DALL-E 3 Got A MAJOR Downgrade Time To Go Back To Midjourney?
HTML-код
- Опубликовано: 4 окт 2024
- Dall-E 3, OpenAI's text-to-image AI generator is out now, but is it the best AI Image generator After the recent downgrades? Here are the Biggest Rules for DALLE 3.
So many damn restrictions and guidelines…if you pay you should be able to use it how you want.
For me it’s the annoying limitation on usage, why don’t they create an Image subscription?
I can see how it might be slightly frustrating if your prompts are quite broad in nature, but personally I don’t mind the 2 image limit due to just how good DALLE 3 often is at following prompts. When there are more than two or three specific items/colours etc in Midjourney, I waste a lot of time and generations due to most of them being of no use. DALLE 3 is on a completely different level in its ability to follow prompts. The main issue I had with DALLE 3 is a problem with aspect ratios recently, in that it was fine producing square and wide images, but just would not create 9:16 or 2:3, getting confused with the meaning of portrait format/aspect ratio and hallucinating when I asked it what aspect ratio the images it had just created were. I haven’t tried it since the 2 image limit changes, so maybe that’s been resolved now. I hope so, as it was very frustrating.
I also have aspect ratio problems. More often that not, it's just 16:9 but sideways, and if it's with text, it's mirrored for some reason. Never had that problem with MidJourney.
You can also 'regenerate'. I use this. I create two images and use one by copying it then hit the regenerate button and it does two more with the same prompt.
This is an ultimate hack!! thanks for sharing lol
that counts as an additional message.
the fact they charge a monthly sub and limit you to 50 messages per 3 hrs is a disgrace tbh
Recently been dropped to 40/3 hrs.
Funny: 1792x1024 is the format generated when asking 16/9 (and it's not actually 16/9). When asking why, DALL-E recognizes the error, says that 16/9 should be 1820x1024 and proposes to make one on this ratio... Ok, do it... And it makes a 1792x1024 again, commenting it the 16/9 I asked. After some exchanges, it even ended up making 1024x1024, still pretending it to be 16/9, never saying "sorry i can't do it" or "sorry, i chose the closest format I can do"
GPT's are still, relatively speaking, very young. There are glitches to be worked out. While not technically true, we the users are both benefiting from what it can do well, and kind of 2nd phase 'user-testing' as we go.
I hit the "too fast" limit. That's usually couple minutes of timeout. You can hit "daily limit", and then it writes nonsense like you have to wait 20 hours. When I regenerated it told me even more (21 hours), but it worked after 8 hours (sleep). So it doesn't know correctly. 'Regenerate' eats the daily limit as well (obviously).
What a great video. I've been bumping up against my time restrictions also. I've been trying to create character consistency. Bouncing back-and-forth between DAlle and mid journey. I've been working on some book projects. And have been really struggling with a consistent theme imagery to extend through the book. I noticed some. Prompts are themes all of a sudden drop. Support. Like if you go into mid journey and type in a certain prompt, it'll only give you. Certain styles. And you type in a different prompt and it'll give you a different set of styles. It's like certain ideas or paragraphs are restricted to certain types of styles.
I have a character consistency video for Dalle here on the channel, not sure if you taken a look at it
I tried to force 4 images so it started to give me 4 designs into a single image for both images lol, it was for a logo so kinda ok
Lol sometimes AI does what AI wants to do
My 9:16 requests *almost* always come back as 16:9, sideways, and if there's text, 100% it's mirrored (on the 'vertical' ones). Text on 1:1 and 16:9 look fine.
The rate at which you can create because you have to work with its restrictive chatbot, means that the true rate at which you can create images is far slower than midjourney. additionally you soon run into the GPT-4 request limit.
Its now down to 1 image per message
so lame!!
I still got 4 image per batch for free user, sometimes got 3 or 2
but mostly still 4
When it first came out and you could do famous people it was awesome. Showing what it is capable of and then taking it away was foolish on their part. Kind of ruined it for me.
I don’t use Dall E 3 that much but on the part of the video where you ask for 6 images it looks like they give you 3 sets of 2 images you used 3 generation 3 messages
Now it's only 1 image at a time
I know Felicia, it’s the worst!
how about to use BING?
Same problem here, thanks
Ive been getting 1 every now and then.
lol you have to beat OpenAI up for doing you wrong like that
love the update but on the website it still gives you 4 images
on bing generator?
@@JoseNajarroAI yes
I can’t get any aspect other than 1;1 😢😢😢 🇦🇺❤️
Even if you ask for 16:9 size?
@@JoseNajarroAI yes. Or tall, or portrait, anything I tried just comes back 1:1. I’m pretty sure I don’t have the last few updates either. I’m in Australia and don’t get updates the same time as the us. 🇦🇺❤️
Check audio levels! Use basic mic till you find the solution.
Strong, those random pockets. Is it that my voice level got too low for the mic, or too loud?
@JoseNajarroAI There are two distinct levels interchanging. Like if you have directional mike with narrow field, and you get in and out of it.
Maybe you have noise cancelation "ON", and it makes more harm than good.
@@strong8705 thank you soo much!!