How to Create Chapter Illustrations with ChatGPT Without Losing Your Damn Mind

An author’s practical guide to character consistency, visual storytelling, fewer wasted generations, and knowing when to tell the AI to leave well enough alone.

-By Hawk

There’s something ridiculously satisfying about finally seeing a scene you’ve written come to life.

Maybe it’s the first time your characters meet. Maybe it’s a devastating emotional moment, a fight scene, a long-awaited reunion, or two idiots who’ve spent half the book dancing around their feelings finally kissing.

Whatever the moment, you can see it in your head. You know exactly what your characters look like, what they’re wearing, where they’re standing, and what the scene is supposed to feel like.

Then you ask an AI to illustrate it.

And suddenly your protagonist has a completely different face, somebody’s got six fingers, the room looks nothing like you described, and the character who was supposed to be sitting on the bed is somehow standing in the doorway.

Fantastic.

Welcome to AI-assisted chapter illustrations.

I’ve spent a considerable amount of time working with ChatGPT to create illustrations for my stories, and if there’s one thing I’ve learned, it’s that good results have a lot less to do with throwing enormous prompts at an image generator than people might think.

In fact, my actual process is considerably less complicated than some people might expect.

Most of the time, I give ChatGPT my chapter, upload my established character references, and essentially tell it to help me make something pretty.

Seriously.

I don’t sit around manually constructing elaborate technical prompts or memorizing photography terminology. I use ChatGPT as a creative collaborator. We discuss the chapter, figure out what would make a good illustration, and work through the details together.

ChatGPT helps translate my ideas into visual instructions. I provide the creative direction, established references, and corrections when something isn’t right.

It’s not foolproof. There’s absolutely some trial and error involved.

But it’s a process that’s helped me turn written scenes into artwork while maintaining considerably more control over the results.

Here’s how I do it.

---

1. Don’t Start by Generating an Image

This is probably the biggest piece of advice I can give anyone.

Have a conversation with ChatGPT first.

I use ChatGPT as a creative partner. Before we generate anything, we discuss what I’m trying to illustrate.

And quite often, that starts with me simply giving ChatGPT the chapter.

Not a meticulously constructed image prompt. Not a detailed technical breakdown of camera angles and lighting.

The actual chapter.

I’ll upload my character references, tell ChatGPT that I want to create some illustrations, and we start talking about which moments would work best.

Sometimes I already know which scene I want illustrated. Other times, I ask ChatGPT to help identify the strongest visual opportunities.

Then we work through the visual details together.

Which moment in the scene would make the strongest illustration? Should the image be a close-up or a full-body composition? Where should each character be positioned? What’s the lighting? What’s happening in the background? What emotion should the illustration capture?

These things matter.

A dramatic confrontation, for instance, might work better from a low camera angle, while a quiet, intimate moment may benefit from a closer composition that emphasizes expressions and body language.

And sometimes, the scene you originally thought would make a fantastic illustration turns out to be less visually interesting than another moment from the same chapter.

You can figure that out through conversation without spending a single image generation on it.

You don’t even have to know what kind of composition would look best. That’s one of the things ChatGPT can help you figure out.

One of the simplest things you can tell ChatGPT is:

Here’s my chapter, and here are my character references. I’d like to create some illustrations for it. Help me identify the best scenes, figure out what would look good, and make sure the characters stay consistent with their references. Let’s discuss the illustrations before generating anything.

That’s it.

You can be much more specific if you want, but you don’t have to be.

The important part is making sure ChatGPT understands that you’re planning illustrations together rather than immediately requesting finished images.

You want time to develop the idea without accidentally turning every brainstorming message into another generation request.

2. Build Your Character References Before You Need Them

Character consistency is one of the biggest challenges with AI-generated illustrations.

And it’s especially frustrating when you’re illustrating a long story.

Imagine having twenty chapters illustrated with a particular character’s face, only for Chapter Twenty-One to introduce someone who looks like their distant cousin.

No, thank you.

This is why I maintain character reference images.

Whenever possible, I supply established visual references rather than relying entirely on text descriptions.

A character reference tells the AI much more than “a handsome man with brown hair” or “a woman with bright purple hair.”

It gives the generator a visual target for facial structure, proportions, hairstyle, complexion, and other distinctive features.

But here’s something important: not every reference image serves the same purpose.

A facial reference establishes who the character looks like.

An outfit reference establishes what they’re wearing.

A pose reference shows how their body should be positioned.

An environment reference establishes the setting or atmosphere.

These aren’t interchangeable.

If I upload a facial reference of a character wearing a leather jacket, that doesn’t necessarily mean I want the character wearing that jacket in every illustration.

Likewise, if I upload a reference image because I love the pose, I don’t necessarily want the AI reproducing the model’s face.

I make those distinctions explicitly.

For example:

Image 1 is the facial reference for Character A. Preserve their established facial features and hairstyle.

Image 2 is the facial reference for Character B.

Image 3 is an outfit reference only. Use the clothing, not the person’s face.

Image 4 is a composition reference. Use it to understand the pose and camera angle, not to replace either character’s appearance.

Simple? Absolutely.

Helpful? Also absolutely.

A useful newer feature

As of November 2026, ChatGPT Images also supports saved reference images, which can be named and reused in later conversations.

That could be particularly convenient when you’re illustrating a long-running story with recurring characters.

Just remember that a saved reference isn’t a magical guarantee of perfect visual consistency. You still need to inspect the generated result.

3. Treat Character Continuity Like Story Continuity

If you’re writing an ongoing story, you already know that continuity matters.

You wouldn’t deliberately give a character an injury in Chapter Seven and then forget about it in Chapter Eight.

Your illustrations should receive the same attention.

If a character has a broken arm, the illustration needs to reflect it. If their hair is tied back during a scene, it shouldn’t suddenly be loose. If their shirt was torn during a fight, it shouldn’t miraculously repair itself halfway through the chapter.

And if your story takes place in 1999, your characters probably shouldn’t have a modern smartphone sitting on the table.

I pay particular attention to details that image generators are prone to improvising: clothing, jewelry, injuries, tattoos, weapons, props, and the physical positions of characters.

The important thing is to tell ChatGPT which details are established facts and which ones it can interpret creatively.

For example:

The scene takes place immediately after a fight. Character A still has blood on their sleeve and a bandage around their right hand. Character B is wearing the same outfit described earlier in the chapter. Neither character has changed clothes. The setting is a small, cluttered workshop, not a hospital.

That’s much more useful than discovering afterward that the AI decided everyone needed a wardrobe change.

Your illustration should serve the chapter, not quietly rewrite it.

4. Let ChatGPT Help You Build the Actual Image Brief

Once I’ve settled on the scene and provided my references, I have ChatGPT help translate everything into a clear visual brief.

And here’s the best part.

You don’t need to write elaborate image-generation prompts yourself.

Half the time, my process is simply handing ChatGPT my chapter, uploading my character references, and asking it to help me make something pretty.

ChatGPT can identify the visual details, suggest compositions, and turn our conversation into image-generation instructions.

I don’t need to know every technical term for camera angles, lighting, or composition. I just need to know my characters and what I want the illustration to accomplish.

For instance, I might say that I want the illustration to feel claustrophobic, with the characters illuminated by computer monitors and the rest of the room disappearing into shadow.

ChatGPT can help turn that into more specific visual instructions.

Or I might simply say:

I really love this moment from the chapter, but I’m not sure how to illustrate it. What do you think would work?

And we’ll discuss it.

Sometimes ChatGPT suggests a composition I hadn’t considered. Sometimes I disagree with its suggestions and explain exactly what I want instead.

That’s the collaboration.

I’m not outsourcing my creative decisions. I’m letting ChatGPT handle the technical translation while I remain in control of the artwork.

The goal isn’t to make the prompt as long as humanly possible.

It’s to make the instructions specific enough to communicate what matters without burying those priorities under unnecessary detail.

Here’s the basic framework ChatGPT can help establish:

Subject: Who is in the illustration?

Action: What exactly are they doing?

Composition: Where are they positioned, and what does the camera show?

Character references: Which established appearances must be maintained?

Continuity: What clothing, injuries, props, or environmental details are mandatory?

Lighting and atmosphere: What should the scene feel like?

Format: Is this a vertical chapter illustration, a horizontal banner, or another intended layout?

Restrictions: What absolutely must not change?

You don’t have to fill out that framework like a questionnaire every time you want a picture.

You can simply provide your chapter and references, talk through your ideas, and have ChatGPT help identify the relevant information.

In many cases, the chapter itself already contains most of those details.

And because you’ve discussed the scene beforehand, you’re not trying to invent the entire illustration in a single message.

You don’t need to speak AI. You just need to know your story. Let ChatGPT help translate what’s in your head into something you can see.

5. Lock the Nonnegotiable Details

This is where I become very particular.

If something absolutely must appear in the illustration, I say so.

If something absolutely must not change, I say that, too.

Consider an illustration involving two characters sitting together on a bed.

It sounds straightforward.

But does the AI understand that both characters should be sitting? That one character is leaning against the headboard? That the other is holding a specific object? That they’re looking at each other rather than at the camera?

Not necessarily.

So instead of simply asking for “two characters sitting together on a bed,” I spell out the physical relationship.

Both characters are seated on the bed. Character A is leaning against the headboard, with Character B seated beside them and turned slightly toward them. Character A is holding the book in their left hand. Character B is looking at Character A, not at the viewer. Preserve their established facial appearances and clothing.

You don’t have to micromanage every square inch of the picture.

But if a detail is important to the scene, don’t leave it entirely to interpretation.

And when illustrating scenes with multiple people, clarity about who’s doing what becomes especially important.

6. Generate the Image—and Actually Examine It

Now we get to the fun part.

Once we’ve established the references, composition, continuity, and overall look, I tell ChatGPT to generate the image.

And then I inspect the result.

Not just whether it’s pretty.

Because AI can produce an absolutely gorgeous illustration that’s completely wrong for the chapter.

I look at the faces first. Are these recognizably the same characters?

Then I check their expressions, body positions, clothing, props, and surroundings.

Does the image actually capture the moment we planned?

Are there obvious anatomical mistakes? Did someone’s hand merge into another object? Is a weapon being held incorrectly? Did a piece of jewelry disappear? Did the AI change the lighting so dramatically that the intended mood is gone?

I also make a distinction between actual mistakes and details that are merely different from what I originally imagined.

Sometimes the AI makes a creative choice that works beautifully.

If that choice doesn’t contradict the scene, I’m perfectly happy to keep it.

There’s no point demanding a change just because I hadn’t thought of doing it that way myself.

But if something contradicts the chapter or breaks an established character’s appearance, that’s worth correcting.

7. A Very Important Note for Erotica Writers: ChatGPT Is Not an NSFW Image Generator

Since many of us write romance, erotica, or stories featuring mature relationships, I want to address something before we go any further.

ChatGPT is not a tool for generating pornographic or sexually explicit illustrations.

And no, being an erotica author doesn’t magically change that.

ChatGPT’s image-generation tools have content restrictions. They aren’t intended for producing graphic depictions of sexual activity, and attempting to bypass those restrictions isn’t a productive illustration strategy.

Establish Your Own Illustration Boundaries, Too

Here’s something else worth remembering: just because ChatGPT permits a particular type of image doesn’t mean you have to be comfortable creating it.

The platform’s content restrictions aren’t the only boundaries that matter. Yours do, too.

Personally, I prefer my illustrations to focus on the characters, their relationships, and the emotions behind the scene. Even when I’m illustrating a chapter containing explicit intimacy, I generally don’t want the accompanying artwork to be particularly revealing. I’d much rather let my writing handle those details and use the illustration to capture something meaningful about the characters themselves.

And there’s absolutely nothing wrong with telling ChatGPT that.

For example:

I prefer fully clothed, non-explicit illustrations, even when the chapter contains mature or intimate material. Focus on tenderness, expressions, body language, atmosphere, and emotional connection rather than exposed skin or suggestive posing. Please respect these preferences throughout the illustration process.

You can establish whatever additional boundaries make you comfortable. Maybe you prefer romantic illustrations without revealing clothing. Maybe you don’t want certain types of physical contact depicted. Maybe you simply want your artwork to feel more like illustrations from a novel than promotional images for adult entertainment.

Those are creative decisions you’re entitled to make as the author.

And remember, AI image generators aren’t perfect. Sometimes they’ll misunderstand instructions, exaggerate elements you didn’t ask for, or produce an image that’s more revealing than you intended.

If that happens, you don’t have to accept it, use it, or publish it. You can request a correction, rethink the composition, or discard the image altogether.

Your comfort matters just as much as your creative vision. You shouldn’t have to compromise either one to illustrate your story.

But here’s something I think writers sometimes overlook.

An erotic story does not require pornographic illustrations to be visually compelling.

Erotica and pornography aren’t interchangeable terms. Erotica is also a literary form, one that can explore desire, intimacy, vulnerability, relationships, character development, and emotional connection.

Not every meaningful moment in an erotic story takes place during sex.

And even when you’re illustrating a chapter that contains explicit material, you don’t have to depict the explicit action itself.

In fact, some of the most powerful romantic illustrations focus on everything surrounding intimacy.

Think about the possibilities.

The anticipation. Two characters standing close enough to kiss, holding each other’s gaze, neither quite willing to make the first move.

The tenderness. One character gently brushing the other’s hair away from their face, or resting their forehead against their partner’s.

The emotional vulnerability. A couple embracing after a difficult conversation, with expressions that communicate everything the dialogue couldn’t.

The chemistry. An amused glance across a crowded room, a hand resting at someone’s waist, or a quiet moment where two characters clearly understand each other without saying a word.

The aftermath. Two fully clothed characters sitting together in comfortable silence, sharing coffee, exchanging affectionate looks, or laughing over something only they would understand.

Those moments can be romantic, sensual, intimate, and visually gorgeous without showing graphic nudity or sexual activity.

And sometimes, they’re more revealing about the relationship than an explicit illustration would be.

Let the reader’s imagination do some of the work

As writers, we already understand the value of suggestion.

We don’t need to describe every single detail of an emotion for readers to experience it. We give them meaningful sensory information, establish the characters’ reactions, and trust them to understand the rest.

Visual storytelling works the same way.

Lighting, composition, body language, expressions, and atmosphere can communicate incredible amounts of intimacy.

A couple sharing a kiss beneath the rain can be a gorgeous illustration.

So can two people curled together on a couch, or one character looking at another with an expression that makes their feelings unmistakable.

The important question isn’t how much skin you can show. It’s what the image communicates about the characters.

If you’re illustrating an explicitly erotic chapter, consider choosing a non-explicit moment before or after the intimate encounter, or another emotionally significant beat from the same chapter.

You can also ask ChatGPT to help identify the strongest romantic or emotionally intimate moments that are suitable for illustration.

For example:

This chapter contains mature romantic material, but I want a non-explicit illustration that captures the relationship and emotional atmosphere. Help me identify a visually compelling moment that communicates their chemistry, affection, or vulnerability without depicting nudity or sexual activity.

That’s an excellent starting point.

And if a particular image request isn’t supported, work with ChatGPT to develop a different visual concept rather than treating content safeguards as a puzzle to defeat.

Remember what you’re actually illustrating

Your readers already have the chapter.

The illustration doesn’t need to reproduce every action in it.

Its job is to complement the writing, establish atmosphere, emphasize a meaningful moment, or give readers a clearer picture of the characters they’ve come to know.

Stories don’t need graphic nudity to have great visual storytelling.

Neither do their illustrations.

A little creativity, imagination, and understanding of what makes a scene emotionally effective can go a very long way.

And frankly, if an illustration communicates nothing about a relationship except that two people have taken their clothes off, I think we’re missing an enormous opportunity as storytellers.

8. Make Surgical Corrections, Not Unnecessary Rewrites

Oh, this one.

This is where things can get spectacularly frustrating.

You finally get an illustration you love.

The faces are right. The lighting is gorgeous. The poses are perfect. Everything looks fantastic.

Except somebody has an extra finger.

So you ask the AI to fix the finger.

And it gives you an entirely different picture.

The face changes. The outfit changes. The background changes. The beautiful lighting disappears.

And suddenly you’re wondering why fixing one hand required rebuilding the entire universe.

This is why I try to request small, targeted edits instead of broadly asking for an improved version.

For example:

Use the existing image as the reference for this edit. Correct only Character A’s right hand so the anatomy is natural. Preserve the faces, expressions, clothing, poses, lighting, background, colors, and composition. Do not redesign or reinterpret the scene.

If the image editor allows you to select a specific area, that can help communicate which part needs attention.

But here’s the catch: AI image editing isn’t always pixel-precise.

Even when you ask it to change one detail, it may modify surrounding areas or recreate parts of the image.

Newer image models have improved at preserving details, but no amount of emphatic prompting guarantees that every other pixel will remain untouched.

If exact preservation is essential, traditional image-editing software or compositing may be a better choice.

And sometimes the smartest decision is to leave a tiny imperfection alone rather than risk destroying an image that’s otherwise excellent.

9. Don’t Confuse Expanding an Image with Remaking It

This deserves its own section because it’s a mistake I’ve encountered repeatedly.

Let’s say you have a beautiful vertical illustration, but you need a horizontal banner for an article.

You don’t necessarily want a different illustration.

You want more canvas around the illustration you’ve already approved.

That’s called outpainting, or extending an image beyond its existing boundaries.

The distinction is important.

If you simply ask for a widescreen version, the AI may reinterpret the entire composition. Characters can change faces, poses, clothes, or even locations.

Instead, be specific:

Expand this existing illustration into a horizontal banner. Preserve the original characters, faces, poses, clothing, artwork, lighting, typography, and color palette as closely as possible. Extend the surrounding environment to fill the wider composition. Do not create a new interpretation of the scene.

And establish the intended aspect ratio before generating.

For a coordinated article or chapter series, I recommend selecting consistent image dimensions at the beginning rather than allowing every illustration to end up with a different shape.

There’s another important distinction here, too: telling an AI to preserve an original image isn’t the same as using a method that literally locks the original pixels.

When exact preservation matters, you’ll need to inspect the output carefully—or use a canvas-based workflow that preserves the original artwork and only fills the added space.

10. Know When to Stop

This might be the most underrated part of creating AI illustrations efficiently.

You have to know when the illustration is finished.

It’s tempting to keep tweaking.

Maybe the lighting could be a little better. Maybe you could adjust the framing. Maybe that tiny object in the background could move three inches to the left.

But every new attempt introduces the possibility of losing something you already liked.

I’ve learned that not every minor imperfection deserves another generation.

If the characters are recognizable, the scene is accurate, the composition works, and the image communicates the intended emotion, it’s probably time to move on.

That doesn’t mean settling for something you dislike.

It means knowing the difference between an important correction and endlessly chasing an imaginary perfect result.

Make the edits that matter. Leave the rest alone.

11. Save Your References and Successful Illustrations

When you’re illustrating an ongoing story, organization becomes just as important as image generation.

I strongly recommend maintaining a collection of approved character references, completed chapter illustrations, and any particularly successful compositions or visual styles.

Make it easy to distinguish final artwork from unsuccessful attempts.

And when you move on to another chapter, reuse the established references whenever they’re relevant.

You shouldn’t have to recreate your protagonist’s face from scratch every time they appear.

The same goes for recurring settings.

If a particular room, vehicle, apartment, or workshop appears throughout the story, an approved visual reference can make maintaining its appearance considerably easier.

Your existing images are valuable creative assets. Treat them that way.

12. What About Usage Limits and Image Generation Costs?

This is worth mentioning because ChatGPT’s features, subscription plans, and usage limits can change over time.

Image generation is subject to usage limits, and the exact allowances depend on your account, plan, and the features you’re using.

That’s one reason I prefer doing as much preparation as possible through conversation before generating the actual illustration.

Discussing an idea, checking continuity, and refining a proposed composition can eliminate a lot of unnecessary image attempts.

But I also don’t want to imply that every image needs an absurdly elaborate planning session.

Sometimes you know exactly what you want.

In that case, provide your references, explain the scene, and generate it.

Efficiency doesn’t mean making the process complicated. It means avoiding work that doesn’t improve the result.

For the latest information about ChatGPT’s image features and limits, consult OpenAI’s official documentation rather than relying on outdated tutorials or social-media rumors.

13. Let ChatGPT Be Your Creative Partner, Not Your Creative Director

This is ultimately the philosophy behind my entire process.

I don’t use ChatGPT because I want it to take over my stories.

I use it because I already know my characters, my plot, my settings, and the emotions I want to communicate.

What I need is a tool that helps translate those ideas into images.

Sometimes that means having ChatGPT suggest camera angles I hadn’t considered.

Sometimes it means asking whether a particular composition is physically possible.

Sometimes it’s reviewing an image and realizing that an important detail from the chapter has been overlooked.

And sometimes it means telling ChatGPT, in no uncertain terms, that the character’s original face was perfectly fine and it needs to stop fucking changing it.

That’s all part of the process.

The key is maintaining creative control.

You’re the author. You know the story.

The AI can help you visualize it, but it doesn’t get to decide that your characters should look different, behave differently, or inhabit a completely different scene simply because it thinks the alternative would be prettier.

---

My Actual Workflow, Simplified

For anyone who wants to try this approach without getting overwhelmed, here’s the short version:

1. Give ChatGPT your chapter. Share the scene you want illustrated, or ask it to help identify the strongest visual moments.
2. Upload your character references. Establish who your characters are supposed to look like, and identify any additional clothing, pose, or environment references.
3. Talk about what you want. Discuss the mood, setting, character positions, and overall composition. You don’t need technical terminology.
4. Let ChatGPT help develop the image brief. It can identify relevant details from the chapter and translate your creative decisions into generation instructions.
5. Check the important details. Make sure the characters, clothing, injuries, props, and setting match the story.
6. Generate the illustration. Once you’re happy with the concept, tell ChatGPT to make the image.
7. Review and correct. Check the result against your chapter, request targeted fixes when necessary, and recognize when another generation would do more harm than good.
8. Save the final artwork. Keep your approved images and character references organized for future chapters.

That’s it.

Not a mystical secret prompting formula.

Not a requirement to become a professional graphic designer.

And definitely not a requirement to become an expert in prompt engineering.

Just a practical process that puts your story first and lets ChatGPT handle some of the technical work.

Ready to Try It? Here’s Your Starting Point

Now that we’ve covered the process, you might be wondering how to actually get started.

Good news: you don’t need to memorize everything in this article, learn complicated prompting techniques, or write an enormous list of technical instructions before you can make your first illustration.

Remember, half the time I start by handing ChatGPT my chapter, uploading my character references, and asking it to help me make something pretty.

You can do exactly the same thing.

I’ve put together a simple starter message you can copy and paste into a new ChatGPT conversation. It establishes the basic workflow, encourages collaboration, and reminds ChatGPT to focus on your story and characters rather than inventing details you never asked for.

Copy and paste this into ChatGPT to get started:

[I’m going to share a chapter from my story along with my established character reference images. I’d like your help creating chapter illustrations that accurately reflect the story.

Please read the chapter, identify a few visually compelling moments, and help me decide which would make the strongest illustrations. Preserve my characters’ established appearances and the chapter’s continuity, including clothing, injuries, settings, and important props.

I prefer non-explicit illustrations that focus on storytelling, character relationships, atmosphere, and emotion.

We can discuss composition, lighting, and poses together. Please don’t generate any images until I explicitly tell you to. I don’t know technical image-prompt terminology, so help me translate my ideas into visual instructions.]

And that’s your starting point!

Once ChatGPT responds, share your chapter and upload your character references. You can tell it which scenes interest you, ask for suggestions, discuss possible compositions, or simply explain what you’re imagining.

You don’t have to follow a rigid formula, either. Adapt the message to your own preferences, writing style, and creative process.

And if you already know exactly which scene you want to illustrate? Tell ChatGPT! There’s no reason to spend twenty minutes choosing a scene you’ve already chosen.

The goal is to make the process easier, not bury yourself beneath another mountain of instructions.

You’re the author. Bring the story. Let ChatGPT help you bring the pictures to life.

Final Thoughts: Trial and Error Is Part of the Process

I’d love to tell you that following these steps will produce perfect illustrations on the first attempt, every single time.

It won’t.

Sometimes an AI will misunderstand a pose. Sometimes it’ll forget a prop. Sometimes it’ll produce something beautiful that bears absolutely no resemblance to the scene you described.

And yes, sometimes you will spend an unreasonable amount of time trying to convince it that your character has the same damn face they had five minutes ago.

But the more consistently you establish references, communicate priorities, and check the results against your actual writing, the easier it becomes to recognize problems before they spiral into endless regenerations.

The point isn’t to eliminate experimentation.

It’s to make your experimentation productive.

Your chapters already contain the moments, emotions, and characters that make your stories worth illustrating.

ChatGPT can help bring those moments to life.

Just remember who’s in charge of the story.

And for the love of all things literary, save the version you actually liked before asking the AI to fix that fucking hand.

— HAWK