Which is the BEST AI image generator?

GenesisOne

Elite Member
Jr. VIP
Joined
Nov 25, 2017
Messages
11,570
Reaction score
7,545
Hello guys,

I haven't used AI image-generated yet.
I am working on a new project and this kind of images should do a good job on it as it is fantasy related.

In your experience, which would be the best and easiest one to use?

Thanks for your guidance :)
 
Hmm. Depends on what you mean by best. Stable Diffusion, extremely versatile. Midjourney, best at people/faces imo. I haven't had enough time with the new DALLE 3 to feel it out, but the fact that it can handle some text is impressive.
 
Hello guys,

I haven't used AI image-generated yet.
I am working on a new project and this kind of images should do a good job on it as it is fantasy related.

In your experience, which would be the best and easiest one to use?

Thanks for your guidance :)
I genuinely think that any of the latest versions will be good enough. But my opinions on the main models are:

SDXL has almost caught up to Midjourney on people and faces. Loads of options to use it as well, including self-hosting if you're really going to be producing images at scale. It can produce text in images as well. I'd probably go with a trial on Dreamstudio first (you'll get some free credits to test it out) and just see what you think of the images. You can use specific styles (cinematic/anime/etc) without having to write a full prompt describing the style. And if you're wanting to get certain poses, or to create a scene based on a photo of a landscape, you can upload a reference image and set the noise so it can just about see the outline, then generate images with a prompt based on that. That feature is why I cancelled Midjourney and stuck with vanilla Dreamstudio.

Dall-E 3 is very impressive and completely free in Bing Chat, so it's worth using. Again, pretty much catching up to Midjourney for people and faces (although, it does seem a little more "cartoon-y", which I think may be a limitation OpenAI have put in to discourage deepfakes).

Midjourney does seem to be a lot easier to get what you're looking for, which may be down to the way they've trained it on human feedback. Their decision to produce 4 low res images at a time and then asking the user to upscale the ones they like was a stroke of genius. They have years and years of data of what outputs people liked and didn't like. So it's usually very easy to get exactly what you want by using Midjourney.

Plus, there are hundreds and hundreds of tools built on Stable Diffusion which have recently upgraded to XL. Supermachine is pretty good and has a lot of fine tuned models, there's a few (including one on Appsumo now, I can't remember the name) which allow you to generate consistent characters. So, if you want something where you just select a model and don't have to work too hard on the prompt you should be able to find it out there.

Personally, I think they're all good enough at pretty much anything now. There isn't really a "best" in my experience. I've chosen Dreamstudio, but I don't actually think it's better than any of the others, it just seems to work for me. Anytime prior to 6 months ago, I'd have said Midjourney was by far the best. But now, I'd probably say to try the free trial on Dreamstudio first, there's very little between that and Midjourney. But if you struggle to get what you want using Dreamstudio, then you'll probably have more luck on Midjourney.
 
which I think may be a limitation OpenAI have put in to discourage deepfakes).
Interesting.

And yeah, same re: they're all pretty good. There isn't one that jumps out at me as clearly superior. Just pros and cons, and mostly minor ones.
 
Interesting.

And yeah, same re: they're all pretty good. There isn't one that jumps out at me as clearly superior. Just pros and cons, and mostly minor ones.
Agreed. And with every update they're getting better and better. They were all almost unusable for anything other than horror images just over a year ago, but they've pretty much replaced stock images for me now. The only times they all seem to struggle is when you try to generate an image with multiple humans which are meant to be the focus of the image. But even then, you can still get something usable after a few goes.

I'll be interested to see what the next Midjourney update looks like.
 
I genuinely think that any of the latest versions will be good enough. But my opinions on the main models are:

SDXL has almost caught up to Midjourney on people and faces. Loads of options to use it as well, including self-hosting if you're really going to be producing images at scale. It can produce text in images as well. I'd probably go with a trial on Dreamstudio first (you'll get some free credits to test it out) and just see what you think of the images. You can use specific styles (cinematic/anime/etc) without having to write a full prompt describing the style. And if you're wanting to get certain poses, or to create a scene based on a photo of a landscape, you can upload a reference image and set the noise so it can just about see the outline, then generate images with a prompt based on that. That feature is why I cancelled Midjourney and stuck with vanilla Dreamstudio.

Dall-E 3 is very impressive and completely free in Bing Chat, so it's worth using. Again, pretty much catching up to Midjourney for people and faces (although, it does seem a little more "cartoon-y", which I think may be a limitation OpenAI have put in to discourage deepfakes).

Midjourney does seem to be a lot easier to get what you're looking for, which may be down to the way they've trained it on human feedback. Their decision to produce 4 low res images at a time and then asking the user to upscale the ones they like was a stroke of genius. They have years and years of data of what outputs people liked and didn't like. So it's usually very easy to get exactly what you want by using Midjourney.

Plus, there are hundreds and hundreds of tools built on Stable Diffusion which have recently upgraded to XL. Supermachine is pretty good and has a lot of fine tuned models, there's a few (including one on Appsumo now, I can't remember the name) which allow you to generate consistent characters. So, if you want something where you just select a model and don't have to work too hard on the prompt you should be able to find it out there.

Personally, I think they're all good enough at pretty much anything now. There isn't really a "best" in my experience. I've chosen Dreamstudio, but I don't actually think it's better than any of the others, it just seems to work for me. Anytime prior to 6 months ago, I'd have said Midjourney was by far the best. But now, I'd probably say to try the free trial on Dreamstudio first, there's very little between that and Midjourney. But if you struggle to get what you want using Dreamstudio, then you'll probably have more luck on Midjourney.

Thanks for your detailed reply - really appreciated.
 
I have been using Midjourney for a very longtime, I think it's the best to me so far.
 
Midjourney is pretty good. I was using it for creating cartoon profile pictures and it did a great job.
 
Hello guys,

I haven't used AI image-generated yet.
I am working on a new project and this kind of images should do a good job on it as it is fantasy related.

In your experience, which would be the best and easiest one to use?

Thanks for your guidance :)
You can use aipronet.com very conveniently
 
You can try akool, facewapper.ai, instantpotrait for image swapping.
for generating purpose, I recommend Midjourney
 
Back
Top