[PR] Thinking with ChatGPT, generating with Pollo AI, and lining them up

*This article is a sponsored post.

🌅 Prologue | Why do we stop in our tracks?

When I decide to create something with generative AI, I sometimes find myself strangely unable to move forward.

How should I write the prompt? Which model is the right choice? What should I adjust to improve the quality? The more I think about it, the more stuck I become.

I used to be like this. I would write a prompt, delete it, and rewrite it again. I would generate something, but it wouldn't turn out as I imagined. So, what should I fix? I would start thinking from scratch again. Before I knew it, hours had passed, and nothing had taken shape.

Perhaps I was convinced that unless I wrote the perfect instructions, I wouldn't get good results.

But do I really need to decide everything before I can start?

What I tried this time was an approach of moving forward without deciding everything. First, I talk with ChatGPT to organize my thoughts. Then, I try to give it shape with Pollo AI. I use the same instructions to generate with several models. And then, I line them up and look at them.

Instead of searching for the right answer, I look at the differences. In that process, my sense of distance from generative AI gradually changed.

*I have summarized information about Pollo AI. Please feel free to use it as a reference!

💭 Interacting with ChatGPT

Before giving instructions to the generative AI, I decided to try talking with ChatGPT first.

Previously, I would try to write a prompt right away. But when I'm just thinking in my head, it inevitably remains vague. What does "a gentle feel" actually mean? What elements make something look "futuristic"?

When I talk to ChatGPT, I can gradually put those ambiguities into words.

"I want to create a video with this kind of atmosphere." "Something gentle, but futuristic." "A quiet world that doesn't explain too much."

ChatGPT receives that carefully and gradually gives it shape. It might ask, "Is the gentle feel about the softness of the light, or the color palette?" or confirm, "Would you prefer there to be a human presence in the quiet world?"

This exchange felt more like a conversation than a task. I don't even know clearly what I want to make yet. But as I talk, the outline starts to emerge. That's the kind of time it is.

Perhaps the important thing here is not to decide on a single correct answer. If I solidify it by saying "it must be this way," I feel like what I can see later becomes narrower.

In the past, I would have spent hours at this stage trying to write the "perfect prompt." But this time, I decided to just determine the direction through the conversation with ChatGPT and move on to the next step while the details were still vague.

📝 Trying to make it a meta-prompt

As I continue to interact with ChatGPT, the outline of the 'kind of world I want to create' gradually becomes clearer. So, I decided to summarize that outline into a meta-prompt.

A meta-prompt is, roughly speaking, 'words that define a worldview.' Instead of giving detailed instructions, it conveys an atmosphere and direction. It is a collection of words that serves that role.

The prompt I ended up with this time looked something like this.

A calm, cinematic short video set in a softly futuristic city at dusk.
On a quiet rooftop terrace, a stylized Shiba Inu mascot and a small acorn mascot sit side by side.
They are not performing or posing — just being there naturally, sharing the same space and time.
The atmosphere is gentle and warm, with subtle sunset light reflecting on surrounding buildings.

In summary: A gentle and cinematic short video. A soft future city at dusk. Small characters are sitting side-by-side on a quiet rooftop terrace. They aren't posing; they are just naturally there, sharing the same space and time. A gentle and warm atmosphere. The light of the setting sun is reflecting off the buildings.

It's not too rigid, but it's not careless either. Should I call it just the right amount of ambiguity? With this much room for interpretation, what is born from it changes little by little.

🎬 Passing it to Pollo AI

Once the meta-prompt is ready, the next step is to pass it to a video generation tool. The service I used this time was Pollo AI.

✨ It stays close to you while you think, and you can properly leave the creation part to it.

Pollo AI was chosen because it allows you to try multiple AI models in one place. By using the same prompt to generate with several models, you start to see something like the 'expression' of each model.

Here, the atmosphere changes a little.

When I was talking with ChatGPT, I felt like I had the initiative. But the moment I passed the instructions to Pollo AI, the initiative shifted slightly to the other side. I couldn't say anything more. I just had to wait.

This feeling of 'leaving it to someone else' was surprisingly fresh.

🔄 Trying generation with multiple models

What I tried this time was passing the same meta-prompt to three models.

I used Pollo 2.0, Sora 2, and Kling 2.6.

Since the instructions were the same, I somehow thought they would return similar results. But in reality, they were quite different.

When the generation was finished and I lined up the three videos, the first thing I thought was, 'Oh, they're more different than I expected.'

None of them are wrong. But they are all slightly different. That's how it feels.

👀 What I saw by lining up the three outputs

As I lined up and looked at the three videos, I began to see something like a "personality" in each of them.

One has a well-balanced overall composition and a sense of stability. Both the framing and the lighting turned out close to what I expected.

Another had a slightly different atmosphere. It was more realistic, with the textures of the details rendered carefully. The movement also felt natural, with a sense of weight.

And the third one showed a unique directorial style. The colors, textures, and camera movements were clearly different from the other two. I can feel a certain "flavor" to it as a video.

It's not about which one is good or bad; they are just different. Just by looking at those differences, I felt like I was beginning to understand something little by little.

In the past, I would have tried to decide "which one is the correct answer" at this point. And I might have thought of the ones I didn't choose as "failures." But this time, I started to feel like I don't have to choose.

I'll keep all three and look back at them from time to time. I'll think, "This one is good for today," depending on my mood. I think that's fine.

Even with the same words, the scenery that comes to mind differs depending on the person receiving them. It might be similar to that. You can think of the differences as errors, or you can think of them as richness.

This time, I decided to accept them as richness.

Effort before use:

• Rewriting prompts over and over (several hours)

• Not satisfied with one result, starting over from the beginning

• Not knowing what to fix, wandering aimlessly

Flow after use:

• Organizing the direction with ChatGPT (about 15 minutes)

• Pollo AI sending to multiple models simultaneously

• Lining them up and enjoying the differences

• The option of not having to choose

It's not just that the time has been shortened. Being freed from the pressure of "searching for the correct answer" might be the biggest change.

🧭 What I saw by comparing

As I compared the three videos, I realized that the version of myself that was searching for the 'correct answer' had disappeared.

I had intended to choose just one, but as I lined them up and looked at them, I started to feel like there was no need to choose at all.

Each has its own merits. Each has its own differences. But none of them are wrong. When I was able to think that way, I felt a little more at ease.

🌱 Rather than the result, I felt that the time spent experimenting was truly adding up.

When using generative AI, we tend to make the 'deliverable' our goal. But perhaps what is important is the process itself: thinking, creating, and observing.

I believe the deliverable is not an answer, but a part of a relationship.

🤝 Rather than creating, it's about keeping company

If we think of our relationship with generative AI as 'using a tool,' we inevitably head in the direction of 'mastering it.' Giving the right instructions to get the desired result. That is important, but if that is all there is, it feels a bit stifling.

Maybe it's okay to be a little more relaxed in how we interact with it.

Generating multiple results from the same prompt and looking at them. Sometimes, I discover preferences I hadn't even realized I had. Or, I might get the feeling that while they are all different, none of them are wrong.

This feels closer to keeping company than to creating.

Watching how the other side responds and making small adjustments on my end as well. Not trying to get it right on the first try, but finding the right balance through several exchanges. That is the kind of relationship it is.

When we create anything, not just with AI, we tend to decide on the 'finished form' first. But in reality, don't we often think while we create, or change direction after seeing what we've made?

Generating multiple results from the same prompt is, I think, one way to create the space needed for that kind of 'thinking while creating.'

Without deciding on a correct answer, you keep the options lined up. You can choose from them, or you can look at them all and think of a new prompt. There is no need to rush.

There is surely a way of interacting that is unique to you.

🌙 Epilogue | Ending with 'Oh well, that's fine'

In the end, which video did I choose?

To tell you the truth, I haven't decided yet. I've kept them all. I look back at them occasionally, and depending on my mood that day, I think, 'This one is good for today.'

I don't think there is a single right way to interact with generative AI. Sometimes you get tired trying to master it, and other times you try things randomly and make unexpected discoveries.

Generating multiple results from the same prompt and just looking at them is one way to interact with it. Whether it's efficient is debatable, and whether it's productive makes me tilt my head. But this 'time spent looking' is actually not bad at all.

🪐 For those who want to think things through, having a place to take a detour can make things a little easier.

You don't have to aim for perfection. You don't have to settle on just one. Enjoy the differences, and just move forward with a 'well, this is fine' attitude.

If you can interact with it that way, generative AI might feel a little more approachable.

You don't have to start by creating something right away. Talk to it first, then think. In the meantime, something will change little by little. I think that's just fine.

Afterword

Thank you for reading this far.

I wrote this article about the flow and insights of creating prompts with ChatGPT and using them in Pollo AI.

Pollo AI I hope this conveys at least a little of the appeal of

Following or liking this post would be a great encouragement!
I look forward to your continued support!

Hashtags

#PolloAI #ChatGPT #Sora #Kling #GenerativeAI #AIVideoGeneration #Prompt #HowToInteractWithAI #MultipleModels #AIArt #TrialAndError #CreativeProcess #AIBeginner #Creative #VideoProduction #MetaPrompt #AICreator #OrganizingThoughts #Dialogue #AITools #VideoProduction #Detour #Margin #PR