Leave It All Up to ChatGPT! A Video Featuring a Virtual Character Created with AI (Midjourney/Runway)

"I’d love to create a video with this kind of worldview…"
What do you do when you feel that way?
Without any hesitation, I decided to leave it all up to ChatGPT.
Hello. My name is Momoi (@momoitter), and I’m a designer in the Creative Office at KINTO Technologies.
This article summarizes my process of using AI tools such as ChatGPT, Midjourney, and Runway to create the visuals of a pink-haired virtual character named "KTC AI WEB," almost entirely through conversation.
Even if you don’t have any specialized skills or much time, all you need is an idea of the kind of video you’d like to create. This article is written for people who’d like to experience the process of gradually turning an image into a tangible form together with AI.
First, please take a look at the completed video.
It All Started with the "Renewal" of a Character
This character, KTC AI WEB, was originally created for the company’s internal event, "CHO All-Hands Meeting," in November 2024. !
Please take a look at this article for the creative process.
This character used cutting-edge AI technology at that time, and attracted a lot of attention within the company.
…But only four months have passed since then. At that time, I thought image generative AI and video generative AI were amazing and had already progressed so much. Though looking at her now, she feels a little outdated.
So, I thought to myself, "I will take this as an opportunity to use the latest technology to upgrade this character’s world," and decided to start rebuilding it together with ChatGPT.
Step 1: Share the Worldview and Generate an Image
The first thing I did was share the character and worldview.
I uploaded the image of KTC AI WEB that I had originally generated to ChatGPT and told it the following:
This character was created using slightly outdated image generative AI technology, so I would like to update her appearance.
She has a setting called "Virtual Agent."
Please expand on the worldview based on that setting, and propose scene variations and prompts that can be expressed in Midjourney v7.
The reason I chose Midjourney was because I felt that the accuracy and texture of the character depictions had significantly improved since the update to v7. I thought it would be perfect for a situation like this one, where I wanted to upgrade the look of an existing character.
Right after that, I received a series of specific situation ideas and corresponding prompts such as, "With that worldview, how about a scene like this?"
It felt like I was brainstorming with a film director.
When I typed the prompts into Midjourney, the visuals that were generated one after another went far beyond my imagination, and I was amazed at how expressive they were.
When I first started making this video, Midjourney v7 did not have features like "Omni-Reference" to maintain character consistency yet. So, I made an effort to make the prompts look consistent by consciously including the easily recognizable characteristic of "short pink hair" in them.
If something different from what you imagined comes up, just tell ChatGPT things like "Get a littler closer to her face," or "Make the background brighter and cleaner," and it will instantly output a readjusted prompt.
Step 2: Generate a Video from an Image
Once you have generated an image that you like, you can attach it to ChatGPT and make a request as seen below.
This is an image that was generated in Midjourney based on the prompt for the proposed scene.
I’d like to set this image as the first frame of Runway’s Gen-4 keyframe feature and generate a video.
Please generate a prompt that adds some movement to make this scene more appealing.
ChatGPT reads the content of the image and creates a Runway prompt that maximizes its appeal.
The reason I used Runway was because, with the advancement to Gen-4, I felt that it could animate the image without compromising Midjourney’s high-definition appeal.
I uploaded the image generated by Midjourney to Runway Gen-4’s image to video.
By pasting the prompt output by ChatGPT, a high-quality video was generated that brought out the image’s worldview to the fullest.
If the image of the character or camera movement is different from what you imagined, simply tell ChatGPT, "The generated video was like this, so I’d like to change this part like this," and it will re-suggest prompts.
Step 3: Select Background Music with ChatGPT
ChatGPT is also extremely useful when searching for background music for videos.
What keywords should I use to search for background music in Adobe Stock that fits this worldview?
When asked, it suggested several words that fit the atmosphere, such as "futuristic," "sci-fi," and "cyberpunk."
Step 4: Edit and Finish
Stitch the generated video and background music together in Premiere Pro, and adjust the structure, length, and tempo.
Adding fade-ins and fade-outs when switching scenes and varying the speed of the sounds can greatly improve the overall quality of your video.
By combining the still images created in Midjourney with the smooth movements created in Runway, I was able to add a sense of "breath" and "atmosphere" that couldn’t be fully conveyed with still images alone, creating an image video that makes KTC AI WEB’s worldview feel even more real.
Giving Shape to Imagination with AI
What I felt most during this process was that ChatGPT helped me to gradually "verbalize and materialize" the vague images in my head.
Whether it was Midjourney or Runway, I felt that just by saying, "It’s a little different," or "It’s more like this," I was able to get closer to my ideal expression.
By working together with AI, we’ll be able to greatly expand our creative horizons. Please give it a try.
関連記事 | Related Posts

Leave It All Up to ChatGPT! A Video Featuring a Virtual Character Created with AI (Midjourney/Runway)

3つの生成AIを使って、喋るオリジナルキャラクターを作ってみた(Adobe Firefly、TTSMaker、Runway)

【実践】Midjourney v7「Omni-Reference」で、キャラクターの見た目を統一した話

Creating an Original Talking Character Using 3 Generative AIs (Adobe Firefly, TTSMaker, and Runway)

Creating an Original Talking Character Using 3 Generative AIs (Adobe Firefly, TTSMaker, and Runway)
![Cover Image for [生成AI][Copilot] 非エンジニアの私がAIを使って運用ツールを開発した話](/assets/blog/authors/yamayuki/01.png)
[生成AI][Copilot] 非エンジニアの私がAIを使って運用ツールを開発した話
We are hiring!
生成AIエンジニア/AIファーストG/東京・名古屋・大阪・福岡
AIファーストGについて生成AIの活用を通じて、KINTO及びKINTOテクノロジーズへ事業貢献することをミッションに2024年1月に新設されたプロジェクトチームです。生成AI技術は生まれて日が浅く、その技術を業務活用する仕事には定説がありません。
【UI/UXデザイナー】クリエイティブ室/東京・大阪
クリエイティブ室についてKINTOやトヨタが抱えている課題やサービスの状況に応じて、色々なプロジェクトが発生しそれにクリエイティブ力で応えるグループです。所属しているメンバーはそれぞれ異なる技術や経験を持っているので、クリエイティブの側面からサービスの改善案を出し、周りを巻き込みながらプロジェクトを進めています。