Hi Michael,

Indeed, it is not able to generate better ones.

My theory is that after the publication of a paper that they were able to generate an image with a single step (older algorithms required around 25 or 60 steps), they replaced the DALL·E 3 in the ChatGPT by the new algorithm. It shows all the telltales for that.

But, the sad story is that the original DALL·E 3 is still the best image generator that I have found (yes, even better than flux, probably not better than SORA). My sensation is that DALL·E 3 has far more parameters, and it has a "better understanding" of your intentions because it has deeper "thoughts".

The API version is still the original DALL·E 3, but it is not good enough, specially because it rewrites your prompts (very focused on making them more inclusive, leaving out some details that you wanted).

So, for now, the best that I found is the MS copilot. Which, sadly, only makes squared images.

And now, I don't even like to use GPT-4o because it annoys me.

For the rest, I believe that they do great products —I'm not sure what will happen now that so many people have quit from OpenAI—, but great products too expensive to sell.

About if it will survive... I am not sure. It seems that they are pivoting toward governments. It looks like that they will get our taxes as form of payment. 😔

I hope to be wrong.