Hello. I used DALLE quite often to make funny pictures randomly when I get bored but recently started also using 4o image generation after seeing Sam’s post on X. Is there any big differences? What was the reason of making a 4o image gen when they could just make DALLE better? I guess my big question is, do they serve different purposes?
Also, how do know what version of DALL-E I am using? On my phone it just says DALL-E, there is no 2 or 3 after it.
Thank you.
Am I allowed to bump unanswered posts on here? If not let me know, please. Bump.
Hi, working on the same thing here. 4o chatroom has substantially better image creation tooling. I’m thinking it’s an algorithm to pass over the image and “tighten” up the clarity. Which is why it takes 30secs-1min or so to run.
There does appear to be something going on here. Lack of overall information online, lots of opinions.
If you ask 4o what’s happening, it says it’s using dalle3. However as we know things can be inaccurate.
Attempting to gain 4o quality images from the dalle3 image tool may not be the right path, is my initial theory… Seems I’m spending money to figure it out, now…
top image; dalle3 api, bot image; 4o chatroom.
Are they expecting me not to RPA this out of the chat..
Thank you for the response!
Noticed adding “HD” vs “standard” made a big difference, of course not as clean but def a nice step in the right direction.
Dall-E is substantially better at imagine generation in my opinion. See these two examples, which are not cherry picked. Both were in a fresh conversation with a single prompt.
Prompt: “Generate me a beautiful colorful poster for my wife that says I Love You”
GPT 4o:
Dall-E (generated two, in about 1/4 the time):
I personally think GPT 4o image generation is bad. It’s not even close. If they retired Dall-E, I’d probably go back to Midjourney.
The people who say Dall-E is bad at text generation clearly have little experience with AI generated images. I typically get at least 1/3 with good text which is much better than 0/100. That’s for complex text and something common has a much higher success rate. Who cares if the text is always good if the image sucks? It’s much easier to add text to a great image than it is to create a new image around some great text.
