‍

1. Get clear on your goals and current workflow

First up, think about your endgame with this library.  Are you doing it to keep tabs on all those fancy advanced ChatGPT prompts or to hoard your favorite image styles? 

Now, think about your personal style with these prompts. How are you using the tool, and more importantly, what is working well for you? 

For image styles, my library is a mix of the things I work on—mostly patterns (I make clothes) and storyboards (I write science fiction). I've cataloged art movements and color palettes I like for patterns, and design, photography and lighting-related terms for storyboards.  I also have several that I'm still actively experimenting with and testing their performance and functionality on a standard set of subjects. Yours should reflect your projects and interests.

For text prompts, figure out if you're more of a “Centaur” or “Cyborg” prompter. Don't know which is which? Centaurs split tasks between themselves and AI, while Cyborgs are all about a lot of back-and-forth with the tool—less prompt engineering, more prompt repeating.

As a Centaur with ChatGPT, I go for long, detailed prompts. That's different from a Cyborg style, where you'd go for a series of follow-up prompts to add to the output. My faves? The “Flipped Interaction”—where ChatGPT asks and you answer—and the “Persona Pattern” where ChatGPT takes on a specific persona for the interaction. But hey, that's me. Your style, your rules.

‍

2. Look at other people’s prompts

Even if you're pretty good at prompting already, it's always a good idea to see what other people are doing and try it for yourself. This isn't the SATs, after all. 

Word of warning though: although there are lots of prompt libraries available, make sure you test them first—half of the prompt libraries I see look like someone just put "list 10 prompts for a prompt library" into ChatGPT. Obviously that's not going to get you what you want.

What works better, I've found, is following groups where people share and discuss prompt techniques that do work. Look for specialty groups on Reddit, Facebook and Discord, and, of course, you can never go wrong with the official groups for the tools you use a lot. Talk to your friends and colleagues about how they work with the tool, too.  And remember, while an AI chatbot can spit out a prompt, it might be worse than one you design yourself – especially for image generation.

When you review other peoples' prompts, take an analytical approach. Did they get a good result? Add that trick to your arsenal.  

I'll give you an example: I recently went through a colleague's ChatGPT history where he built a specific technical component from the ground up. Turns out, he had a huge difference in the order of his workflow compared to mine – while I focus on one part and complete it before moving on to the next, he went back and forth between several tasks simultaneously. You don't know what you don't know.

Finally, consider keeping a section in your library just for inspiration – I've got a collection of Midjourney images I liked that I might want to riff off in the future.

3. Learn the tool’s nuances

Here's where it gets weird.  We've only really had the ability to create advanced prompts for about a year, and in that time researchers and hobbyists alike have found all kinds of odd things that help give better results. 

Here are a few quirky techniques that are surprisingly effective:

• If you ask the LLM to think about it step-by-step, it helps the LLM reason better.
• Adding emotional comments like "This is very important to my career" improves accuracy.
• Don't bother with word count. LLMs understand things in tokens, not words, so any time you ask for a specific word count it's just not going to listen. 

Often, these depend on the models you're using, so keep up to date on what new things people are discovering.

4. Save more than just prompts

If you're just starting out, don't stress too much. You'll probably start by dumping a bunch of stuff into a big word doc. That's totally fine.

But here are some more things you should consider saving:

1. TL;DRs: The most useful thing I've found to add to your prompt library isn't even a prompt. It's a project tl;dr (aka, a summary) to guide the LLM at the beginning of your work. Think of these as your project cheat sheets: a 100–200 word rundown that sets the stage for your work with the AI.  Tip: If you find yourself leaning on these tl;drs a lot, why not level them up into agents?
2. Terms and Styles: The nice thing about prompting is that it teaches you the vocabulary needed to describe various writing or art styles. But for us mere humans, the sheer volume of styles can be totally overwhelming. Save styles you like along with a short description in your library so you don't mix up your Bauhaus with your Brocade. Tip: ChatGPT does a pretty good job of describing obscure styles you can put in your library.
3. Your Voice: If you want to use ChatGPT to write drafts, it's best to steer it with a description of a particular voice to lighten your editing load. You can also set custom instructions if you want to have it always write in a specific voice (although if you're working on a lot of different projects you might not want this).
4. Templates: Especially if you're a heavy user of advanced prompts, having a template that you can fill in is a handy time saver. Here are five advanced prompt templates to get you started. 
5. Metadata: If, like me, you use a lot of different AI models, you probably want to record at least basic metadata to help you remember the settings or specifics of the tools the prompts are for. That way, you can reroll them from scratch if you need to.

‍

‍

5. Keep it organized

The next question is: where do you keep all this? At first, a simple notes app might suffice. But eventually you'll hit a moment where it all gets unwieldy and hard to find. Then you have to think about the best way to organize it. 

Your organizational system depends on your workflow. If you're working on a lot of different projects, each of which have a series of prompts you regularly use, you might want to create a prompt book for each one and add it to a bigger library. Bonus: this makes it easier to share with colleagues if you need to, too.

I personally organize my library in a series of files dedicated to each part of the project with huge searchable headings separating each set of prompts so I can scroll through it fast or search for keywords. The point is to be able to easily find the prompts again when you need them in the most painless way possible. You can also add tags or keywords to make them easier to find.

It's best to think of your prompt library as a living document and keep adding to it as you go. If you leave it too long, it's a massive time-waster to try and go back through all your ChatGPT conversations or Midjourney renders to try to find something specific. Remember, your prompt library is like a garden—it needs regular tending. Don't let it become a jungle where you can't find anything. (Rich advice from someone who killed a cactus through neglect.)

Conclusion

Setting up your own prompt library is a bit of work, but trust me, it's totally worth it if you're regularly using AI tools. Think of it as your personal artificial intelligence toolkit that gets better the more you use it. Just start by throwing in prompts that clicked for you, and keep tweaking it as you go. And don't forget to have some fun with it and add some inspiration—after all, it's your own creative space in the AI world.