How to Create a Consistent Character in Comic Strips with AI: A Beginner’s Guide
Discover 5 unique tips that will help you create an amazing story with a consistent main protagonist.
As an avid enthusiast of artificial intelligence (AI) and its creative applications, I have immersed myself in the world of digital tools since the launch of ChatGPT (Chat Generative Pre-Trained Transformer) in November 2022. While my journey began with conversational AI, my passion, however, lies in the realm of AI art, and I enjoy exploring AI art generators such as ArtSpace.ai, Midjourney, DALL-E 2, Bing Image Creator, and Leonardo AI during my leisure hours.
One intriguing aspect of AI art that perplexes me is its capability — or lack thereof — to consistently create a character or characters across diverse scenarios.
So, I decided to learn more about this topic. In this article, I will share with you what I found, and guide you through the fascinating process of harnessing AI to create consistent characters, especially for comic strips or books.
Please note that I’ve used ArtSpace.ai, a powerful AI tool for art that I really enjoy, in all my experiments. However, the methods don’t differ much across all AI art generators.
What Exactly Is Character Consistency?
In comic strips, keeping a character consistent means their outer appearance, or looks stay pretty much the same, about 80% similar, no matter what’s happening around them. This matters because as readers follow the main character’s journey, it helps them connect and believe in the story. Consistency makes the character more relatable and the story more professional and realistic too.
How to Use AI to Create a Consistent Character in a Comic Strip
In order to create a consistent character in a comic book or comic strip, here are two main methods which had worked for me so far.
(1) Using the Same Prompt But with Variations
One of the most popular methods of generating a consistent character with AI for different scenarios is to use the same prompt but with variations. What do I mean? Let me explain further. For example, here is my master prompt:
Blaze the white mouse, [wearing a sleeveless yellow vest with red tie, is in his car driving home after work]. Draw this in ghibli-style.
You can include negative prompts like “text,” “logo,” or “words” to prevent them from appearing in the image. This won’t significantly impact the overall result.
Here is what I’ve generated using ArtSpace:
Please note that you may need to generate a few times to get the correct image that you want.
Once you’ve generated an image using the initial prompt, you can change the prompt text highlighted in the brackets above to generate another image. The important guideline here is not to change the prompt too drastically, or else the result will not be ideal. For example, the second prompt can be:
Blaze the white mouse, [wearing a sleeveless yellow vest with red tie, is at the gym working out]. Draw this in ghibli-style.
Here is the second image that I’ve generated (remember to remove the brackets [ ] above! before you start generating the images!). You can see that the facial features and overall look of the mouse is similar to the initial character in the master image.
* Please note that all these images have not been digitally altered in Adobe Photoshop or other imaging software so you may see the usual AI defects in areas such as the hands or feet.
Now, let’s try a third prompt. You can use any of the original prompts to try, such as the initial prompt, or the second or third prompt we’ve used before:
Blaze the white mouse, [wearing a yellow pyjamas, is at home getting into bed.] Draw this in ghibli-style.
Here’s the third image. Notice how the original character in all three images looks quite similar? You can keep generating more images using the same method, with slight variations in future prompts.
(2) Using a Seed Number or a Seed Image with Different Prompts
The second way to create a consistent character with AI is to use a seed number or a seed image but with slightly different prompts. However, I will prefer to use a seed number rather than a reference image. Let me explain further why.
(a) Using a Seed Image
First, let’s use the original image that we have created above as the reference image here:
Let’s take one of the previous prompts to use, such as this:
Blaze the white mouse, wearing a sleeveless yellow vest with red tie, is at the gym working out in weight lifting. Draw this in ghibli-style.
The image above is what I generated with the seed image. It maintains the same pose as the seed image, and even after multiple attempts with different prompts, the pose doesn’t seem to vary much. This issue with seed images isn’t exclusive to ArtSpace; I’ve encountered it in other AI art generators like Leonardo AI as well.
(b) Using a Seed Number
Now, let’s attempt to create a consistent character using a seed number. I’ll use the same seed number as the master image above, which is 1985894727, along with the same prompt:
Blaze the white mouse, wearing a sleeveless yellow vest with red tie, is at the gym working out in weight lifting. Draw this in ghibli-style.
Here’s one of the better images I generated after a few tries. It has some defects and errors, but nothing a Photoshop expert can’t fix.
(3) Using a Specific Prompt In A Single Image
I learnt this prompt trick from Wei Mao. While he achieved character consistency using DALL-E, I’ve tried it with various AI art tools. Although some editing is still required, the approach generally worked well for me.
Using this method, you need to specify a prompt in a particular format. I’ve amended his template a bit to suit my own use below.
A [layout] of [describe your main character] [top left description] [top right description] [bottom left description] [bottom right description]
Layout above refers to how you want your collage to be arranged. Wei Mao suggested that you can use phrases such as ‘photo montage’, ‘grid’, ‘arrangement’, ‘collage’, ‘quad-diptych’, ‘storyboard’, ‘panorama’, ‘split-screen’, ‘mosaic’, ‘film strip’, ‘comic strip’, etc.
Using the above, my prompt can be:
A photo grid of Blaze the white mouse, wearing a sleeveless yellow vest with red tie. Top-left shows him reading a book. Top-right shows him eating a chocolate bar. Bottom-left shows him jogging. Bottom-right shows him playing guitar at home. Draw everything in ghibli-style.
The following is what I obtained from ArtSpace.ai. You definitely need to do a little bit of editing here and there, but in general, the character remains consistent and most of them are good enough to use.
(4) Generating Multiple Expressions and Poses In a Single Image
Here is a 4th method by Christie C. Although she tested it using Midjourney, I’ve also tested it using other AI art generators such as ArtSpace.ai and it worked similarly, though the processes of getting your final consistent characters might be slightly different.
I love it that Christie’s concept is easy to understand for beginners, and you can also create a consistent character with multiple expressions and poses. Refer to her article’s Part 3. First, you need to use a master prompt to generate a character with multiple expressions and poses by using specific keywords like ‘multiple poses, facial expressions’ in your prompt. After that, you just need to take each individual expression and pose, mask all the other expressions and poses, and use ‘Vary Region’ with prompts to regenerate the background of each image. Next, just select another expression and pose for the next scene, and do the same process again.
If you are using other AI artwork generators, you can generate your main character in different poses and expressions, and then either use inpainting or Photoshop to regenerate a different background or surrounding. For example, my prompt was:
Blaze the white mouse wearing a red t-shirt, multiple poses, facial expressions, plain background. Draw this in ghibli-style.
Here is what I obtained from ArtSpace.ai:
(5) Using GPT Apps from the ChatGPT Store
Another method works only in ChatGPT with Dall-E 3, using GPT applications from the ChatGPT store.
According to Jim Clyde Monge, we can use GPT applications that creators made to create consistent characters. In his recent article, he showed us how he created a consistent avatar based on his own image using an aplication called ‘Consistent Character GPT’. You may want to read more about it here.
Tips for Creating Better Consistent Characters
To further maintain character consistency, here are some tips that I’ve learnt.
(1) Do Not Change Your Art Style
Stay consistent with your art style when generating different images. For instance, if you’ve been using a Ghibli-style in one prompt, avoid switching to another art style in another prompt.
(2) Do Not Use Vastly Different Words In Different Prompts
Also, try to keep the words in each prompt relatively consistent. For instance, avoid changing “mouse” in one prompt to “rat” in another, even if they refer to similar items.
(3) Fix Errors with Inpainting, Outpainting or Photoshop
Sometimes, your AI-generated images may have minor unwanted elements or random noise. To address this, you can utilise inpainting or outpainting features provided by most AI image generation tools to clean up the images. For instance, in ArtSpace, simply highlight the area you want to fix with a lasso, click the inpainting or outpainting button, and it will brush over the parts that need fixing. Alternatively, Photoshop experts can use their skills to refine the images.
(4) Generate As Many Images As Possible
Generate various images to understand how your AI art generator operates. As you become familiar with these tools, your ability to create consistent images will improve as you perfect your creative process.
Update (15 April 2024):
Midjourney recently released a ‘Consistent Character’ feature. You can now use the same character across multiple images and styles without deviating too far from their original design. You can find out more from Midjourney’s Discord channel.
Conclusion
Creating character consistency or consistent character designs with AI requires some trial and error, and a lot of patience and creativity. While tools like Midjourney, Stable Diffusion or Leonardo AI may have slight variations from the ArtSpace that I’m using here, and you can even train your AI to generate the same character with them, the two core concepts I’ve highlighted above apply universally, and can help you achieve better results. I won’t dive into overly technical details here, as I want to approach AI from the viewpoint of a novice.
So that’s all for today. As mentioned, I’m still in the early stages of learning about AI, absorbing new knowledge daily. If you have any suggestions on enhancing the output of a consistent character or improving this article, feel free to share your input below!
P.S. For further insight into how I create these remarkable projects, take a look at this detailed review of ArtSpace.ai.
Related Stories:
Want more of such stories? You can follow me here, or find out more here.
👉Join a global community of 1,000+ marketers and business owners to receive two monthly marketing strategies for small business growth, covering topics that I’m very passionate about such as China marketing (Chinese market marketing), content marketing, SEO, GenAI, metaverses, and social media. Sign up here.👈
Disclaimer: This article contains affiliate links. I may earn a small commission, at no extra cost to you, if you click through my links and make a purchase.
