e-Learning Blog

This time, I will introduce how we created this year’s ISMS training. I will slowly explain what kind of AI we used and what difficulties we encountered.

"Oh, Ryuno-kun made it? I’m looking forward to it."

In this year's ISMS training, we fully utilized generative AI to create two pieces of content. Last year, we made animations using Vyond, but there were limitations in expressive power, and with video generation AI becoming practical from now on, I thought it was important to gain experience.

"Oh, it's all done with AI. Indeed, the quality of generative AI is steadily improving, so we have to keep challenging ourselves."

This time, we created three parts. The first is a Yukkuri-style video, the second is an animation somewhat inspired by a certain famous anime, and the third is content on countermeasures against supply chain attacks.

"Each one has a completely different style. Specifically, what kind of AI did you use?"

The first one I used was ChatGPT. I used it to organize the scenario and script.

"Ah, that famous one, right?"

After organizing the key points here, I had it generate a draft script, which I then manually reviewed. Next is GenSpark. This is an AI agent that can combine and use various AIs.

"So you can integrate and use multiple AIs. That sounds convenient."

The slides for the supply chain attack countermeasures, both the text and the slides themselves, were created using GenSpark.

"You generated the video based on that scenario, right?"

Actually, we first create a single illustration as an image. This is because it’s not possible to make a consistent video just from scripts or text. The image generation tool we used is NanoBanana. This is Google’s image generation AI. We used it to create the character drafts and the very first frame of the video, the so-called 1st frame.

"Google, huh? Impressive. By the way, what exactly is the 1st frame?"

By creating the 1st frame as an image, the character's appearance can be unified, and the final image becomes clearer.

"I see, so you make a video from still images. Did you also create the voice with AI?"

We used two types of voices. For the Yukkuri parts, we used Yukkuri Voice Maker. This is a standard software actually used in Yukkuri videos. However, the other voices were created with Gemini 2.5 TTS. We also used Gemini 2.5 TTS for other anime parts and for countermeasures against supply chain attacks. The voice is quite natural, but the drawback is that it always includes misreadings.

"What kind of misreading?"

For example, multi-factor authentication was read as "tayousei" authentication.

"What kind of AI did you use for the main video generation?"

We used three AIs for video generation. At that time, the most stable one was Kling. It's made in China, but we use it via GenSpark without inputting any confidential information.

"Security considerations are necessary no matter which service you use."

It was good that they followed instructions straightforwardly and didn’t do unnecessary things. Veo is Google's video generation AI, and Sora is OpenAI's video generation AI. Both have advanced technology, but they often behaved unpredictably.

"What do you mean by unexpected behavior?"

Like a stranger suddenly appearing in the background, or in a scene where someone picks up the receiver, for some reason the receiver is placed standing upright.

"Wow, that's way too mysterious (laughs)"

We also used video editing tools. We fine-tuned images with PhotoShop, cut and pasted videos with Filmora while combining audio, and finalized the content using Articulate Rise and Storyline.

"In the end, human hands are still necessary, huh? How did you create each part?"

First is the Yukkuri part. Normally, I would use a dedicated software called Yukkuri Movie Maker, but since I'm not familiar with the software, I created the movements with video generation AI and combined the audio using video editing software.

"Haha, that's honest."

Next is the anime part.

"I've been curious about this!"

Using NanoBanana, I created a character that combines the president and an anime character.

"What, a mix of me and that character?"

Finally, the e-learning for supply chain attack countermeasures. This doesn’t have anime-like movements but rather simple motions closer to typical e-learning. Items that needed to loop were converted into GIF animations using PhotoShop, and everything else was used as video. The slides were created with GenSpark, and the videos were placed there.

"You’re using various technologies selectively. Were there any difficulties during production?"

Both image and video generation struggled to maintain quality, and before we knew it, the outfits had changed. Also, with AI-generated synthetic voices, there were parts that couldn’t be read correctly, so this time there are sections where mispronunciations were left as is.

"It's important not to aim for perfection and to set compromises according to the project."