Dalle-3

Prompt: monkey with Gavan cigar, sunglasses, holding small F1 grenade, colorful fog, editorial photography

#aigeneration #dalle3

Dalle-3

Prompt: New York street, waterfall from the building, river between buildings, editorial photography

#aigeneration #dalle3

Dalle-3

Prompt female Spartacus, ancient, contrasting shadows, atmospheric

#dalle3 #aigeneration

#dalle3 - opinion and what's next for visual models

A long time ago (a year ago, lol) Dalle-2 set the bar for quality in picture generation, but lost the media race to Midjourney (in terms of quality) and Stable Diffusion (in terms of flexibility).

And now, OpenAI is back in the visual modeling game and setting the bar again.

First of all, the level of query understanding and what AI bloggers call coherency (cohesion or coherence), roughly speaking logical and visual integrity.

From field tests - what I spent hours to achieve in MJ and Stable, in Dalle-3 I was able to achieve in one attempt.

What you write is what you get.

What's next?

Midjourney is about to release version six, which, according to information I heard, will have a lot more style flexibility (not just very, very pretty) and that coherncy. And also, some semblance of ControlNet. And if I were them, I would certainly go in the direction of more control, because otherwise they will release Dalle-3 clone.

Stable Diffusion continues to be the people's favorite, because it is expanding to infinite possibilities at the expense of enthusiasts. But now with a new generation, the XL. Last year's 1.5 model smoothly gives up the reins of popularity.

So if you haven't tried it yet, I suggest https://www.bing.com/create

#dalle3