Free YouTube Transcribe

Video transcript

This Advanced AI Writing System Writes ENTIRE Books (Step by Step Tutorial)

The Nerdy Novelist · 4,440 words · 21 min read

Want to search this transcript, jump the video from any line, or download it as TXT, SRT, or VTT?

Open in the transcript tool

Full transcript

0:00Welcome everybody to another video going

0:02over another N8 automation. And this is

0:04the big one. This is the one I've been

0:06building up to for a number of the last

0:08videos that I've done on automations.

0:10And this is the one that can essentially

0:13create an entire book in just minutes.

0:15Usually about an hour to actually go

0:17through the whole book depending on the

0:19length of your book. Now, I wouldn't

0:20necessarily recommend you do it that

0:23fast, uh, because usually there's some

0:25errors that come in and I tend to go a

0:27chapter at a time or maybe two or three

0:29chapters at a time at most. So,

0:31reviewing it as I go through this

0:33automation, but technically speaking,

0:35you can give it all of your info and

0:37just push a button and it will go

0:39through chapter by chapter and write the

0:41entire thing. If you trust it enough to

0:43be able to do that all the way through

0:45and then you go back through and edit

0:47it, that's fine. Uh, I personally don't

0:49do that. I just go through a couple of

0:50chapters at a time, but I still use this

0:52automation because it still saves me a

0:54huge amount of time. So, let's dive in.

0:56So, the automation in question here is

0:58this particular automation that I've

1:00built. I call it full book automation

1:02colon outline to chapters. And, uh, it's

1:05a great automation. Uh, I have mentioned

1:07it here on the channel before, but I

1:10have updated it a little bit so that it

1:12works a little better. Now, in the past,

1:14I have also shown you these automations.

1:17This one takes all of your brain dump

1:19information and constructs a more

1:21structured dossier for your story about

1:24all the different elements that you

1:25need. And it it's a fun one. I did a

1:28whole video about this one. I did also

1:30another video about this one where you

1:32take that dossier that was just created

1:34in the previous automation. And you run

1:36it through this and it creates some

1:38characters and fully fleshes those out

1:40as well as all of your world world

1:42building elements. Those get all fleshed

1:44out as well. And then you can create a

1:47full outline in this automation as well.

1:49So this is kind of like step one and

1:51step two, right? We go this far and by

1:53the end of this you have a dossier, you

1:55have a character sheet, a world building

1:57sheet and an outline. And for this

1:59automation, this is the natural

2:01continuation of this process where you

2:03take all of those documents and you

2:07create your full book. Okay, so let's

2:09walk through exactly how this works.

2:10First of all, it starts with a form

2:12submission where I just input the name

2:14of the book as well as any additional

2:16notes that I might have, which can be

2:18random. Uh, you know, I don't know

2:21exactly what your needs are going to be.

2:23So, it can be any number of things. So,

2:25I just have a space there for author

2:26notes. And then I have a couple of

2:29documents that we pull in. We get the

2:30pros style guide. And uh I've done

2:33videos about style guides before, but

2:35essentially this is just examples of

2:37your work and uh perhaps even a

2:39structured like you can take samples of

2:42your work and give it to AI and say

2:43develop a style guide based on this pros

2:46that you see and it'll do that. And so I

2:48usually put all of those things together

2:51in one document. And that's my pros

2:53style guide that I pull in here. I also

2:55have a list of forbidden words uh AISMs

2:58and things like that that I wanted to

2:59cut out. So, I have that list that I

3:01pull in. And then we have the three

3:04lists that were created in the previous

3:06step. The character sheet, the world

3:08building sheet, and the outline. And and

3:10so here you can actually see some of

3:12those documents right here. Uh I have uh

3:14for one book that I'm working on. Um

3:17this is my world building document. So

3:19it's uh this was created by AI with a

3:22lot of input from me. So it's basically

3:24just taking what I gave it and sort of

3:26structuring it in a little bit more

3:28organized fashion. and uh it gives me

3:30all of this information. And the same

3:32goes is true for the characters here. So

3:35here's like one character. This is my

3:36main character. This is all the

3:38information we need to know about this

3:39character. Quite a bit of information.

3:41And then I also have an outline. So

3:42here's chapter 1, chapter 2, etc. So

3:47that's what the outline looks like. So

3:49those documents, all three of those are

3:50pulled into this automation via these

3:53nodes right here. And then I have this

3:55node that says get blank first draft

3:57doc. This means that before you start

4:00using this automation, you want to come

4:03in here and actually create a doc where

4:06the first draft is going to go. And it

4:09needs to be blank just to start with

4:11because, you know, obviously we haven't

4:12written the book yet, but we're going to

4:14actually have this one created here. So

4:17now we've got that. We are then ready to

4:19go into the very next step, which is to

4:22find the chapter names in the outline.

4:24So what this one does is it takes a

4:26really low inensity model, looks through

4:28your entire outline, and makes a list of

4:30all of the chapters. The prompt goes

4:33something like this. You are a parser.

4:34Given the above book outline text,

4:36return a JSON array named chapters

4:38containing each chapter title exactly as

4:41it appears, one element per title,

4:43including those without numbers like

4:44prologue and epilog. Make sure the

4:46chapter numbers are also present where

4:47listed, and do not include any

4:49additional keys. Only output valid JSON.

4:52And then you can just leave it at that

4:53if you wanted to do the entire book at

4:55once. But I also add this additional

4:58sentence if I only wanted to work on a

5:00couple of chapters at a time. So I say

5:03in this specific instance, I only want

5:04you to do chapters 1 through three.

5:06Okay. Uh you could also potentially put

5:09this in the initial form and then have

5:11it autopop populate here. That might be

5:13easier for some, but I I don't see that

5:15as being any easier than just coming in

5:17here and modifying this. So in this

5:19case, it will only return chapters one

5:22and through three. But if I remove the

5:24sentence, it would do everything for the

5:26entire book, uh, the entire outline. So

5:29once it has that list of chapters that

5:31it's going to be working on, it parses

5:32[clears throat] them out with this

5:33little code node and then starts a loop

5:36where the first thing that happens in

5:38the loop is it gets the last 2,000 words

5:41from the first draft document. Now, if

5:44we're writing chapter 1, there is no

5:47past 2,000 words. So, it'll just bring

5:50in nothing here. But assuming we're

5:52writing chapter 2 or chapter 3, I want

5:54it to have the context of the last 2,000

5:56words or so, so it knows where it left

5:59off and where to begin. I don't do more

6:02than 2,000 words because that

6:04overextends the AI a little bit. It

6:07already knows from the outline. it'll

6:09have access to the full outline so it

6:10knows what the story is that happened

6:12before. I really only want it to know

6:15the last 2,000 words or so. And even

6:16that might be generous just to pick up

6:19where it left off. And the reason I

6:21don't do more is because when you have

6:22too many words in a prompt, it sort of

6:25waters down the prompt and makes it less

6:27effective. And so I only put 2,000 words

6:29in there. AI models are getting better

6:30every day. There was a time when putting

6:322,000 words would have been too much.

6:34And now it can handle 2,000 words. So

6:36maybe at some point we'll get to a point

6:37where you can have it read your entire

6:39book and still be able to write a really

6:42good next chapter, but I don't think

6:44we're quite there yet. All right, so it

6:46analyzes those last 2,000 words.

6:48Actually, what this does, what this node

6:50does is it gets the entirety of your

6:52document, and then this code node just

6:54selects the last 2,000 words, the most

6:56recent 2,000 words. All right. Then we

6:59have these four steps, and these are

7:02where the majority of the work happens.

7:04to give you a highle overview. What it

7:06does is it creates a scene brief of

7:09based on your outline. It'll create a

7:11more fleshed out outline essentially of

7:13what that specific scene should look

7:14like and then it moves on to a first

7:17draft and writes that first draft. Then

7:20it looks at the that first draft and

7:22says, "Okay, there are a couple of

7:23things wrong with this. I have it make

7:24some critiques." Uh but it doesn't

7:27rewrite anything. It just goes through

7:28and makes the critiques and then it

7:30rewrites. And all of these are using uh

7:33I believe most of them at the moment are

7:35using Gemini 3 except for this last one

7:37which is actually I believe it's Gemini

7:402.5 Flash. I can actually update that

7:42now to Gemini 3 Flash. It'll be a little

7:45bit better because the the rewrite

7:47doesn't actually need to be that

7:49intensive oddly because the work has

7:52already been done in writing the first

7:54draft and creating the improvement plan.

7:56All that needs to happen is it needs to

7:58rewrite and implement that improvement

8:00plan, but it's not really making any

8:02decisions, which is why I'm able to have

8:04it be a lower model and save a little

8:07bit of money that way. All right, so

8:08let's actually look at the prompts for

8:10this. Uh, they're very lengthy prompts,

8:12so I'm not going to read out the whole

8:13thing, but just to give you an idea,

8:15this is the prompt for the for the scene

8:18brief. I have it pulling the outline,

8:20the characters, the world building, and

8:21the previous chapter text. This is the

8:23last 2,000 words here. Uh, and it's all

8:25contained in these little XML boxes to

8:29sort of help the AI be able to parse

8:31through all of that data. And then in an

8:34instruction XML box, I say given the

8:36above outline and

8:37characters/worldbuilding information,

8:39your task is to flesh out a scene brief

8:41for. And then it pulls in the name of

8:44the chapter that it's working on in this

8:46specific loop. Make sure that the

8:47chapter is clearly labeled with the

8:49words, you know, name of the chapter. Do

8:51not name the chapter anything besides

8:52this. The scene brief should include the

8:54following. Uh I have it pull in the POV.

8:57So in this case, I uh give it specific

9:00instructions because sometimes it gives

9:03me the wrong thing. So in this case, I

9:05wanted first person past tense from the

9:07perspective of Elaine. So I added this

9:09in myself because this is the the story

9:11that we're doing right now. U but you

9:13could in theory change this to whoever

9:16your character is. But I know for this

9:18book that I'm writing, I always wanted

9:19to be first person past tense from the

9:21perspective of Elaine. Okay. [snorts]

9:23Then I pull in the genre, which it pulls

9:25in. Uh, here's the plot verbatim where I

9:28ask it to extract the exact section of

9:31the outline that it's doing. And then it

9:34fleshes that out with additional beats

9:37and blocking. And then I have it

9:39describe what the scene function is, as

9:42well as make sure the plot for this

9:43scene picks up appropriately after the

9:45end of the previous chapter text.

9:47Disregard if this is the first chapter

9:48and there is no data for the previous

9:49chapter text. uh characters, list all

9:52the characters appearing in this

9:53chapter. For each provide their name and

9:55role, physical appearance, etc. And I

9:58make sure it lists all of these things

9:59because their physical appearance might

10:01be different in this scene versus other

10:04like more general scenes, right? I

10:06already have the character sheet, but

10:08that character sheet is going to be

10:10different depending on what scene we're

10:12in. So, I have it kind of create all of

10:14that. Uh it does the same for tone and

10:17style and setting and uh main source of

10:20conflict. It basically goes through and

10:21lists quite a few things to be able to

10:23effectively write a good scene. And then

10:26we move on to the next step which is the

10:29first draft. And this one is also a

10:32fairly complicated one. I have it pull

10:33in the stylesheet, the list of forbidden

10:36words, as well as the outline

10:37characters, world building, and the

10:39previous chapter text and the scene

10:40brief that it had in the last chapter.

10:42So, this is by far the step with the

10:45most context being handled. Um, but most

10:48of the the beefier models like Claude

10:514.5 Opus and Gemini 3 can handle it.

10:54I've seen people do even larger prompts

10:57than this. Uh, I tend to keep it by

11:00comparison on the lean side, but it's

11:02still quite a bit of information. And

11:03here I say your task is to write the

11:05entire chapter name based on the scene

11:08brief and to cover it thoroughly from

11:09deep point of view. Writing the scene as

11:11if it's written by a bestselling

11:12novelist and not rushing through the

11:13scene. Use the pro style examples

11:15samples to know what the pro style of

11:17your chapter should be. Chapter should

11:19be as long as it is needed to properly

11:20properly flesh everything out. Pay

11:22special care to show deep point of view

11:24and showing not telling in order to

11:26fully flesh out the scene without

11:27skipping over important details. The

11:29reader should feel like they are fully

11:30immersed in the scene, seeing the events

11:32through the lens of the viewpoint

11:33character rather than being simply told

11:34what happened. Always keep the follow

11:37following rules in mind. And then I've

11:39over the years added a couple of things

11:41here, taken away some, added more. I ask

11:44it not to use m dashes, you know, and

11:46say tell it to use commas or an ellipses

11:49instead, stuff like that. This section

11:52is probably the section I am tweaking

11:54the most. And I often tweak it on a per

11:56model basis because some models do like

11:59I'll catch something it's doing that I

12:01don't want it to do and I'll just add

12:03that in here. But in general, that's the

12:06essentially what the prompt looks like.

12:08All right. And then the next one is a

12:11little bit simpler. It's the improvement

12:12plan. So for this, I bring in the

12:14previous chapter text, the chapters, the

12:15scene brief, uh the so the first draft

12:18of the chapter, the scene brief, the pro

12:20style example, and the prohibited words.

12:22and I say, "Given the above chapter, I

12:24want you to critique the chapter on a

12:26lineby-line basis and find ways to

12:27improve the chapter. Give specific

12:29examples. Here are a few things to look

12:30out for." And then I give it a big long

12:33list of all of the things that I want it

12:35to look for. I actually think that I may

12:38be rethinking this in the future and

12:39actually creating some automations that

12:40are specific for editing where I have a

12:44different AI agent that is specifically

12:47focused on one specific thing and it

12:50gives you recommendations. Um, that way

12:52I could probably use a less powerful

12:54model because it's only focusing on one

12:56thing, but for now I'm just using this

12:58and uh I will refine those editing

13:01automations in the future. And then from

13:03there, it creates an improvement plan

13:04and then it goes to rewrite the chapter.

13:06This one's really simple. I give it the

13:08original chapter that was written and

13:09the improvement plan and say using the

13:11text of the original chapter and the

13:12improvement plan, I want you to

13:13implement the suggestions in the

13:14improvement plan. Only implement the

13:16suggested changes and do not change

13:18anything else about the original

13:19chapter. reproduce the entire chapter

13:21with the suggestions made. A chapter

13:23should begin with a chapter header

13:24written and markdown as an H2 heading

13:26like this. And then I just give it the

13:28name of the chapter. All right. And

13:31that's essentially it. And once it

13:32finishes that for the chapter it's

13:34working on, it will loop back. Well,

13:37first it'll add that chapter to your

13:39document, your blank [snorts] document.

13:40And then it'll loop back around and

13:43start on the next chapter. So why don't

13:45we go ahead and run this automation?

13:49remember here in where I ask it to find

13:51the names of all the chapters. I'm

13:53asking it specifically to do just

13:54chapters one through three. So, I'm

13:56going to go ahead and run that for you

13:58right now so you can see what that looks

13:59like. And um yeah, let's go. So, I'll go

14:02ahead and hit execute workflow. And then

14:05it'll have this little popup come come

14:07up and I'll say this the name is of the

14:09book is The Chosen of Avalon. And I

14:13don't really have any author notes. So,

14:15I'm just going to hit submit. And now

14:16it's going through grabbing all of those

14:18documents and it selected the chapter

14:21names. If we go through and actually

14:22look at the output for that node where

14:24it selected the chapter names, we can

14:27see that it successfully selected just

14:29the first three chapters here in this

14:31output. And it's outputting in JSON like

14:33I asked it to. And if I had removed that

14:36one line in the prompt, it would have

14:38actually gone through and given me all

14:39the chapters that are in the outline.

14:41But for now, it's just doing chapters 1

14:42through three. And now it's working on

14:45the first step, which is the scene

14:47brief. And from there, it'll continue on

14:50and I'll get back to you when this is

14:52finished. Quick pause. If you're an

14:54author who wants to start writing more

14:56books, I put together my complete AI

14:58prompt pack that helped me and many

15:00others in my community write a book in a

15:02month, often more. It's completely free.

15:04Hit the link in the description, join

15:06the free weight list for Story Hacker

15:07Gold. Then you just answer a couple

15:09quick questions and I'll send you the

15:11full prompt list plus a few extra free

15:13goodies. I'll also email you when I open

15:16the paid Story Hacker Gold group so you

15:18can join if it feels right. There's no

15:20commitment to join though and you'll

15:21still be able to get all of the freebies

15:22you signed up for either way. All right,

15:24back to the video. All right, so it's

15:26gone ahead. It's already finished

15:28chapter 1 and has started work on

15:30chapter 2. But since we've got chapter 1

15:33already, we can go ahead and go into our

15:35first draft. And you can see that it has

15:37successfully pasted chapter 1 in here.

15:40Now it's pasted in in markdown. There's

15:42a simple way that you can convert that

15:44either through an a free online tool or

15:47something like that. I believe there's a

15:49way to do it inside of Google Docs as

15:50well. But regardless, we can take this

15:53and actually start to go through it and

15:55modify it to our heart's content. So

15:57this is what I would do if I were going

15:58through and doing a couple chapters at a

16:00time. I'd say. All right. Dirt packed

16:02under my fingernails, cool and gritty,

16:04kneeling in the lom felt right, far

16:06better than the stiff wooden chair back

16:07at the cottage. Hello there, Mr. Moss.

16:10The patch of green clinging to the

16:12limestone didn't answer, but it looked

16:14particularly fluffy. I poked it. It

16:15sprang back, spongy and damp. You look

16:18very dashing today, I told it. Mrs.

16:20Fungus is just over that rotting log.

16:21You know, it's not a long trip if you

16:23stretch. A squirrel chattered from high

16:25in the canopy, dropping a chewed acorn

16:27shell inches from my knee. I glanced up

16:29at the shaking branch. Rude. I am having

16:32a conversation down here. Pine needles

16:34and damp earth filled the air a scent

16:35that always settled my stomach. Sunlight

16:37cut through the leaves, warming the back

16:38of my neck. The basket beside me sat

16:40half full with wild garlic nettles and a

16:43few precious roots I'd pried from the

16:44soil without snapping them. I shifted my

16:46weight to stand. My foot slid on the

16:48slip pack on a slick patch of mud. My

16:51boot came down heavy. A wet snap sounded

16:53under my soul. I froze. Um, so a couple

16:56of things here. Um, this might be more

16:58of an a problem with the outline, but

17:00I'm realizing we're kind of not really

17:02getting into the action very quickly.

17:05Um, there's very little hook in this

17:08first page. So, I I might go back to the

17:11drawing board actually and make sure I

17:12have something that happens right away.

17:15Or I could go f further and find a place

17:19where it would make sense to start the

17:21chapter uh a little bit more in medius

17:23res. Also, a couple of things. I liked

17:26the pros overall. This last bit though

17:28needed some work. Uh, lots of very short

17:31sentences. So, I would modify this a

17:34little bit. I'd say my shifting my

17:35weight to stand just as my foot slid on

17:39a slid on a slick patch of mud. That's a

17:42little bit of a what we call that

17:45redundant right there. My foot slid on a

17:50patch of mud. Right. Um, my boot came

17:55down heavy. See the See the the foot

17:58already slipped on the patch of mud. So,

17:59the boot doesn't really make sense. So,

18:03I'd say I just delete that entirely. A

18:05wet snap sounded under my soul. So, I

18:07changed this to say, "And I froze."

18:10Dread curled in my stomach. I lifted my

18:12foot. A single blue bell lay crushed

18:13against the dark earth. Its delicate

18:15stem bent as a sharp unnatural angle.

18:17Oh, no. I dropped it back to my knees,

18:19discarding the back the basket. No, no,

18:21no. I'm so sorry. All right. So, you get

18:23the idea. Little edits like that

18:25throughout. Probably a lot for some

18:27people uh depending on how much you uh

18:31like the pros and how close it is to

18:32your vision. There's always going to be

18:34a fair amount of editing needed even as

18:37these large language models get better

18:39and better. And you can also try

18:41experimenting with different models. Um

18:44for instance, this was written with

18:46Gemini 3. I find that Claude 4.5 Opus is

18:49also very good. slightly different, a

18:51little bit more dramatic, which it would

18:53not be appropriate for this scene, I

18:55think, but for like a climactic end of

18:59the book kind of scene, it would be very

19:02appropriate. Uh, so that those are just

19:04my thoughts there. And that is

19:06essentially what we've got for this full

19:08thing. It's continuing on with chapter 3

19:10now. Looks like it finished. If we go

19:13down, scroll down and we can get to

19:15chapter two. And here it is, chapter 2.

19:18And then chapter 3 will stop here. Then

19:22chapter 3 will start here at the end

19:24there. And if we look at the word count,

19:27the words for these first two chapters

19:28is 5600 words. So that's a very good

19:30good length for a full chapter. So I'm

19:34happy with that. And yeah, so that's

19:36been this full book automation process.

19:40You may have noticed that uh if we were

19:43to somehow combine the brain dump to

19:45dossier and the dossier to full outline

19:48and the outline to chapters automations,

19:52you might be able to actually create a

19:53whole system that creates a book in one

19:55click. Now, it's not necessarily

19:56something I recommend, but I do have

19:59this tradition now of every year I will

20:02try to write a sci-fi beach romance

20:05book. Don't ask. using automations,

20:09using prompts completely with no human

20:11input, just to kind of get a sense and a

20:14measure of how much these tools are

20:17improving and how much like our

20:19prompting methods and and all of that

20:21are improving year-over-year.

20:24And we're getting close to the time of

20:27year to do that again. And I plan on

20:31essentially putting these three

20:32automations together and being able to

20:33do it with one click. Now again, I don't

20:35recommend that for most people if you're

20:38doing that for your own book because

20:39you're going to end up with a lot of

20:41logical issues. But we are moving to an

20:45age where that might be possible. Uh

20:47although I do think and I will I have

20:50said this a lot, but an AI doesn't

20:53matter how good the AI gets, it will

20:55always be better when paired with a

20:56human and having a human guiding it. And

20:59even if it can create a cohesive story

21:01that actually works, it won't

21:03necessarily be your story. So making

21:06sure that you actually spend the time to

21:08do the thinking and let AI do the doing

21:11is absolutely crucially important as we

21:13move forward into this very AI saturated

21:15world. All right, those are just my two

21:17cents. If you want to get your hands on

21:19this specific automation, it is

21:20available in my story hacker gold group.

21:23The link for that is down below. And

21:24again, if you sign up for the waiting

21:27list, assuming that the uh that it's not

21:30open at the time that this video goes

21:32out, then you'll get my prompts for free

21:34when you join the waiting list. You

21:36obviously get a whole lot more if you

21:38actually join. Uh but if you join the

21:42waiting list, you'll get those free

21:44prompts as well, regardless of whether

21:46you actually want to join the group or

21:48not. So check that out and I'll see you

21:49in the next

More from The Nerdy Novelist

Recently added transcripts

Browse the whole transcript library

This transcript was generated from the captions YouTube publishes for this video. Get the transcript of any YouTube video atfreeyoutubetranscribe.com, free, unlimited, no sign-up.