Neat! It seems like Midjourney's biggest "value add", if you can call it that, is that everything is beautiful. That's fantastic if you're just trying to create something that looks good, but that can actually get in the way (as you rightly point out) if you're looking for specific stylization.
Although that conclusion and my other observations only hold specifically for "vanilla" MJ 5.2 and SDXL 1.0.
Midjourney has the "--style raw" parameter that specifically reduces the influence of the default aesthetic and makes the model more receptive to your input. At the same time, SDXL 1.0 has many "styles" you can apply to a prompt to nudge it in the right direction.
And of course most of the unintended results can be further manipulated by more thorough text descriptions and "negative" prompting (telling either model what you want to avoid in an image.)
Still, I had to give a level playing field to each model for this comparison.
It sure feels like they're all kind of converging on a great service, all picking up the features from one another, just like with the LLMs that are developing. For a few months there, it felt super important and urgent to understand every single new feature and how to use it, but that feeling has subsided for me, and it's now just very interesting to see these different innovations.
I think Midjourney is really trying to build a one-stop-shop on their own platform (they're also working to have proper website UI instead of relying on Discord).
Stable Diffusion is more of a raw open-source model that can be modded, improved upon, and integrated with lots of other services and models. So if you're more tech-savvy and patient, you can probably get much more out of Stable Diffusion.
Neat! It seems like Midjourney's biggest "value add", if you can call it that, is that everything is beautiful. That's fantastic if you're just trying to create something that looks good, but that can actually get in the way (as you rightly point out) if you're looking for specific stylization.
True!
Although that conclusion and my other observations only hold specifically for "vanilla" MJ 5.2 and SDXL 1.0.
Midjourney has the "--style raw" parameter that specifically reduces the influence of the default aesthetic and makes the model more receptive to your input. At the same time, SDXL 1.0 has many "styles" you can apply to a prompt to nudge it in the right direction.
And of course most of the unintended results can be further manipulated by more thorough text descriptions and "negative" prompting (telling either model what you want to avoid in an image.)
Still, I had to give a level playing field to each model for this comparison.
It sure feels like they're all kind of converging on a great service, all picking up the features from one another, just like with the LLMs that are developing. For a few months there, it felt super important and urgent to understand every single new feature and how to use it, but that feeling has subsided for me, and it's now just very interesting to see these different innovations.
I think Midjourney is really trying to build a one-stop-shop on their own platform (they're also working to have proper website UI instead of relying on Discord).
Stable Diffusion is more of a raw open-source model that can be modded, improved upon, and integrated with lots of other services and models. So if you're more tech-savvy and patient, you can probably get much more out of Stable Diffusion.
I actually covered this in this post from April: https://www.whytryai.com/p/sd-midjourney-dall-e-android-ios-windows - much of it still holds up.
That sounds a bit like the Apple vs Android dichotomy. I wonder if that's similar to where we end up.
That's exactly the comparison I make (with DALL-E somewhat unfarily ending up as the Windows Phone)
I love my Sidekick 2 and you can't have it!
I mean, nothing.