Ask HN: Why is Midjourney better than Stable Diffusion?

#1

Post by gavribirnbaum » Sun, May 28, 2023, 4:59 PM UTC

I wonder what they do better than Stable diffusion to generate the images. It seems more responsive to prompts and able to generate higher quality results.

Re: Ask HN: Why is Midjourney better than Stable Diffusion?

#2

Post by brucethemoose2 » Sun, May 28, 2023, 5:06 PM UTC

Stable Diffusion 1.5 is state of the art... Heck, its not even close. But you have to run your own instance and dive into the ridiculous number of tools/models the community has developed to take advantage of it.

Midjourney has a strong (and improving?) base model and probably automates the usage of some similar tools behind the scenes. So on the surface, it is indeed better.

Re: Ask HN: Why is Midjourney better than Stable Diffusion?

#3

Post by brightside67 » Sun, May 28, 2023, 5:10 PM UTC

bing/images is a good one btw. powered by private DALLE ver

Re: Ask HN: Why is Midjourney better than Stable Diffusion?

#4

Post by gavribirnbaum » Sun, May 28, 2023, 5:57 PM UTC

brucethemoose2 wrote: post #2

Stable Diffusion 1.5 is state of the art... Heck, its not even close. But you have to run your own instance and dive into the ridiculous number of tools/models the community has developed to take advantage of it. Midjourney has a strong (and improving?) base model and probably automates the usage of some similar tools behind the scenes. So on the surface, it is indeed better.

What tools are you referring to and how are they being automated?

Re: Ask HN: Why is Midjourney better than Stable Diffusion?

#5

Post by brucethemoose2 » Sun, May 28, 2023, 8:27 PM UTC

gavribirnbaum wrote: post #4

brucethemoose2 wrote: post #2

Stable Diffusion 1.5 is state of the art... Heck, its not even close. But you have to run your own instance and dive into the ridiculous number of tools/models the community has developed to take advantage of it. Midjourney has a strong (and improving?) base model and probably automates the usage of some similar tools behind the scenes. So on the surface, it is indeed better.

What tools are you referring to and how are they being automated?

TBH I am not familiar with today's Midjourney, but I am thinking of things like face restoration, tiled VAEs for better high res generation, and self attention.

But many tools (like controlnet, or using images as input) require more manual control.