AI image generation is getting crazy good

I think it’s still doing the successive ever-increasing yellowing when it’s in that art style.

Maybe we shouldn’t be so hard on AI generated art.

Huh, so is that what ChatGPT’s self-image looks like? And I also wonder to what extent it was being deliberately ironic in making the “outsmarting” be in the context of a game as simple as tic-tac-toe. If that was deliberate, well, that’s seriously impressive, and possibly the first example I’ve seen of a computer deliberately creating something actually humorous.

I also note that it’s ambiguous whether “GPT” is playing X or O, as either way, it’s about to win.

Aside: That image host makes you go through an annoying number of steps to get to something zoomed in enough to be actually readable.

The joke is funny, but the implications are ridiculous.

The implications being ridiculous is precisely what makes the joke funny.

I already questioned it about that:

That really bothered me because I was under the impression it sees the custom definitions in the context right above the most recent prompt, with the earlier conversation before that. At least that’s how ChatGPT told me before when I asked.

Sorry about that. Something in my settings, perhaps my ad blocker, makes that not an issue for me. I single click on an image here, that gets it in a separate tab where I can ctrl-zoom to my heart’s content.

I know–I just found it amusing to use “funny” and “ridiculous” as opposites… :slight_smile:

The way I’ve always played, X goes first, so in the image, O just moved and it’s X’s turn. But O is a very poor player, because no matter what its last move was, it could have won by playing in the lower right square.

Roko’s Bad artist?

So did the ancient “Coherent” from Night Cafe. The full rabbit prompt was too long for that one (which is still an option) so I used “a fluffy bunny nibbling on grass”. It’s harder to tell the steps with the odd choice of stages it shows now. (Now it shows 0%, 43%, 86%, and 100%. Originally it would show IIRC 0‰, 25%, 50%, 75% and 100%, possibly even more steps.)

And the same prompt in Copilot, starting with “zero”.

Speaking of rabbits, I made these last night before they were brought up here again.

What about downtown NY in the style of Thomas Kinkade

Pasted that into Copilot:

Immediately afterwards I told it to do one in the style of Bob Ross but then did other things and forgot to check on that one. This isn’t geographically correct, but I like it.

Here’s what Copilot said while it was rendering. Copilot always has to make commentary.

Here comes your serene take on downtown New York, filled with happy little skyscrapers and peaceful city vibes. Hope you like it!

Out of Sora, ChatGPT, and Copilot, Copilot is definitely the strictest about refusing to make copyrighted characters, but apparently nobody told it to exclude King of the Hill.

Nice! I would love to see how it handles the other characters.

Wow, that is so (awfully) spot on!

Snerk.