一旦调试成功，恭喜你，你将获得一个前所未有的聪明简便的打标工具！现在开始享受它吧！
While writing this article, I learned about the emergence of GPTs and there will be more GPT-based apps in the future, which is simply fantastic, isn't it? I'm also trying to use it to create my own prompt robot. Right now, it's still a work in progress, and it doesn't always cooperate as expected. I'm constantly fine-tuning it. When the time is right, I think I might share it with others to benefit more people.
Well, that's the sharing for today. I have to admit that there were many challenges in debugging GPT. I simplified some of the steps here, but the overall logic remains unchanged. Once you successfully debug it, congratulations, you'll have an unprecedented, smart, and convenient labeling tool at your disposal!