top of page
Search
  • Writer's pictureHarshal

From Idea To Illustrated Storybook: My 5 Steps With ChatGPT And Midjourney

Updated: Sep 28, 2023

Celebratory Charm, Computational Craft, And… Clearly Challenging - Part 1


Do you like receiving gifts? Do you appreciate personalized gifts, like hand-stamped metal jewelry, a hand-chosen antique from a night market, or a custom woodwork box? Like many, my wife and I have a soft spot for personalized gifts. But, I don’t have the skills to craft such gifts. So, I tried a different approach. Below is my chronicle of learning new tools to create a new type of gift I have never created before.


For my wife's birthday and our anniversary, I decided to use ChatGPT and Midjourney to make a gift. We both love these tools, so I wanted to make something special with them. I aimed to craft something reflecting our shared experiences, hobbies, and even quirks.


I learned a lot, and I’ll share my process and learnings below. If you want to create a personalized storybook for your children, parents, or significant other, stay with me. Stay with me and discover how 2023’s technology simplifies creating personal treasures.


Human and Robot co-creating a storybook.
Human and Robot co-creating a storybook.

The 3 Pillars You Need For Personalized Gifts

I thought of a framework. You need 3 core elements for a personalized gift.

  1. A Vanilla Gift - This could be anything from a coffee mug, a T-shirt, or a water bottle, to perfumes and books.

  2. Understanding Preferences - It's pivotal to know the likes and dislikes of the recipient to ensure the gift resonates with them.

  3. A Technology - A tool to blend the first two components. This could be software but also be your woodworking handsaw or 3D printer.

Personal Preferences, Vanilla Gift, and Technology are required for a personalized gift.
Personal Preferences, Vanilla Gift, and Technology are required for a personalized gift.

In the past, my wife and I have gifted each other personalized coffee mugs, T-shirts, book covers, and more. But after getting comfortable using generative AI or large language models (LLMs), I realized the potential to go beyond usual personalized gifts.


ChatGPT’s Take On Personalized Gift Ideas

My started my project by seeking gift recommendations from ChatGPT. I wanted to know about gifts that could be personalized. I provided some of my wife's preferences.


Give me ideas to celebrate my wife's bday where I can use ChatGPT and Midjourney AI tools to do something for her. My wife loves food touring, harry potter universe, poetry, crime stories, movies, and ice cream. She doesn't like nature, traveling (unless for food touring), fashion, recipes.


The suggestions I received varied from interactive game, a photo album, to an animated chatbot. Although crafting a DIY story initially seemed overwhelming, the concept of an illustrated storybook, inspired by the endeavors of some community members, seemed like the perfect fit.


Craft a mini crime story where the two of you are detectives searching for a stolen, legendary ice cream recipe in a setting inspired by the magical world of Harry Potter.


How To Use Generative AI Tools For The Best Creative Output?

I choose to generate a story using ChatGPT and create illustrations for it using Midjourney. My goal was to minimize using other technology or my knowledge and maximize using only these 2 tools. I asked it for the next step.


help me do #1. what should I prompt chatgpt? How should I break it down into parts to improve output quality? Let’s Think Step by Step


A few tips I learned:


1 - Ask questions you cannot ask Google. For example, you can ask Google for “Give me ideas to celebrate my wife's bday” or even “Give me ideas to celebrate my wife's bday where I can use ChatGPT.” You might find some SEO-optimized blogs for it. But you cannot ask it to give ideas based on her interests.


2 - During my interactions with ChatGPT, I noticed its tendency to occasionally "hallucinate". It may offer short-term predictions rather than a coherent, overarching narrative. This is because of its autoregressive nature — where each subsequent word depends on prior context. Solution/ It's essential to make the LLM not only provides relevant responses but also lay the groundwork for comprehensive answers. It's analogous to constructing a scaffold before diving into the intricate details of an artwork. Ask ChatGPT to build a scaffolding for its answer before it generates the answer. Manuel Odendahl has a detailed blog on this on The Scapegoat Dev.


3 - When not sure how to do #2, ask ChatGPT. When unsure how to ask ChatGPT, tell it to think step by step. One of my most used keyboard shortcut prints “Let’s Think Step by Step”. You can read more about One-shot reasoning and improving ChatGPT’s accuracy in this research paper (I haven’t) or this github cookbook (I like this).


ChatGPT - Person Or Software? Collaborator Or Tool?

ChatGPT - Person or Tool? Can Turing Test differentiate between a mix?
ChatGPT - Person or Tool? Can Turing Test differentiate between a mix?

Sometimes I find it hard to say "I did x" because ChatGPT helped me. At the same time, I find it hard to say “ChatGPT did x” because it did not do anything on its own. In reality, it was a combination of what I did and what ChatGPT helped with. It was a collaboration.


Let’s call ChatGPT C. Surprisingly, I’ve seen a few early adopters name ChatGPT when working with it on a big project. Working with C feels like having another collaborator by my side — a blend of both software and a human touch. In the rest of my article(s), I'll refer to ChatGPT as C and use ‘it’ pronoun. But, I’ll also give its full name once in each section because a reader may skip to some section.


Also, I will not share every prompt I used, but I will cover some of the early ones. As we move through the process, I'll highlight how I formulated these prompts without delving too deep to keep the article concise. It will still be a multi-part article.


A 5-Step Scaffold For A Personalized Gift

I asked C (ChatGPT) to guide me in crafting the storybook. It offered a 5 step process.

  1. Formulate a Story Outline: Use ChatGPT's assistance to create a compelling storyline.

  2. Create Visual Elements: Design illustrations relevant to the story using Midjourney.

  3. Merge Storyline and Visuals: Combine the text and visuals.

  4. Set an Immersive Mood: Dim the lights. Play enchanting background music. Consider dressing up when giving or reading the gift.

  5. Enjoy a Post-Story Treat: Consider having a special ice cream treat after reading the story.

I wasn’t interested in step 4 or 5 at this stage, but I needed more help on formulating the story.


A 6-Step Framework For Crafting The Story

Together, C (ChatGPT) and I developed a 6 step framework:

  1. Story Premise

  2. Character Development

  3. Setting and Locations

  4. Clues and Crime Details

  5. Major Plot Points

  6. Conclusion and Resolution

Initially, C gave me five overarching steps. But to accomplish just the first of those five, I was handed another six. What dawned on me was the layering. Things were getting complicated and we hadn’t even started.

Exploding steps. Each step results in 5 more steps.
Exploding steps. Each step results in 5 more steps.

Story Skeleton: AI-Generated Plot, Locations, And Characters

Utilizing C (ChatGPT) to develop a plot premise worked very well. I adopted most of the plot it generated. Now, I had an overarching story arc, the locations, and the main characters.


Help me create a premise for a Harry Potter universe-inspired crime story. The main plot revolves around a stolen legendary ice cream recipe from Florean Fortescue's Ice Cream Parlour in Diagon Alley.


I also needed 3 - 4 locations within the Harry Potter magical universe for key events. After providing C the primary plot details in an independent chat, it generated further information on each location.


For character development, I wrote personal details about both my wife and I.


Provide character profiles for two magical detectives in the Harry Potter universe. One should be based on me and the other on my wife. Include names, backgrounds, magical abilities, and personalities.


me: name starting with H. analytical, wears a specs, black tossled hair, look at lot of data, slow to decide, slow to think, speaks less, martial arts, musician.


wife: name starting with S. emotional, long black hair, detective, can unearth any information, fast to respond, dancer, caring, loving, doesn't like touching plants.


Describe the antagonist of the story: someone from a rival ice cream shop who might have motive to steal the recipe. Provide name, background, motive, and distinctive traits.


I did not like the initial names C provided, so I requested alternatives using our initials HP and SS.


Harry Potter fans may have heard the troupes that the sinister characters often come from Slytherin and the brave from Gryffindor. The heroes are Gryffindor, and their sidekicks are from Ravenclaw. I do not like these troupes, so I adjusted the character narratives. Read more about Hogwarts houses here.


The Story Wasn’t Complete Yet

In my project to craft a magical narrative, C (ChatGPT) worked to lay out an intricate skeleton. I got plots in the grand universe of Harry Potter but adapted to my purpose. But, as with every artist and their toolkit, there were hits and misses. In my next article, I write about:

  1. Navigating AI's Creative Flow: From overwhelming ideas to the intricate details that were challenging to use.

  2. Structuring the Narrative: Merging the introduction, aligning C's output, and summarizing work to pave the way for future sections.

  3. Infusing Personal Touches: Including a sprinkle of our Irish home, personal traits, and combatting gender biases in C (ChatGPT).

  4. The Art of Editing: 4 challenges I faced in refining the plotline and addressing issues in the story.

Read more here.


76 views

Comments


Commenting has been turned off.
Post: Blog2_Post
bottom of page