Tips for image generation to read before you fight with AI and smash your PC
I work as a company employee while also creating videos as a side job.
I have overtime too.
There are days when I get home and just don't want to do anything anymore.
Even so, deadlines don't wait.
In my production process, the biggest bottleneck was image generation.
Image generation stops halfway through.
Even though I told it to make them one by one, it makes a collage on its own.
Repeating the same instructions over and over, stuck in an apology loop.
Creating new chats over and over again.
And it still doesn't listen to instructions.
I have no choice but to generate them one by one.
I check each image as it's completed, then ask for the next one.
Time just melts away.
I've come close to smashing my PC many times.
To put it briefly, I stopped asking for them one by one.
What I stopped doing was the method of having images created one by one in a regular chat.
Instead, I use ChatGPT's new feature, "Work," to:
Organize the script by cut
Generate the necessary images in order
Give the images easy-to-understand file names
Compile the production details into a spreadsheet
Zip the generated images
I have started handing over everything up to this point as a single task from the beginning.
I don't need to come back and give the next instruction every time one image is completed.
I don't need to watch the screen the whole time to make sure it doesn't stop halfway through.
Once I hand over the work at the start, all I have to do is wait.
For me, it wasn't just a matter of saving a few hours.
The production time changed by days.
And above all, I no longer have to fight with the AI.
Even though I was using AI, I was the one working the whole time.
Until now, I would hand over the script, have it divided into cuts, and have the necessary images created in order.
If it didn't go well, I would ask for corrections.
If it stopped in the middle, I would prompt it to continue.
I thought I was communicating the necessary conditions in quite a lot of detail, too.
Portrait orientation.
No text.
One scene per image.
No collages.
Do not add characters on your own.
Create in the specified order.
Even so, when the instructions got a little long, some parts would be left out.
Comparison images I didn't ask for would appear.
Multiple scenes would be crammed into one image.
It would generate halfway and then just stop.
Every time it failed, I added more prohibitions.
But if I created a new chat, I would have to explain everything all over again from the beginning.
Check after one image is done.
Ask for the next one.
If it's wrong, instruct it to fix it.
If it stops midway, prompt it to continue.
Read the apology, then write the same instruction again.
Even though I was supposed to be using AI, I spent the whole time standing by and managing the progress.
Far from automation, my job had actually increased to taking care of the AI.
What I wanted was not a polite apology.
It was for it to follow the instructions I gave at the start.
Not to give up halfway.
And to finish the job to the end.
What is ChatGPT's "Work"?
ChatGPT Work is a new feature announced in July 2026.
While regular chat is suited for proceeding in a conversational format while asking questions or seeking advice, Work is suited for tasks that involve multiple steps or creating a finished product.
It can handle documents, spreadsheets, images, and more, allowing it to carry out work from start to finish.
In my case, perhaps because I had interacted with ChatGPT many times regarding image production, a suggestion appeared on the screen asking, "Would you like to create a work like this?"
However, I do not know under what conditions this suggestion appears, or if the same suggestion is shown to all users.
Using that suggestion as a starting point, I decided to restructure the image production process, which I had previously been doing one image at a time, into a single workflow.
What did I entrust to Work?
What I handed over to Work was not just a prompt for image generation.
It was the entire production flow, including the steps before and after creating the images.
First, divide the video script into cuts.
For each cut,
Which line of dialogue the image is for
What kind of scene to create
Whether to include characters
What kind of composition to use
Instructions for image generation
Instructions for animating in Runway
Instructions for use in Canva
File name after completion
and so on, create a list.
Follow that list to generate the necessary images in order.
Name the generated images so you know which cut they are used for.
Finally, compile the list into a spreadsheet and have the generated images zipped.
Only after this is done can I finally start video editing.
Previously, I used to give instructions for this process one by one.
Now, I provide them all at once at the beginning.
What was important was the process, not the prompt.
Of course, image generation instructions are important.
Portrait orientation.
No text.
One scene per image.
No collages.
Do not add characters without permission.
If you don't clarify these conditions, it's hard to get the image you want.
But that alone wasn't enough.
No matter how detailed a prompt I wrote, if I was creating images one by one while conversing, I would have to be there the whole time.
The real problem wasn't the speed of image generation.
Prompting it to continue every time the AI stops.
Reverting it every time the composition changes.
Writing the same prohibitions over and over.
Reading apologies and giving the same instructions again.
This interaction itself was draining my time and energy.
What was needed wasn't just writing an even longer and more perfect prompt.
Where to start.
In what order to proceed.
What must not be done.
To what state to finish it in the end.
I needed to decide the process from the beginning to the end of the work in advance.
Transforming image generation from a "conversation with AI" into a "production process."
For me, this was the answer.
First, write down all your tasks
If you are similarly exhausted by interacting with AI, the first thing to do is to write down the tasks you usually perform.
For example, in image production:
Check the script
Divide into cuts
Decide on the necessary images
Create instructions for image generation
Generate images
Recreate failed images
Name the files
Record them in a list
Organize the images
Import them into video editing software
In this way, try lining up the tasks in order.
Then, look for the parts you do the same way every time.
Group those parts together, and say,
"Please complete the work up to this point in this order."
Hand over the work until its final endpoint.
That was what I changed.
Instead of asking the AI for one image at a time, I tell it first what needs to be finished for the job to be complete.
Just by doing that, the time spent talking to the AI was greatly reduced.
Using Work doesn't mean everything will be perfect
Of course, leaving things to Work doesn't mean everything will automatically become perfect.
So far, I have used it in the production of about three videos.
I am making improvements little by little while actually using it.
What kind of items are useful to include in the list?
How should I name files so that I can use them later without getting confused?
Which prohibitions should be fixed from the beginning?
If I consider using it in Runway or Canva, what information should I keep?
I am reflecting the problems I encountered in actual production into the process one by one.
Even so, the burden is completely different from when I used to request images one by one and check them each time.
Previously, image generation alone took several days.
Now, if I hand the work over to the AI and wait, it proceeds all at once, not just with image generation, but also with creating the list, organizing file names, and zipping them.
While it is working, I can do other things.
I can also leave it to the AI while I am at the office.
There is no need to open the chat repeatedly to request the next step.
It is not just about saving time; the fact that I no longer have to keep my attention on the AI is a big deal.
Even if you use my workflow as it is, it probably won't fit.
The workflow I am using now is something I have nurtured little by little to match my video production.
The format of the script.
The number of images required.
Settings for people and characters.
How to name files.
Items to include in the list.
How to distinguish between using Runway and Canva.
These conditions vary depending on the video being created and the person using them.
Therefore, I think there are many parts that will not fit well even if others use the form I am currently using as is.
So the important thing is not to create the exact same workflow as mine.
It is to find the tasks you perform every time and create a process that suits your own production.
Where is your time getting stuck?
What instructions are you repeating over and over?
What tasks cannot proceed unless you are in front of the screen?
Finally, in what state do you want to receive the output?
If you organize those points and pass them to the Work, you will get closer to a production process tailored to you.
Things to know before using
I am using the paid ChatGPT Plus plan.
The Work I introduce in this article is also used in the Plus environment.
Work cannot be used indefinitely, and executing it consumes the usage quota for the agent features included in the plan.
After that quota is used up, the option to purchase additional credits may be displayed to eligible users.
If you do not purchase additional credits, you will have to wait until your usage limit is restored.
Also, unless you have enabled auto-recharge yourself, additional credits will not be purchased automatically just because you have exceeded your usage limit.
However, the available plans, usage limits, conditions for additional credits, and reset times may vary depending on the user's plan and service status.
I myself have not been able to find a place where I can check in advance exactly how many times I can use it with Plus or when the usage limit will be restored.
Therefore, when you actually use it, please check the terms of use and purchase information displayed on your screen.
To avoid wasting my usage limit, I use it for things like:
consultations
brainstorming ideas
checking scripts
considering structures
editing text
in regular chats.
And I leave only the actual work that I want to complete in bulk to Work, such as:
batch image generation
creating spreadsheets
organizing file names
zipping files
and so on.
My time and energy for side jobs are also limited.
After finishing my main job, I don't always have the energy to work on a side hustle every day.
There are days when I'm tired.
There are even days when it's exhausting just to bring my computer to the desk.
In such times, I don't have the luxury of spending time in an endless conversation with AI.
What I wanted wasn't a way to talk nicely and get along with AI.
It was a system that would carry out a set process to the end on my behalf when I'm tired.
What I want is a completion report, not an apology.
To achieve that, rather than continuing to ask AI more politely, it's better to first create a workflow that AI can complete to the end.
First, try writing out the work you do every time from start to finish.
Then, try grouping the repetitive tasks within that into a single process.
At least in my case, that alone significantly changed the bottleneck in image generation.
Not in hours, but in days.
And I was able to reduce the time I spent fighting with AI to the point of wanting to smash my PC.
Honestly, my articles are barely seen by anyone.
So, you who found this far might be a little lucky.
I feel that many of the "AI-powered automations" seen often these days involve combining multiple services or paying for each one.
Of course, spending money increases what you can do.
But I am doing a side hustle because I simply don't have enough money to live on.
I am not in a position where I can pay money to make things easier.
That is precisely why I have been thinking about how to lighten my workload using what I currently have, with as little expense as possible.
If this reaches someone in a similar situation, and it helps someone who is struggling just a little bit, I would be very happy.
If you found this article helpful, please give it a like.
It is very encouraging!