[Utilizing ChatGPT] Have you experienced the evolution of AI image generation yet?

Thank you for coming to read again today♪

The update to ChatGPT-4o has been a hot topic lately, hasn't it?
On X, it seems like everyone is enjoying and celebrating the update's features, with feeds overflowing with four-panel comics and Ghibli-style art.

The precision has improved to the point where you can really feel how amazing it is when you try it yourself; it was able to do things that were impossible before, like transparent backgrounds and even creating GIF images. Perhaps what made me happiest was that it outputs as PNG images. I used to have to have it converted to JPEG every time (which it could do), or convert it myself after saving, so being able to use what I created as-is is very much appreciated.

And here is something I tried out for a test.

The music is from the free distribution announcement in my previous article, but please turn a blind eye to the low quality of the video🙈lol
This slime-like character and the background were created by ChatGPT.
Then, I had it create a GIF image based on those images, and combined the two.
Actually, after this, I had it remake it into an anime-style image, but whether it was because the processing was difficult or my instructions were bad, it kept generating things that made me think, "Why did it turn out like that!?", and after about two outputs, I ended up in a situation where I was hit with a generation limit for several minutes...
Even though something like the above was completed in just a few dozen minutes, once it became anime-style, I couldn't even make it in a whole day.

I still feel like I haven't quite mastered it yet, but one thing I find a little disappointing with this update is the change in art style.
As you can see if you look at the articles on this account, I have been generating characters with as much consistency as possible using DALL-E 3.
However, the prompts I was using have become completely unusable.
If you look at the image in this heading, you can see that this is an image output using the exact same prompt I used before.

When you compare them like this, it's obvious, isn't it...
But I feel like GPT-4o understands the meaning of the prompts better. Furthermore, when I asked it to "close its eyes" just like this, DALL-E said a technical error occurred and wouldn't generate it lol

For better or worse, there are pros and cons.
It seems I have no choice but to keep going while taking the best of both worlds.
No... it would be solved if I could get my ideal art style with GPT-4o, though.
Still, as I tweeted on X the other day, it really is amazing~
It's no wonder X was overflowing with this commotion this time.

Well, rather than writing here about what happened and how, there are many people who have summarized it wonderfully, so please refer to them.
I personally recommend this article in particular~

Alpaka-san is a wonderful creator whose art style I have been using as a reference since I started generating images with DALL-E 3.
I am secretly looking forward to seeing what kind of art style they will use to create new works from now on♪

And since I think I will be referring to them a lot again, I will express my apologies and gratitude here in advance(♥︎︎ᴗ͈ˬᴗ͈)⁾⁾⁾

In any case, the development of AI is remarkable, isn't it...
Rather than being desperate to master the current situation, if we don't evolve flexibly together while anticipating that it will be improved further, it seems like we will be left behind immediately.

It's exactly this state of mind.
I was thinking about how to master prompts, but now it says it will even take into account the ambiguity of those prompts...
While I think it's amazing, it's also terrifying, isn't it.

But, well.
I'm just going to enjoy it♪
Let's all enjoy it together♪