I let a 600 yen/month AI handle my thumbnails and 30 minutes of my day disappeared

Today's thumbnail is
I'm Taka, wide-eyed at
the Gemini that landed
in the palm of my hand✨

Until a little while ago, I
would spend over 30 minutes
on a single note thumbnail.

Opening Canva,
searching for assets,
placing text,
and adjusting colors.

If you keep that up every day,
it calculates to nearly 15 hours a month
just spent on
making thumbnails.

Now I leave it to generative AI and
my only task is to check the image
.

Back when 30 minutes a day disappeared in Canva

My previous thumbnail creation
process went like this.

1. Generate character images with the web version of Gemini

2. Run them through a background removal tool

3. Paste them into Canva

4. Think of and arrange the background and text

5. Adjust the size and colors

I would always get stuck somewhere,
and before I knew it, 30 minutes had passed.

What's more, the vibe was different
for every article,
so looking back, there was no consistency.

I was spending more time
drained by design
than by writing.

The arrival of a 600 yen/month AI made me decide

The turning point was when the options for
image generation models increased by one.

• Nanobanana Pro: High precision, but about 100 yen per image
  

  
  

• Nanobanana2 (newcomer): A little under 20 yen per image
  

Since I post every day,
Pro would cost 3000 yen a month.
As someone cautious about fixed costs,
it was an amount I couldn't commit to.

However, Nanobanana2 is
only 600 yen per month, even for daily posts.

"I can use this every day."

So, I integrated it into
Claude Code's image generation skills
so I could call it up
from my daily articles.

The work has become just "checking the image"

Now, as soon as I finish writing an article,
I just ask Claude Code
to generate a thumbnail.

The image appears in a few dozen seconds,
and I check the result.
Canva work is zero.
Background removal is also unnecessary.

Since I am generating them
with the same prompt structure,
a consistent look and feel
was a side benefit.

Getting back 30 minutes every day
and being able to redirect that time
to writing, improvingSysteme.ioproduct landing pages,
creating Udemy courses,
and testing new tools
has been the biggest change.

4 tips I learned from delegating images to AI

What I learned from trying this
were some simple tips
for giving instructions to AI.

1. Have the AI create the prompt first

Convey the image you want to create
to the AI in Japanese,
and have it create only the prompt first.

When you generate an image with that prompt,
you get an image closer to what you imagined
than if you instructed it directly yourself.

This was my biggest discovery.

2. Have the prompt written in English

I'm not sure if it's related,
but I feel that instructions structured in English
tend to yield higher accuracy.

You can just have the AI
translate your Japanese ideas into English
and format them at the same time.

3. Generate in 16:9 and keep the subject in the center

Since I cannot output directly
in the recommended ratio for note thumbnails,
I generate them in 16:9.

On top of that,
I include an instruction in the prompt
"to keep the main elements
within the central 80% area."

Even when cropped on note,
the face and small items
will not be cut off.

4. Always save your prompts

I save the generated prompts
along with the article draft
in a text file.

It allows for reproducibility,
and it's convenient to look back
at how I changed things from last time.

The actual prompt structure
looks like this.

The image above
came from these instructions.
Writing in a 5-part structure of
character, scene, background, lighting, and constraints
is the most stable method so far.

Start by having the AI create just the prompt

If you are also
wasting time
on daily image creation,
please try at least step 1
for now!

Tell ChatGPT or Claude
in Japanese
about the image you want to create,
and have them write
only the prompt.

Copy and paste that
into Gemini.
With just this extra step,
the accuracy of the output
will improve significantly.

For me personally,
just by stopping the method of
trying hard to write prompts myself,
image creation became a completely different experience.

When you get back 30 minutes every day,
you can start working on the articles you want to write
and the things you want to try.

Please feel free to start
with step 1 of prompt creation!

🎁 My Udemy course is now live!

This is a practical course packed with the entire process
of building an automated sales system from scratch.
If you're interested, please take a look.

If you have taken the course,
I would appreciate a high rating and a kind review🙏

👇 Click here for course details and how to enroll

👨‍💻 Click here if you're wondering who Taka is

👇 Recommended Articles