#0036 [For Beginners] The 5 Things I Stumbled Over When Starting AI Video Production｜B型事業所Diamonds｜クリエイター

I want to start making AI videos.

When I thought that and actually started using AI tools, I stumbled in unexpected places.

I myself experienced “things not going as planned” many times while producing the AI film “LIFE”.

This time, for those who are about to start AI video production, I will introduce the 5 things I first stumbled over.

1. I was confused by the sheer number of AI tools

The first thing I felt was,

“Are there really this many AI tools?”

That was it.

Image generation AI, video generation AI, voice generation AI, editing AI, etc.—the more I researched, the more new tools appeared.

At first,

“I have to learn them all”

I thought.

However, when I actually started production, I didn’t need to learn every tool.

I just needed to choose the tools to use according to the necessary tasks.

Once I realized this, it became much easier.

2. Prompts didn’t convey what I intended

The next thing I struggled with was prompts.

“Please create an image like this”

Even when I instructed the AI, images different from what I had imagined came out.

The character’s clothing was wrong.

The facial expressions are different.

The backgrounds are different.

In some cases, the person themselves changes entirely.

At first, I

“Why won’t it make what I told it to?”

I used to think.

However, AI does not understand vague images
like humans do.

What kind of person is it?

What kind of place is it?

What kind of composition is it?

What kind of expression is it?

You need to convey these as specifically as possible.

I have gradually come to understand that prompts are
a way of “having a conversation” with AI.

3. I couldn’t create the same character

What I found particularly difficult in AI video production was

“having the same person appear multiple times”

it was.

In movies, the same protagonist appears in various scenes.

However, when generating images with AI,

“The face is different from the one before…”

This is what happens.

The hairstyle changes.

The clothing changes.

The facial features change.

Sometimes even the age looks different.

Even in the production of “LIFE,” I struggled quite a bit with character consistency.

So, I organized the character settings and used reference images to try and make them look like the same person as much as possible.

In AI video production,

I learned that it is important not just to “create characters,” but also to “manage characters.”

That is also important.

4. It was difficult to turn images into videos

Once the image was finished,

I thought, “If I turn this into a video, it’s done!”

That’s what I thought at first.

However, when I actually tried turning the images into videos, other problems arose.

The person moves unnaturally.

The hands and fingers look strange.

The face changes.

The background moves.

It makes movements I didn’t specify.

Things that looked natural in a still image
can look unnatural when they become a video.

In other words,

Image generation and video generation have different challenges.

This was also something I only realized after I started AI video production.

5. I thought it would be “finished in one go”

And this was the biggest misunderstanding.

“If it’s AI, it should be finished once I give the instructions.”

That’s what I thought.

In reality, it was different.

Generate it many times.

Check it.

Correct it.

Generate it again.

And then, correct it again.

It’s a cycle of this.

AI is convenient, but it doesn’t necessarily
create the ideal video on the first try.

Rather,

“Getting closer to completion through trial and error”

is a more accurate feeling, I think.

◆ What I learned from these 5 stumbles

To summarize the things I first stumbled over, there are these five.

1. Too many AI tools → Learn the necessary ones first

2. Prompts are difficult → Give specific instructions

3. Characters change → Manage settings and reference images

4. Video generation is difficult → Image generation and video generation are different things

5. It doesn’t finish on the first try → Trial and error is a given

When I first started AI video production, there were times I thought, “Is this too difficult for me?”

However, looking back now, each of the things I stumbled over at the beginning has been useful for my subsequent production work.

◆ AI video production is about “learning while failing”

There is one thing I want to tell those who are starting AI video production.

You don’t have to try to make it perfect from the start.

I wasn’t able to do it well from the beginning either.

I failed many times, tried things out, corrected them, and repeated that process.

AI is evolving rapidly.

That is why there is no need to try to learn everything from the start.

First,

“Try creating one image”

“Try creating one short video”

Start from such small things.

I think that is enough.

◆ In conclusion

AI video production was a much deeper world than I had imagined.

But it was precisely because I had those initial stumbles that I gradually began to understand “how to make things work.”

Even now at 64, I am still learning.

And to those who are about to start making AI videos,

“It’s okay not to be good at it from the start”

is what I want to convey.

Making mistakes is also part of AI video production.

Start with one small video.

I, too, will continue to take on new challenges.