How I create Fan Fiction

This is not a tutorial.

I am not writing this to tell anyone how they should use AI, how they should write, or what their own creative process should look like. This is only a description of how I personally create The Monk and the Witch, my Dungeons & Dragons fan fiction saga, using AI as part of my workflow.

I also want to be transparent.

AI is a major part of how I create this saga. I use it constantly. I use it for thinking, planning, organizing, drafting, revising, documenting, worldbuilding, character creation, technical publishing work, and image generation. It is not a minor tool in the background. It is central to how the project exists.

But that does not mean the work is easy.

One of the strange assumptions people sometimes make is that using AI means pressing a button and receiving a finished book. My experience is almost the opposite. AI makes this project possible for me, but it does not make it effortless. It changes the kind of work I do. It gives me leverage. It lets me create at a scale I could not otherwise manage. But the process still takes a massive amount of time, attention, judgment, emotional investment, revision, and love.

I do this for free. I do it because I love the characters, the story, the world, and the act of creating it. AI has not removed that love. It has made it easier for me to build something large enough to hold it.

And I do not think people should feel shame for using this technology creatively. I am enthusiastic about it. I have fun with it. It has become one of the most important creative tools I have ever used.

This is how I use it.

For my workflow: ChatGPT

When I say I use AI, I mostly mean ChatGPT.

For my workflow, ChatGPT is not interchangeable with any random chatbot. I have tried other tools, especially Gemini, and they have not worked nearly as well for what I need.

The issue is not only the quality of the generated prose, although that matters. The larger issue is the whole working environment: the web app, the mobile app, projects, files, memory, conversations, and the ease of moving between different kinds of work.

My process involves an enormous amount of context. I need to organize documents, return to old conversations, upload and review files, maintain process guides, discuss images, handle technical publishing tasks, and work from both my computer and my phone. If the app experience creates friction, I lose too much time. Tiny obstacles become impossible at this scale.

ChatGPT is the environment where I can actually work.

I also use the strongest paid version available to me. I need the depth, the context handling, the continuity, and the tooling. For this kind of project, a lighter version is not enough.

This is important because the article is not really about casually asking an AI to write a scene. It is about a whole working system. For me, that system depends on ChatGPT specifically.

Before writing: I learned about writing

The first serious phase of using AI was not prose generation.

It was learning.

Before I tried to create the saga with AI, I spent about six months using ChatGPT to talk about writing, prose, fantasy literature, genre, authors, and books.

I would do chores and speak into my phone. I asked about books I did not know, series I had never read, different fantasy styles, and different traditions of fantasy writing. I revisited my favorite authors and favorite books. I asked how those authors wrote, how they became writers, what kind of prose they used, what mattered to them, how they became known, and what their lives as authors looked like.

Because The Monk and the Witch is Dungeons & Dragons fan fiction, fantasy mattered especially. I wanted to understand the genre I was working inside. I wanted to know what I loved, what I did not love, what kind of fantasy tradition I wanted to belong to, and what kind of prose I wanted to avoid.

So AI was first a literary companion.

It was a way to educate myself, broaden my map of fantasy, and think about writing before I ever tried to make it produce usable scenes.

AI as Voice-to-Notes translator

One of my biggest uses of ChatGPT is not prose generation at all.

It is turning speech into usable notes.

I have an enormous amount of story material to manage: character ideas, corrections, scene logic, emotional beats, lore decisions, worldbuilding details, process documents, publication plans, and random thoughts that matter only because they might become important later.

Typing all of that manually would be impossible for me. Or at least it would be so slow that I would spend my life doing it.

So I speak into my phone.

A lot.

I use ChatGPT as a tool that can listen to messy spoken thought and turn it into coherent written material. The important part is not only transcription. It is organization. I can speak naturally, repeat myself, jump topics, correct myself, follow a sudden idea, and then ask the AI to clean that into notes I can keep on my computer.

This is essential to the project.

Without that, I would lose too many ideas. I would forget too many decisions. I would not be able to maintain a saga of this size.

AI gives me a way to think out loud and then keep the result.

Living documents

The second major use is document creation, document review, and document maintenance.

After hours of speaking into my phone, I end up with a lot of documents. Those documents cannot just pile up forever. They need to be reviewed, deleted, updated, merged, rewritten, compressed, or clarified.

So I create dedicated conversations, put documents into them, and talk through what needs to change. I might say that a section is outdated, that a character decision evolved, that a process needs a caveat, or that a whole document needs to be reorganized.

Sometimes I ask the AI whether a document needs updating. Sometimes I ask whether a section has become obsolete. Sometimes I ask it to compare a document against the current way I work and tell me what no longer fits.

This is especially important for process documents.

The story documents matter, but the process documents are what allow the AI and me to keep working together. Those guides explain how I create scene notes, how I review prose, how I handle final polishing, how I want drafts corrected, and how the AI should behave in specific phases of the project.

Those documents constantly evolve. The more I work, the more edge cases I discover. The process becomes more precise, more complicated, and more conditional.

So part of the project is maintaining the documentation that makes the project possible.

Story and prose are different things

Before talking about writing the actual books, I need to separate two things.

First, there is creating the story.

Second, there is writing the prose.

Those are not the same thing.

Creating the story means building the characters, arcs, relationships, emotional logic, conflicts, worldbuilding, history, and structure. It means knowing what matters and why.

Writing the prose means deciding how that story appears on the page: narration, rhythm, point of view, dialogue, description, action, pacing, and emotional delivery.

Both matter. But for me, story creation comes first.

And story creation begins with character.

It starts with character

My characters are the heart of the saga.

I like them complex. I want them to feel like people with long lives, deep histories, contradictions, habits, wounds, preferences, skills, memories, relationships, and ways of seeing the world.

So I use ChatGPT heavily for character creation.

I speak into my phone and give too much context on purpose. I describe where a character lived, what happened to them, what they care about, what they eat, what they fear, what objects they carry, what books they read or wrote, what poems they might have composed, how they speak, how they fight, how they love, how they avoid pain, how they joke, and what shaped their psychology.

The goal is not just to fill a character profile.

The goal is to build a person.

And then, through discussion, exploratory scenes, questions, and corrections, I keep refining that person until they become alive enough that I can recognize when something is wrong.

That is one of the most important parts of the workflow: reaching the point where I know when a character would not say something, would not feel something that way, would not make that choice, or would not stand in that part of the scene.

The character sheet matters

Because this is Dungeons & Dragons fan fiction, the character sheet matters a lot.

I do not treat the character sheet as a disposable game document. I build my characters from level 1 to level 20. I think about their skills, class features, powers, proficiencies, tools, spells, combat style, and how the build changes as they grow.

But the goal is not to copy mechanics directly into prose.

The goal is to translate mechanics into lived reality.

I want the characters’ actions to feel embodied, practical, and believable inside the fictional world. They are still Dungeons & Dragons heroes, so they are powerful people. But I try to avoid making the narration feel inflated or mechanically literal. In many ways, I am trying to be anti-epic.

The character sheet is the hidden structure. The prose must feel like life.

Ṛṣiśūra is a good example.

In the game, monks fight very well with unarmed strikes. But in the fiction, Ṛṣi uses a staff. That makes sense to me. A reinforced staff with metal caps gives him range. It lets him maneuver more safely against armed opponents. It lets him intercept weapons, control space, and survive against swords, shields, and armor.

Having him constantly fight dangerous armed enemies barehanded would feel too mechanical to me. It may work in the rules, but I want the narration to make sense as action.

His healing works the same way.

Ṛṣi is a Mercy monk and an Aasimar, so mechanically he has abilities like Hands of Healing and Healing Hands. But in the prose, I translate that into practical medicine, herbalism, emergency response, and triage.

He stops bleeding. He checks breathing. He stabilizes people. He makes sure someone survives long enough for deeper healing to matter.

This is personal to me because I was an emergency first responder for three years, with around twenty missions. I have real experience with first response. So when I write Ṛṣi’s healing, I care deeply that it does not become vague glowing fantasy magic. I want it to feel like a body being cared for under pressure.

This is one of the central principles of how I use D&D mechanics: the rules matter, but the prose must translate them into reality.

Building the world

Because I write Forgotten Realms fan fiction, I am not creating the world from nothing.

A lot of the lore already exists.

That is a gift, but it is also a challenge.

I need to decide how much canon to use, which sources matter, how to handle contradictions between editions, where I want to respect existing lore, and where I need to create my own material inside the gaps.

My main characters are original characters. I try to avoid using canonical characters unless it makes sense. The saga is not built around famous Forgotten Realms figures. It is built around my protagonists moving through the Realms.

Also, the narration is not a travel guide. We are not exploring the world from above. We follow the protagonists closely, in limited perspective. The world matters because of what they see, hear, touch, smell, suffer, misunderstand, and live through.

That means the details need to feel real from inside the scene.

I use AI constantly to help with that. I ask what information exists about a place. I ask whether there are sourcebooks, novels, games, wikis, or older edition references. I ask the AI to compile sources, compare them, identify contradictions, and help me understand what is established and what is empty.

A lot of the Forgotten Realms has enormous gaps.

A forest may have a name, a few paragraphs of lore, some history, maybe a note about who lives there, but no concrete description of what it feels like to walk through it. In those cases, there is a lot of creative freedom. I need to invent, but I want to invent in a way that feels like it belongs.

Even places with a lot of lore are complicated.

Baldur’s Gate, for example, has a huge amount of material across editions, games, gazetteers, campaign books, and modernized lore. The information is not always identical. Sometimes it is reworked. Sometimes the tone changes. Sometimes the city has a different emphasis depending on the source.

AI helps me sort that out.

It helps me ask: what version of this place am I writing? What does the character actually perceive? What matters to the scene? What details would someone living there know? What details are only useful for a Dungeon Master or a lore article and should stay out of the prose?

The gods as a worldbuilding question

The gods of the Forgotten Realms are another major example.

The way the Realms treats deities and worship has changed a lot over time. Older material often gives faith a very strong presence. The gods feel central to ordinary life, politics, morality, afterlife, culture, and identity. Later material can feel different, sometimes more modernized, sometimes less intense in how it handles worship.

So when I write the saga, I need to decide what tone I want.

Do I want the older, more strongly religious feeling? Do I want divine presence to feel immediate and culturally central? Do I want a more modern, less god-saturated tone? How much do ordinary people think about worship? How much does a temple matter? How much does a god shape a person’s choices?

These are not small details.

They shape the emotional and philosophical atmosphere of the story.

ChatGPT helps me have those discussions. It helps me compare approaches, think through implications, and decide how hard I want to lean into different versions of the setting.

Exploration before structure

Once I had characters and a world, I started exploring.

I generated a huge amount of draft prose around anything I felt curious about. Much of it was not meant to become final text. It was exploration.

I created random character interactions. I tested conversations. I asked what a character would answer, what kind of joke they would make, how they would react to a stranger, how they would speak when tired, what they would do in a town, how they would behave on the road.

As I discovered things, I updated the character profiles.

This was a very long process. Months and months of testing, refining, and building. The characters are the masterpiece for me, so this mattered deeply.

Then I began exploring the world as the characters.

If they traveled for days in one direction, what would they see? If they stopped in a place, what would matter? If they encountered a problem, how would they respond? If they had no adventure that day, what would they talk about?

This is how I began discovering The Monk and the Witch as a saga.

Not through a clean outline.

Through exploration.

I saved everything.

A lot of it was bad. A lot of it was boring. Some of it was basically a huge travel log. But inside that mass of material were scenes that were alive.

That process taught me taste. I learned to feel what was garbage and what was worth keeping. I learned what made a scene matter to me.

The first campaign and the the first volume

At first, I did not know I was writing multiple books.

I only had a movement: the characters meet, leave together, travel south, and do things according to their nature.

I generated an enormous amount of material around that. I imagined their life day by day. Did they meet someone? Did they fight? Did they sit together? Did they travel in silence? Did they speak at night? Did something small happen that mattered only because they were together?

Eventually, this became the first volume, Spirit Threads, and the first three books of the saga.

I did not know that at the beginning.

I only knew that the story had grown enough to become something larger than a sequence of travel scenes.

The major shift came when I decided to give the characters something harder to solve. By that point, they had almost a year of shared history. They had become a pair. They had lived together, traveled together, suffered together, and changed each other.

So I wanted to give them something that looked simple at first. Something they would think they could solve and then continue on their way.

But it would turn into something much deeper.

That became the third book.

And that changed how I understood story.

The Monk and the Witch narrative unit

At some point, I realized that the two protagonists were not only two separate characters.

They were also a narrative unit.

Ṛṣi and Maeril are separate people, of course. They have different histories, bodies, voices, skills, wounds, and desires. But when they are together, the story often treats them as one combined force: the Monk and the Witch.

That became important to how I think about the saga.

In real life, this can happen too. Two people can become something together that neither of them is alone. But in prose, it becomes even more powerful, because the narration can focus on the bond as an active force.

The story is not only “what happens to him” and “what happens to her.”

It is also “what happens through them together.”

This changed how I built later material.

I also started seeing places, cultures, villages, towns, political movements, and groups as narrative forces. They are not just background. They pressure the protagonists. They create the shape of the story. They become active in the narrative.

So the question changed.

Instead of asking only, “What would they see if they traveled here?” I started asking, “What happens if this place collides with them?”

Thinking like a Dungeon Master

At that stage, I began thinking more like a Dungeon Master.

If these characters were the party in a Dungeons & Dragons campaign, what would I do with them?

What situation would challenge them? What would seem easy but become difficult? What would matter locally? What would fit the place, the history, the people, the politics, the dangers, and the emotional state of the protagonists?

I started building what I called campaigns.

I used Dungeons & Dragons tools, including D&D Beyond and the official VTT, to roll things, track things, test situations, and see what happened. In a way, I was playing the game alone, as both player and Dungeon Master.

But it was not like a published adventure module.

The pacing was much more prose-driven. Travel took time. Long rests mattered. Injuries hurt. Damage had consequences. Failure did not vanish after the encounter ended. The story did not rush from one set piece to the next.

It became a hybrid process: part D&D play, part prose creation, part world simulation, part character study.

Campaign two and slow life

After the first major movement of the saga, I started what I thought of as campaign two.

I wanted it to feel different.

I wanted to honor The Earth’s Children by Jean M. Auel, which is my favorite series of all time. I read all six books twice, and I loved the experience of living slowly with characters through skills, daily life, environment, culture, and survival.

So for the second major movement of the saga, I changed the mode.

Instead of the protagonists traveling constantly through the world, I settled them in one place for a long time. I wanted to explore what happens when they live slowly, day by day, with a very small group in a specific region of the Realms.

That changed the questions.

How do they learn? How do they train? How do they improve? What does a day look like when nothing dramatic happens? What matters to the people they live with? How do relationships shift over a year? How does belonging form?

To make that work, I placed them with wood elves.

That created a whole new layer of work. These were elves, with long lives, long memories, and complex relationships with history, grief, family, place, and time. I had to create many characters with deep backstories and weave them into the environment and the canonical history of the Realms.

This became Volume Two, The Twilight Stream.

And this is where my method changed again.

I started thinking in scenes.

The scene note method

Scene notes became the central pillar of my writing process.

At that point, I had character profiles, environment profiles, plan profiles, background profiles, cultural profiles, and different versions of documents for different books or sub-campaigns.

But I needed a better way to turn all of that into actual fiction.

So I stopped going directly from idea to prose.

Instead, I began creating scene notes first.

A scene, in this system, is usually about 10 to 30 minutes of reading time. Each scene is divided into movements. Often there is an opening movement and a closing movement, though not always. Sometimes two scenes connect directly, or a scene needs to end with a strong transition or cliffhanger.

Each movement has a purpose. It has an entering state and an exit state. It may do several things, but it has one main goal. Something must change.

Then each movement is divided into beats.

The beat became the smallest useful unit of narration for me.

A beat is small enough that I can know what it needs to do. It tells me what happens, what changes, what the reader should understand, what the character experiences, and what the prose will later need to carry.

This changed everything.

I stopped thinking first about finished paragraphs. I started thinking about how the scene was built.

From Book 4 onward, every scene that has prose also has scene notes attached to it.

Later, when I returned to the first three books, I applied the same method retroactively. I broke apart the old exploratory drafts, kept the strongest moments, discarded what did not work, and rebuilt the first three books from scene notes.

In many ways, I started over.

And the books became much stronger because of it.

Scene notes: The master working document

Scene notes matter because they tell both me and the AI how to work.

They are not just summaries.

They are the master working document for the scene.

They include lore, emotional direction, what must be included, what must be avoided, what the characters should not do, what they would not say, possible dialogue, example lines, warnings, pacing notes, and structural goals.

They also allow uncertainty.

Sometimes I know exactly where a book is going. Sometimes I only know the next scene. Sometimes I know I need an action scene, a transition scene, an opening scene, or a quiet scene, but I do not yet know its exact shape.

So the scene notes become an ongoing discussion.

I ask the AI: how should this be built? What are the risks? What could be boring? What could be fun? What would make this scene fail? What does this movement need? What would this character actually do?

Then I add more context: profiles, ideas, images, fragments of prose, lore notes, contradictions, emotional goals.

A lot of the real creative work happens there.

Sometimes I spend weeks or months creating scene notes without generating the final prose. I might listen to them while doing chores and think, “This is going to be amazing,” even though no publishable prose exists yet.

That is a huge part of the joy of the process.

Timeline: The fiction and the narration

Scene beats also solve a major prose problem: simultaneity.

In the fictional world, many things can happen at the same time. In a fight, one character may be handling a group of enemies while another is reacting to a new person appearing. Emotion, action, movement, and danger can all happen together.

But prose does not happen all at once.

The reader reads from top to bottom.

So I need to decide how the reader experiences the scene. What do they see first? What comes second? What information is delayed? What action needs to be isolated? When do I cut from one character to another? When do I stay with one line of action?

If I try to explain everything at once, the scene becomes unreadable.

This is one of the reasons beats became so important. They let me translate fictional simultaneity into readable sequence.

The timeline of the world and the timeline of the page are not the same thing.

Scene notes help me manage that.

Later books changed earlier books

Another reason scene notes matter is that the saga is long.

When I was polishing Book 1, I already had scene notes finished for Book 7. I was six books ahead in story understanding. I also had major ideas for Books 8 and 9 and for how the whole saga would end.

That changed how I returned to Book 1.

I was not rewriting it from the perspective of the person who first drafted it. I was rewriting it with knowledge of where the saga was going.

That is why scene notes are so useful. They let me return to earlier material with the later story in mind. They let me rebuild old exploratory drafts into something intentional.

Without notes, I would have to hold too much in my head.

And that is impossible. This saga is nine books. It is roughly thirty hours of reading. Nobody can keep all of that perfectly in memory.

Scene notes hold the story when I cannot.

Prose generation drafts

Once the scene notes are ready, I ask the AI to generate prose.

That result is a draft.

Not a final draft.

Not a publishable scene.

Not something I would show readers.

A draft.

This is another place where people misunderstand AI writing. They imagine the AI produces the finished text. That is not my experience.

The result is hit-or-miss. It is almost a gamble.

Sometimes it is astonishing. Sometimes the AI produces a joke, an emotional turn, a sentence, or a scene shape that is better than what I had in mind. Sometimes I cry. Sometimes I laugh. Sometimes I am genuinely shocked by how alive something feels.

And sometimes the output is unusable.

It may be confusing, bland, wrong in tone, structurally broken, too generic, too epic, too mechanical, or simply not my story.

That is part of the process.

The draft is not the end. It is raw material.

More freedom usually means worse results

One of the clearest lessons I learned is that the more freedom I give the AI, the worse the results usually become.

The AI needs constraints.

It needs character profiles, prose guides, scene notes, correction guides, context, lore notes, and clear instructions. It needs to know what not to do. It needs to know what a character would never say. It needs to know the scene’s function, emotional direction, and structural order.

If I leave too much open, the AI tries to do everything at once.

And then the scene fails.

I also need to be aware of what context the AI actually has. What is in the current conversation? What is in the project? What is in memory? What documents has it seen? What has it forgotten? What assumptions is it making?

The AI does not magically know the saga. It works from available context.

So part of using AI well, for this project, is managing context.

Draft correction

After the AI produces a prose draft, the correction process begins.

Even if the draft is good, it needs work.

If I am very lucky, and my notes were excellent, maybe the draft is 80% usable. That still means I need to change around 20% myself.

I reread. I correct. I ask questions. I rewrite sentences. I remove sections. I add missing details. I change dialogue. I fix narration.

At this stage, the work becomes extremely detailed.

Does this respect my narrative style?

Did the narration drift into omniscient voice?

Did the AI forget a character?

Is a detail from a previous scene missing?

Would this character actually say this?

Would they use these words?

Would they feel this emotion in this way?

Is the environment correct?

Is this paragraph too long?

Is this description bland?

Does this whole section need to be rewritten?

This is slow work.

For important scenes, I can spend four hours reworking a scene that takes twenty-five minutes to read. Sometimes I revise the prose. Sometimes I realize the scene notes were wrong or unclear, so I go back and fix those too.

The draft exposes problems in my own planning. If the AI cannot make sense of the scene, sometimes that is because I had not made sense of it yet.

Correction guides

Because correction became such a major time sink, I built correction guides.

After reviewing chapters, I would ask the AI to summarize what happened: what the generated prose did, what I asked to change, what final decisions I made, and what the final prose became.

I wanted those lessons recorded.

At first, this created a huge amount of correction material. Too much. It became unusable by itself. So I had to compress it, reorganize it, and turn it into digestible guidance the AI could actually use later.

That became another part of the process: not only correcting prose, but extracting lessons from correction.

The AI helps me write better AI instructions.

It is recursive, but it works.

Reviewed drafts are not final

After the first major correction pass, I do not call the result final.

I call it a reviewed draft.

That distinction matters.

The reviewed draft is readable. It may be good. It may be close. But it still needs distance.

Because I work far ahead in the saga, I often return to a scene much later. By then, I no longer remember all the scene notes. I no longer remember every correction. I can sit down and read the scene more like a reader.

That is the next test.

The emotional reread

The emotional reread is one of the most important parts of my process.

I sit at my computer. Sometimes I listen to music. Sometimes I read in silence. I take my time.

The goal is not only to check for errors.

The goal is to live the story again.

This matters because I love the story. I love the characters. I love what I created. I do this for free, and I want the experience of rereading it to matter.

This reread is also my benchmark.

If a scene makes me laugh or cry, it works.

If it gives me a strong emotion and I enjoy the time I spend reading it, then it is a success.

If I start skim-reading, if I get bored, if I want to skip ahead, then the scene has failed.

And if it fails, I cannot publish it.

Not every scene needs to make me cry. Some scenes are quiet. Some carry information. Some let the story breathe after intense emotional moments. If I am engaged and entertained, even without a huge emotional reaction, the scene may be doing its job.

But if a scene is supposed to carry a major emotional payload and I feel nothing, it failed.

That means it needs to be rewritten, reworked, shortened, removed, or rebuilt.

When a scene fails

When a scene fails at this stage, it is expensive.

Sometimes I have to start over almost completely.

I create a new conversation. I give the AI the context again. I review the scene notes. I rethink what the scene needs to do. I rebuild the draft.

Depending on the scene, that can take another 10 to 15 hours.

If I am lucky, maybe it takes four.

And even then, the new polished version is not automatically final. It has to go back into the reread process later.

So AI does not remove the labor.

It changes where the labor happens.

The process can include story planning, scene notes, prose generation, correction, reviewed draft, emotional reread, failure, reconstruction, and reread again.

Only after that can a scene become something I am willing to publish.

Text Has to Be Finished

Once a scene passes, I move it into my finalized material.

At some point, I need to declare the text set in stone.

That is important for me.

I cannot spend my entire life endlessly revising the same scenes. I already spend a massive amount of time on this project. I want to move forward with the saga, and I also want to move forward with my life.

So eventually, a finished piece becomes finished.

Not perfect.

Finished.

That is different.

Publishing: Another whole process

Once the scenes are polished, the writing is still not over.

There is the publication process.

I need to create the e-book. I need to build the EPUB. I need to maintain the website. I need to publish on fanfiction sites. I need to summarize books, structure the saga for readers, create download pages, troubleshoot formatting, and handle technical problems.

AI helps with all of that too.

I ask for Linux commands. I ask for scripts. I ask how to compile files. I ask how to structure a Hugo website. I ask how to present the saga clearly. I ask for help making summaries and reader-facing descriptions.

This is a different kind of work, but it still matters. A finished story needs to become something people can actually access.

So AI is not only part of the creative process. It is part of the production and publication pipeline.

Images and Covers

I also use AI to create images, and that has become its own workflow.

At first, it is easy to think you can simply ask for a picture and get what you want.

That has not been my experience.

If I want consistent characters, I need a visual process.

I start by describing the character: how they look, how they should feel, what kind of image I want, what matters about their body, face, presence, clothing, and mood.

Then I create a turnover.

A turnover is a front, left, right, and back view of the character in very minimal clothing, usually on a clean studio or white background. The goal is not a beautiful picture. The goal is a reference: a neutral doll-model version of the character.

After that, I create closer references: upper body and head, then face and head.

Once I am happy with the body and face, I move to clothing.

The clothing changes as the saga progresses. Characters travel, age, develop, change gear, and move through different phases of the story. So for the main protagonists, I create multiple turnovers with different clothing and equipment.

For secondary characters, I usually only need one turnover with the clothing already included. But for protagonists, I may have nine or ten different visual references for different stages of the story.

I also create separate gear references: staffs, swords, important objects, or anything the character might hold.

This matters because without references, the AI changes everything. The face changes. The clothing changes. The gear changes. The character stops looking like themselves.

Finished dramatic pictures are not enough as references. They are too specific. What works best is clean visual documentation: body turnover, face reference, clothing turnover, gear image.

That is how I create more consistent art.

It is also how I work toward book covers. I gather references for the characters in the scene, provide the context, and then build the image from that.

Another important part of my process is that I do not ask the AI to generate the image immediately.

First, I discuss the image. I provide the references, explain the intention, describe the mood, the pose, the scene, the constraints, and the feeling I want.

Then I ask the AI to create the prompt.

I save that prompt in a file.

That way, if I need to rework the image later, I still have the prompt and the context. I am not trying to reconstruct the whole thing from memory.

For me, image generation works best when it is treated like the rest of the project: documented, iterative, contextual, and built from references.

What AI actually gives me

AI gives me scale.

It gives me a way to speak instead of type. It gives me a way to organize messy thoughts. It helps me maintain documents, build characters, research lore, compare sources, generate drafts, correct prose, create process guides, plan scenes, test ideas, build images, troubleshoot technical publishing, and keep a nine-book saga coherent.

But it does not decide what matters.

It does not know when a scene makes me cry.

It does not know when a character feels wrong unless I have taught it the character and then judge the result myself.

It does not know what I want the saga to become unless I keep explaining, correcting, refining, and choosing.

AI gives me material.

I decide what survives.

Why I am open about it

I want to be transparent because AI is part of the work.

I also want to be clear because I think people misunderstand what AI-assisted writing can be.

For me, this is not a way to avoid effort. It is a way to make an impossible project possible.

The effort is still enormous.

I spend hours speaking notes. I spend days building scene structure. I spend months refining characters. I spend years planning the saga. I reread, correct, rebuild, discard, regenerate, rewrite, polish, and test scenes emotionally before I publish them.

I am not ashamed of using AI.

I love using it.

It has made me more creative, not less. It has let me build something larger, stranger, more emotional, and more ambitious than I could have managed alone.

The Monk and the Witch exists because I care deeply about the story.

AI is one of the tools I use to carry that care into form.