The latest additions to the regiment of modern technologies can provide significant advances in the creation of the latest Ukrainian book. Ukrinform reports The story of one of the situational developers of the children’s book published on the DOU resource:

– Hello. I am Yura Dzyuban, backend developer of Master of Code Global IT company. This article is an example of how to combine ChatGPT and Midjourney to create a picture book. In my case it was a children’s picture book, but with minor modifications the approach should be suitable for creating comics or, for example, illustrated presentations.

I am a backend developer and never an illustrator, so this article also shows what the average person can achieve in terms of graphic design with modern AI tools and a relatively small investment of time and effort. See below for approximate estimates of resources spent (time, money) as well as a photo of the final result.

BOOK CREATING: BASIC STEPS

1. Description of content: characters, scenes, dialogues

Let’s say we are interested in the idea of ​​​​creating an illustrated story (book, comic, presentation). The first step would be to logically understand exactly what we are going to draw – who our characters are and what our scenes will be – because then these images need to be translated into request format for models to create images and text.

I see several options for coming up with an idea for the visualization:

• If the goal is primarily to draw in the Middle of the Journey (or Dall-E, Leonardo.ai, etc.), you can illustrate an existing story (like Alice in Wonderland or something, some historical event, etc.). At most, if it’s a well-known story, the related images are most likely to be among the images Midjourney has trained on, which should make the graphics easier to create.

  Finally, you can repeat some of the existing examples to learn how to work with Midjourney.

• Don’t worry if there is no ready story, ChatGPT will help us here. You can create a story about a specific topic by instructing ChatGPT to respond as a storyteller (ChatGPT sample chat). You can also specify descriptions of locations, characters in Midjourney, in dialog format with ChatGPT, and select keywords for requests.

• Finally, you may already have a certain story and vision for the characters to tell. In my case it just so happened that 2 goals came together – I wanted to get to know Midjourney and also create something like a “spoiler” for the kids before giving them a dog. Children have been asking for different animals for a long time, and this book is actually an example of our dialogues.

After we pass this step, we must create a list of the characters and scenes we will draw. In my case, the characters consisted of my children, my wife and I, and various animals (horse, crow, lizard, dragon, squirrel). A picture of a gift box was also required.

Part of the story and design is the dialogue between the characters. In my case, children ask their parents to get pet X, and parents stubbornly refuse on various excuses. The final scene – the parents say the best animal was found, but what exactly – it will come as a surprise.

2. Content creation

We move on to the creation of text and visuals, having the idea of ​​\u200b\u200bdialogue as well as descriptions of characters and scenes.

We will create dialogs using ChatGPT. I should point out that this is almost the easiest part of the project and takes about 10-15 minutes. All the dialogues in my story were created using a single template – “write me a sentence in children’s book style <далі йде конкретна тема>» – please see ChatGPT chat and screenshot for example below.

After creating dialogs in ChatGPT I made the following plan. It seems that our book will have 7 forms: 1 with a “tie”, then 5 with different animals, and the last with the promise of gifts. You can proceed to create the necessary images.

Creating the images in Midjourney was the hardest part of this project for me; The process often evoked connotations with the following meme:

Let me remind you that Midjourney is one of the most popular and effective publicly available services for creating images based on text queries (alternatives are DALL-E, Stable Diffusion, Leonardo.ai and others). The service is paid, with monthly plans starting at $8 (they promise ~200 images creation in return). Interaction with the model takes place via the Discord bot.

Requests to the model consist of:

• eg teams /imagine to create an image or /settings to call the settings menu;

• for example keywords that describe the essence and style of the required image little girl character, multiple poses and expressions, children’s book illustration style, full body, character sheet, simple, cute, 6 years old girl, full color, blue children’s clothes, blond hair, solid color;

• Optional additional parameters that specify the version of the Midjourney model (for example, --v5.2), aspect ratios (--at 4:3 am), “minus words” (--no text, font, letter, watermark word, typography, slogan, signature to avoid text on images) and much more;

• requests may include [посилання на] image.

Mid-journey quests are a skill with elements of art. For example, there are many subtleties and techniques for creating consistent characters (using the parameter). -seedthe query is a series of links to pre-selected images, etc. including).

The model creates 4 image variations for each request and generates 4 pairs of buttons (U1/V1 – U4/V4) for scaling or creating variations of related images.

The general approach to production is as follows (and similar to development in general): we try a particular request, if the result is not satisfactory, we make minor changes to it (guided by our own logic, techniques, examples of other people, etc.). ) and repeat until you get an acceptable result.

It was easiest to create animals, especially dragons and lizards. The children were a little harder to come out of, and in the case of parents/adults the result was generally somewhat inconsistent with expectations (I dare to assume this may be due to the fact that there are “significantly more animals and children in the children’s group”). illustration book” model more than adult characters).

Another problem with rendering in Midjourney is that it’s hard to stop as the next query might work even better 😉

Finally, after about 3 hours of active correspondence (over several days) and ~220 requests with the Midjourney bot, I got images of all the characters and objects I had planned.

We enlarge the selected images, save them locally and move on to the next stage.

3. Processing, assembly, printing

It is recommended to scale the image further to get clearer images and better crop the background of the image. The search reveals a significant number of upgrade services, both free and paid (usually with a trial package). I used Pixelbin.io, which has a very user-friendly interface and promises x4 scaling.

In this way, the images recorded from Midjourney and measuring approximately 692×692 pixels were enlarged to 2768×2768 pixels.

After several hours of fairly routine manipulations in Photoshop (alternatively Gimp, pixlr.com or the like would do this), basically concluded:

• cut images from their background;

• finger and pupil artifact corrections (Mid-journey weakness, at least up to v5.1);

• composition of individual images on spreads, their relative scaling, time, reflection;

• work with text. The font Henny Penny was used in the inscriptions (it is a pity that there is no such font for the Cyrillic alphabet).

The original idea was to design this work in the form of a photo book. Therefore, even at the stage of creating a new document in Photoshop, you should understand what size the pages of our photo book will be in order to set the desired canvas size (given that the resolution of the images for printing is 300 pixels per inch) .

One of the mistakes I made in the context of printing was leaving too little space between the images and the page margins and this needed to be fixed.

After that, I used one of the home services for photo printing, selected the parameters of the book (dimensions, materials) in the online editor and placed the pictures on the spreads. I received the printed book 5-7 working days after ordering.

RESULTS

In addition to stock images, a photograph of the printed book can also be viewed in this repository. Below are a few selected photos:

Time required for the project: in my case, figuring out how to draw in Midjourney and completing the steps above took about 8-9 hours of work (over several days) – about 1/3 of learning, rendering and editing.

Cash costs: 0$ to ~30$+

• $8 for a one-month Basic Plan on Midjourney (by the way, I only spent 50% of the quota to create ~220 images). This spend point can hypothetically be reduced to 0 using Stable Diffusion, Leonardo.ai or other free alternatives;

• ~$18 (650 UAH) to print a photobook (optional);

• Other possible costs are fees for services for scaling images, software for rendering graphics, but in both cases there are enough free options.

It was an interesting and sometimes difficult (for Midjourney familiarity) experience; I think I will need this in future projects that involve creating visuals for presentations and converting text to images.

Using ChatGPT in conjunction with Midjourney opens up new possibilities and, if approached correctly, can provide a synergistic effect that is impossible to achieve using them alone.

Well? and I think the announcement of the gift was successful 😉

PS If you prefer to receive such instructions in video format, please watch the detailed video (in English) on YouTube.

YOU