How to Make a Story Dossier
===

Stacey: [00:00:00] Welcome to our class on how to make a story dossier. In this first introductory lesson, we're going to go over what a story dossier is and why this is a foundational skill if you want to learn how to write with ai. Now, if your goal with AI is only to use it for marketing or maybe. Jazzing up descriptions here or there, or only writing short snippets of prose.

You might not need to learn how to write story dossiers, but if you want to use AI for major chunks of brainstorming or for writing any longer passages of prose, then story dossiers are a foundational or a fundamental skill that you need to know. So. To learn what a story dossier is and more importantly why you need to know it.

I'm gonna take you on a quick history [00:01:00] lesson of the past few years with AI and how we got to this point and writing with ai. So if you want to write prose with ai, obviously you can't write a well, obviously at this point in time. You can't write a full novel in one go. The best you can do is either seen by scene or chapter by chapter generally.

Again, depending on your genre or whatnot, some people might be able to get maybe two chapters at a time, but most people feel comfortable seen by scene, chapter by chapter, episode by episode. If you're writing in a serialized format, however you name that,

Just for the ease of vocabulary. I'm just going to call it chapter by chapter in this lesson, but it's about the same concept. So generally, most of us write with AI on a chapter by chapter basis. In the early days of ai, you know, way back in 20 [00:02:00] 21, 20 22, even the biggest, best, most sophisticated models only had a 2000 to 4,000, token context window.

I. If you don't understand what that means, please go back to our free basics class. Understanding context. Windows is fundamental for moving forward in any of these courses, but, because of those small context windows, we can only work beat by beat. Essentially, we had to work in very small chunks in order to write a chapter, you could string those beats together to get out a chapter. You could string those chapters together to get out a book. At the time, that was mind blowing. The cohesion between all those strung together, little bits wasn't the best, but again, we had AI writing for us, so that was amazing at the time.

And this method is technically still [00:03:00] viable. You can still use it, but . I'm going to use a car metaphor here. If you had the most souped up car, a Bugatti, an F1 race car, and rather than turning on the engine, you just put it in neutral and have a bunch of your friends pushing it. That is essentially what you're using.

Today's more sophisticated models, you have way better models, and you're using a less sophisticated method, so therefore you're just not taking advantage of them. You are pushing a car in idle that could go so much faster, about the time GPT-3 0.5 16 K came out, 16 K gave us 16 K context window, and that just blew our minds at the time.

We had more flexibility that we could start writing entire chapters. That gave us flexibility to put more tokens on the input or the prompt, and get more tokens on the output or pros [00:04:00] written. And that's when we started writing these mega prompts, super prompts. A couple different names came about at the time, and that replaced the beat by beat prompting.

So you could give the AI a list of your chapters, settings, outlines, writing styles, all this information, and have it write an entire chapter for you. This was great. So you could have it write chapter one, then give it the same prompt, the same mega prompt. Write chapter two, write chapter three, write chapter four.

Wow. Amazing. Within 25 prompts, you could have 25 chapters and a full novel. , if your Bugatti or F1 race car is today's models, instead of putting the car in neutral and pushing it, him now hooked it up to a tractor.

So it's, it's cruising along and it's a lot less work, but it's still cumbersome as a method [00:05:00] with this mega prompting, super prompting, because some of the drawbacks of this, the chapters strung together in this way, were still kind of clunky. There was still so cohesion, but the biggest issue was bleed over.

Essentially by giving it every character in the book, the entire outline. What happened was you gave it all the information about the book, so it tried to put every character, every plot point in every chapter. So if the. Mafia Boss had a Homecooked pasta dinner at Nona's house in chapter four, and then a gun shootout with the rival mafia boss in chapter 16.

Suddenly he got shot in the shoulder and is talking about, you know, pasta sauce at the same time, which makes no sense. Or the killer [00:06:00] comes waltzing in in chapter five and announces himself in a your cozy mystery because the AI knows who the killer was and just can't help announcing it. And even if he doesn't announce it,

it is so ridiculously obvious that the murder murderer comes in with his shifty murderer eyes and has the knife practically hanging out of his pocket. So there was bleed over with the chapters. The AI cannot hide information very well. So that was the problem with these mega prompt super prompts. So as the context windows got bigger, the models got more sophisticated, we moved on to these chapter briefs, writing briefs.

Essentially this method was to only give the AI the information it needed for that chapter. So instead of giving it every character, every plot point, every everything, [00:07:00] you only gave it the characters for this chapter, the plot points for this chapter, just the information it needs to know.

Now we needed bigger context windows for that because chapter briefs usually have to be longer. When you get to chapter 16, , you have to give it a lot more context.

You can't just tell it what happens in chapter 16. You have to lead up to here's what happened. Here's what happens in this chapter, but here's why. Here was all the false red herrings. Here was the emotional buildup that led to this payoff. 

Even then there were some issues. The biggest drawback of this method is cohesion. Because each chapter's kind of being run as a silo or the AI doesn't have the story as a whole,, there is a bit of a disconnect. So in terms of the authorial voice or the writing style, it [00:08:00] might not match between chapter two and chapter 20. Character voices might not sound exactly the same, especially with multiple points of view characters. You start having four POVs, 

the POV character from chapter one might not sound the same, or have the same dialogue style when they're talking in chapter 16 because you flipped POV so many times, they're just no longer distinct, emotional payoffs or the emotional arcs might start getting a little muddier than you would like.

Again, it is still massive improvements over past methods. So you've actually turned on the engine of your car this time, but there's water in the tank. It's sputtering. It's not hitting its maximum. It's not living up to its potential as a supercar.

So that's where a story dossier [00:09:00] comes in. Now for the story dossiers, you do need larger context windows. So you cannot use these smaller, cheaper models in something like Open Router.

So a few things to keep in mind. You will need a model with a larger context window. At least a hundred thousand tokens or more is a good rule of thumb or more is better and be prepared. Because you will be utilizing more of the context window. There will be a higher cost. So so take a look at the cost, but the results will be substantially better.

So for example, if I used clawed. Claude Haiku is the cheapest.

It still has a high context window. It would work with the story dossier perfectly well, [00:10:00] and because I'm using a good portion of that context window, if I used Haiku, it would be the cheapest. However, I still prefer sonnet even though it costs a little more,

I get substantially better results out of it. So for me, the time I save in editing between using sonnet and using Haiku is well worth the AI cost. However, there is a substantial jump in price between sonnet and Opus, and I don't feel the change in quality is enough to justify that change in price. So test it out for yourself.

these story dossiers are going to be large, therefore, there is a higher cost involved. But again, it is your time versus the money. For the story dossier itself. I tend to [00:11:00] prefer, one of the Open AI models right now, one of the thinking models. 

And we'll get into that as we start running. And I'll go through some of the pricing. But what is a story dossier? A story dossier goes much, much deeper into all of these aspects. It's not just a list of characters, but dialogue style.

The specifics of the genre, it is tropes, it is key points, mandatory scenes. This can be several thousand words. Then it will also include several chapters of your story depending on how much you're willing to spend. Some people include the entire story. Now that does make it exponentially more expensive as the story progresses.

If you tend to write shorter, like novella's, 10 to 15 chapters, it is [00:12:00] well worth it to include the entire story. So as you write, just keep including the previous story along with the story dossier, and you're only getting up to 20, 25,000 words. Not huge. If you write Epic Fantasy and your story dossier and your story so far is getting up to 150,000 words.

You might only wanna include the previous few chapters, and then a summary of the chapters. Before that, we'll get into the specifics, but again, a story dossier is much more specific and you can also use a story dossier.

If you don't write with ai, this story, Dasia, you can use it, the AI to create it. Of course, we're gonna use the AI to create it. This is very detailed and by hand it would take me quite a while to fill this out. So I use the AI for [00:13:00] help of course, but. If you are not a brainstormer, if you're not an idea person, but you love writing prose, you can use the AI to create this very, very detailed dossier outline, character sketches, everything You need to have the story very, very fleshed out, and then just have your prose shine on top of the idea for the story.

So again, it's where your strength, where your joy in the writing comes from. Your joy is the PS and not the particulars of the ideas. Have the AI help you with this part, or if it is the ideas and not the prose specifically, you can feed your ideas into this dossier. You don't have to have the AI come up with the ideas, but

the story dossier is the format that the AI writes the best prose. So I'm [00:14:00] gonna be very clear here. For the story dossier, you can have the AI help you create the ideas, or you can bring all of your own ideas to the table, and the AI doesn't have to help you come up with a darn thing. It's just helping you organize it.

But either way, the story dossier is what makes the AI write good prose. So with that, we'll jump into actually starting to create a story dossier and put it together piece by piece, and we'll jump into that with the next lesson.