How I Used AI to Reimagine Harry Potter Characters as Japanese Anime | by Krupesh Raikar | Aug, 2022

To inspire you with my mindblowing creative journey with a text-to-image AI

Paint a word picture so vivid that I needn’t even draw.

I am sure you have seen a crime series where the witness describes how the perpetrator looks to an artist.

The artist then pieces a sketch together only from the bits and pieces of descriptive information about an unknown face.

And, I find it hard to believe how accurately the artist ends up sketching, only based on these descriptions.

But imagine if that artist were AI!

This is similar to what I experimented with this week with my access to Midjourney — giving text to the AI and getting an image out of it.

For those who haven’t yet heard of it, Midjourney is the name of a research lab and its AI program that creates images from descriptive ext. The Economist used it to generate their cover image in June 2022.

It works as a bot in your Discord server. You have to sign in and get the Discord invite. Then you can add the bot to your server as well. They have a very concise tutorial on how to get started here.

In this post, I have tried something unique — using Midjourney to recreate Harry Potter characters as Japanese Anime.

This post is meant to inspire you, induct you into the fascinating world of creative AI, and share the lessons I learned.

Hence, I have posted all the prompts I have used along with the generated image. You are free to try the same prompts or use variations of them.

For each prompt, the AI generates 4 images — you can either create a variation of the one you choose or upscale the one you like. It is pretty simple.

So sit back and enjoy the images!

First off, just a quick background on the art style I used for the images. I have used something called the Japanese Moe Art Style for generating all the images.

This is done by using moe art in all the prompts. I am sure, even if you have not heard about this art style, you definitely recognize it from some popular animes!

Note — All the prompts in Midjourney begin with the command imagine, like this:

/imagine prompt

Hence, in all examples below I have only mentioned the prompts to avoid repeating the command.

Let’s start with our Hero!

1. Harry Potter

Is one of the most recognizable fictional characters in the world. So the AI worked well even without too detailed prompts. I chose the one I thought looked most fitting (same for all images that follow).

Prompt:

harry potter face + moe art style + anime  --ar 16:9 --uplight

2. Hermione Granger

Just as for Harry, I did not need any more detailed description for Hermione.

Prompt:

hermione granger + Moe art style + anime  --ar 16:9 --uplight

3. Ronald Weasley

In this case, I had to generate more variations with the same prompt to get Ron correctly. Finally, I settled on the image below.

Prompt:

ronald weasley + Moe art style + anime  --ar 16:9 --uplight

4. Severus Snape

I used several more detailed prompts for Snape, but the simple one seemed to generate the best results.

Prompt:

snape + moe art style + anime --ar 16:9 --uplight

5. Albus Dumbledore

Dumbledore is my favorite character from the series. So I wanted to create a more detailed anime version of him. But for some reason, Dumbledore was difficult to generate with the simple prompts as used for Harry. Each time the AI messed up with the beard (I can’t imagine why). So I tried with a more detailed one here, and got a somewhat satisfactory result.

Prompt:

old wizard, dumbledore, benign, smiling, long white beard and hair + moe art style + anime --ar 16:9 --uplight

6. Lord Voldemort

In the case of He-Who-Must-Not-Be-Named, I used several additional details to get a scarier image. There were many generated before I finally chose this one below.

Prompt:

voldemort + moe art style + anime + scary scarlet eyes  + slit pupils + forked tongue + symmetrical bald face --ar 16:9 --uplight

1. Text-to-image tools are impressive

The image generation AI has exploded in recent months. And I can’t help but get impressed!

• Whether it is Dall-E 2 or Midjourney, I found them really quick in generating awesome images from the prompts given to them.
• The AI is pretty good at cross-pollinating and superimposing art styles from one to another. For example (try it out yourself)— a realistic Eiffel Tower portrait in Van Gogh style.
• The AI works well to create an image that can later be fine-tuned and modified later.
• You can use the AI to generate surreal scenes and landscapes — which you can use as inspirations for creating scenes for video games and even movies!

Despite all the advantages, I must confess this: though I called them creative AI at the beginning of this post — I believe calling it such is misleading, at least at the current level.

This brings me to my next takeaway.

2. AI image generation is difficult

Even though it seems very simple in this article, generating all those images was time-consuming, to say the least!

You can never arrive at a perfect prompt without trial and error. And even with the same prompt, the AI will not generate the same images each time.

Look closely at all the above images and you will see that they are not exactly correct. Something seems off — like the eyes do not match, and overall they look a bit weird.

The true art in this case is getting the prompts correctly!

And this is where humans still rule — creativity.

So even with this advancement in image generation, the human mind is still in the driver’s seat.

And I believe we are several years away from an AI that creates a truly meaningful and inspiring image all by itself!

3. AI image generation is expensive

Oh! Did I not mention yet that AI image generation is expensive?

With Dall-E 2 you get 35 images free initially, which is then reduced to 15 free per month. If you need to generate more, you need to spend nearly $15 for around 100 images.

With Midjourney, you only get an initial 25 images for free (which I exhausted very fast), then you have to dish out $10 per month for approximately 200 images (calculated as GPU minutes).

Not bad you think?

The trial and error with the prompts alone are enough to eat your dollars like a hungry Tasmanian devil!

So you really need to be judicious. I spent nearly $10 to finally get to the images I have posted here!

With this in view, I shall leave you with some of my trial images as worthy mentions (as I feel guilty about spending 10 bucks and not being able to use all the generated images😉):

THANK YOU for reading till the end! I hope you loved the article as much as I loved creating it for you.

So, do you like my final picks for the images?

Or do you think I should have chosen some other version from the variations?

Which images from here are your favorite?

Are you inspired now to try your hand at image-generating AI?

And, would you like more details about how such an AI works?

I would love to hear your views!

Till then, happy reading!

Sources and Notes:

• Inspiration for this post came from the creative-commons Harry Potter art here.
• The generated images are strictly for educational purposes only.

To inspire you with my mindblowing creative journey with a text-to-image AI

Paint a word picture so vivid that I needn’t even draw.

I am sure you have seen a crime series where the witness describes how the perpetrator looks to an artist.

The artist then pieces a sketch together only from the bits and pieces of descriptive information about an unknown face.

And, I find it hard to believe how accurately the artist ends up sketching, only based on these descriptions.

But imagine if that artist were AI!

This is similar to what I experimented with this week with my access to Midjourney — giving text to the AI and getting an image out of it.

For those who haven’t yet heard of it, Midjourney is the name of a research lab and its AI program that creates images from descriptive ext. The Economist used it to generate their cover image in June 2022.

It works as a bot in your Discord server. You have to sign in and get the Discord invite. Then you can add the bot to your server as well. They have a very concise tutorial on how to get started here.

In this post, I have tried something unique — using Midjourney to recreate Harry Potter characters as Japanese Anime.

This post is meant to inspire you, induct you into the fascinating world of creative AI, and share the lessons I learned.

Hence, I have posted all the prompts I have used along with the generated image. You are free to try the same prompts or use variations of them.

For each prompt, the AI generates 4 images — you can either create a variation of the one you choose or upscale the one you like. It is pretty simple.

So sit back and enjoy the images!

First off, just a quick background on the art style I used for the images. I have used something called the Japanese Moe Art Style for generating all the images.

This is done by using moe art in all the prompts. I am sure, even if you have not heard about this art style, you definitely recognize it from some popular animes!

Note — All the prompts in Midjourney begin with the command imagine, like this:

/imagine prompt

Hence, in all examples below I have only mentioned the prompts to avoid repeating the command.

Let’s start with our Hero!

1. Harry Potter

Is one of the most recognizable fictional characters in the world. So the AI worked well even without too detailed prompts. I chose the one I thought looked most fitting (same for all images that follow).

Prompt:

harry potter face + moe art style + anime  --ar 16:9 --uplight

2. Hermione Granger

Just as for Harry, I did not need any more detailed description for Hermione.

Prompt:

hermione granger + Moe art style + anime  --ar 16:9 --uplight

3. Ronald Weasley

In this case, I had to generate more variations with the same prompt to get Ron correctly. Finally, I settled on the image below.

Prompt:

ronald weasley + Moe art style + anime  --ar 16:9 --uplight

4. Severus Snape

I used several more detailed prompts for Snape, but the simple one seemed to generate the best results.

Prompt:

snape + moe art style + anime --ar 16:9 --uplight

5. Albus Dumbledore

Dumbledore is my favorite character from the series. So I wanted to create a more detailed anime version of him. But for some reason, Dumbledore was difficult to generate with the simple prompts as used for Harry. Each time the AI messed up with the beard (I can’t imagine why). So I tried with a more detailed one here, and got a somewhat satisfactory result.

Prompt:

old wizard, dumbledore, benign, smiling, long white beard and hair + moe art style + anime --ar 16:9 --uplight

6. Lord Voldemort

In the case of He-Who-Must-Not-Be-Named, I used several additional details to get a scarier image. There were many generated before I finally chose this one below.

Prompt:

voldemort + moe art style + anime + scary scarlet eyes  + slit pupils + forked tongue + symmetrical bald face --ar 16:9 --uplight

1. Text-to-image tools are impressive

The image generation AI has exploded in recent months. And I can’t help but get impressed!

• Whether it is Dall-E 2 or Midjourney, I found them really quick in generating awesome images from the prompts given to them.
• The AI is pretty good at cross-pollinating and superimposing art styles from one to another. For example (try it out yourself)— a realistic Eiffel Tower portrait in Van Gogh style.
• The AI works well to create an image that can later be fine-tuned and modified later.
• You can use the AI to generate surreal scenes and landscapes — which you can use as inspirations for creating scenes for video games and even movies!

Despite all the advantages, I must confess this: though I called them creative AI at the beginning of this post — I believe calling it such is misleading, at least at the current level.

This brings me to my next takeaway.

2. AI image generation is difficult

Even though it seems very simple in this article, generating all those images was time-consuming, to say the least!

You can never arrive at a perfect prompt without trial and error. And even with the same prompt, the AI will not generate the same images each time.

Look closely at all the above images and you will see that they are not exactly correct. Something seems off — like the eyes do not match, and overall they look a bit weird.

The true art in this case is getting the prompts correctly!

And this is where humans still rule — creativity.

So even with this advancement in image generation, the human mind is still in the driver’s seat.

And I believe we are several years away from an AI that creates a truly meaningful and inspiring image all by itself!

3. AI image generation is expensive

Oh! Did I not mention yet that AI image generation is expensive?

With Dall-E 2 you get 35 images free initially, which is then reduced to 15 free per month. If you need to generate more, you need to spend nearly $15 for around 100 images.

With Midjourney, you only get an initial 25 images for free (which I exhausted very fast), then you have to dish out $10 per month for approximately 200 images (calculated as GPU minutes).

Not bad you think?

The trial and error with the prompts alone are enough to eat your dollars like a hungry Tasmanian devil!

So you really need to be judicious. I spent nearly $10 to finally get to the images I have posted here!

With this in view, I shall leave you with some of my trial images as worthy mentions (as I feel guilty about spending 10 bucks and not being able to use all the generated images😉):

THANK YOU for reading till the end! I hope you loved the article as much as I loved creating it for you.

So, do you like my final picks for the images?

Or do you think I should have chosen some other version from the variations?

Which images from here are your favorite?

Are you inspired now to try your hand at image-generating AI?

And, would you like more details about how such an AI works?

I would love to hear your views!

Till then, happy reading!

Sources and Notes:

• Inspiration for this post came from the creative-commons Harry Potter art here.
• The generated images are strictly for educational purposes only.