MidJourney is actually unbelievable
Just did a video review of MidJourney.
Here: vibehut.io/watch/63ae751a35b9ab001517f493
MidJourney is a generative art AI that operates its beta in a discord channel.
The discord has 6m+ users and had 450k online, clearly I'm living under a rock!
Using the prompt
/imagine beautiful website about cars in the year 2100 ux ui ux/ui website
In 20 seconds, it generated these high resolutions images of futuristic car landing pages.
What else should we ask it?
Interestingly you can feed midjourney with a URL of an existing design you already like and request the model to tweak it to your needs. The results have been mind-blowing so far (ran a few tests on dribble samples).
Where I see an opportunity for a SaaS.
A solution that automatically convert the model's output into a usable figma file.
Totally, this is 🤯! Would love to chat more about this
It happened. https://www.usegalileo.ai/
I've seen it too! Requested early access, can chat about it in any of the upcoming discussions here https://vibehut.io/rooms/6304c25f8ac27200164bd65f
@usehappen My hands are rather full. But happy to be the first beta customer if an IH wants to give it a go! 😉
Try my website: https://www.simulai.co we have midjourney style ready.
This is really good. Built on openAI api?
Midjourney is an absolute titan. The only downside is the lack of API, or really any proper UI at all. Discord is CLUNKY!
It's better if you subscribe and use your own private discord to generate the images. I agree though, an API would be best. I expect them to provide this at some point. I literally generated images of an English Quill for the app I'm working on dev.esy.com/about. The idea that I can produce, own, and commercially use images like this for a mere $30 a month rather than hire a freelance artist is insane.
Agree! And I wish the free trial wasn't so quick, I was just starting to get the hang of things!
I've joined the discord of mid journey, and I was shocked.
That’s crazy, after each and every prompt it gets better.
It's so amazing
If you like midjourney, have your tried stable diffusion?
It's much more versatile due to various different models and free to run on your own PC. But all the settings will take some getting used to.
My friend called MJ the Mac of the AI art world and Stable Diffusion the Linux, and I agree with him. But no worries, it's much easier to learn than you think.
And if you're considering creating an AI app with stable diffusion, we're currently finishing up a stable diffusion API at Evoke + more models and dreambooth planned in the future.
Feel free to sign up for our newsletter to get notified when we launch or join our discord
Will try Stable diffusion today, thanks!
I find the results consistent and generally quite remarkable with MJ. SD and Dalle, not so much. but that might just be because SD isn't an "out of the box" tool like MJ. Curious, what's your take?
I've heard some people say MJ isn't just a really good AI, but that there's additional code magic going on behind the scenes for you to get a good result. Whereas Stable Diffusion and DALLE just expose you to the raw model.
For example, if you put in "snowy peaks above alpine forest", you'll get a generic image with SD, but with midjourney, they might inject additional prompting like "volumetric lighting, masterpiece, award winning photography, etc" into your prompt behind the scene to return a better result.
They also automatically upscale all your images so everything comes out at a high resolution.
Combine this with the fact that MJ constantly updates its model with new images by taking the best rated community images and feeding them back into the model.
It's no doubt why MJ produces the best quality.
Major disadvantage is a strict NSFW filter and lack of API. So devs making AI apps are severely limited.
However, there is an open source stable diffusion model that produces MJ-like outputs call MJ v4 you can check out.
We plan to have that model and many others on our AI API service once we launch.
I checked out your website. It's an awesome idea. When are you lunching?
Also just to add, you might need to fix some alignment/UI on your website though.
I suspect you will fix these after lunch any. Just pointing it out incase.
Launching early january!
Is there any reason the believe that the model that SD is trained on is "worse" than the one that MJ is trained on?
Can we chalk up MJ's superior performance entirely to an appendix being added to each prompt behind the scenes?
MJ is closed source, so I don't know.
It's just a high likelihood that the appendix is partially responsible for better results.