How good are ChatGPT 4o’s updated image capabilities?

Why is Sam Altman so excited?

Excitement

Sam Altman, the founder of OpenAI, announced the upgrade to GPT-4o’s image capabilities quite excitedly on social media — look, he even used an exclamation mark!

However, we all know the saying:

Every potter praises his own pot. (Similar to “Don’t trust the salesman praising his own wares.”)

So, blindly trusting the promotional claims of large model vendors isn’t prudent.

Is the actual performance of GPT-4o’s image processing really as exciting as Sam says? Let’s give it a try.

Replacement

First, let’s try a common operation: replacing an object in the foreground of an image.

Look, this is a photo I took on our school’s ecological campus. You can often see geese strolling around. In the photo, there happen to be a few white geese on the grass.