You can rent cloud computing for them very cheap.
In order to download it, nothing more. "Open source" still isn't "totally free for whatever" due to the fact that they are still licensed products. Keeps companies from doing negative things with them, like claiming the product is theirs, patenting it and then suing anyone who releases an open source product based on it. Or releasing a product that was meant to be free forever under a paid license or with DRM etc.
That's exactly what Stable Diffusion is. I run it on my PC. You can even run it on your CPU if you can't on a GPU or a cloud host. It will just be slow.
(((isometric perspective))) and then put the word grid as a negative prompt.
One that gives you permission to use it for free, so long as you aren't using it for something criminal.
No, I get you, you want this to be the one piece of higher-end software you own with zero license at all. Good luck with that.
Of course. That's part of the redistribution license.
You mean like this?
There is potential art-wise, but imo renaissance is a bit of an exaggeration, since you also need devs with the passion to design clever gameplay. Do those devs exist?
There is potential art-wise, but imo renaissance is a bit of an exaggeration, since you also need devs with the passion to design clever gameplay. Do those devs exist?
With the Krita plugins and img2img, it's a very powerful addition to any digital artist's toolbox, just like Photoshop's Content Aware features have been for several years, so I don't think it's a waste of time at all.
People do the same thing as well, both have to be programmed to generate an image either from multiple instances or from just one, the general idea behind 'ai' is not having to specialize doing that or wasting ones time doing it, like a typical tool, just organize a pool and let it do its own thing.A bit of a simplification. More accurate to say it's copying. You are asking the AI to copy something similar to hundreds, thousands or tens of thousands of things it's seen.
It's like asking an experienced artist to draw for you, only this artist is more likely not to understand you and draw a human centipede.
It doesn't do 3D, but I think you could generate almost everything you'd need for a simple 2D Point&Click Adventure game or Visual Novel and the likes with what's currently out there. Highly detailed backgrounds, character art and portraits and items. You'd just need to do stuff like Animations/interpolation and bare-bones coding.
Pretty sure the main difference between the two is more that it's Closed Source and they manipulate the user Input server side to make it look certain ways other people also refer to as the "Midjourney look/style" rather than that they have some sort of Super Advanced model, while in SD you have to reach into the plumbing to make it do something extraordinary, but you seem to have a lot more control over the process. Maybe they also employ a somewhat better language model to interpret prompts better.
But it's already doing that. It doesn't really matter to the model whether the Output is 512x512, 1792x1024 or 3584x2048. It just hallucinates more detail based on the available image and prompt specification. The biggest problem with the current implementation of SDUpscale is that it separates the image into small pieces to iterate over, in the case of the final image above it was 25 different pieces, which will lead to visual aberrations as they're stitched back together (like with the sky at the top) the higher you set the Denoise that you need for it to imagine more detail.
Yeah, visual novels. How much money could that possibly make?
I know. But more so, and more intelligently. To the point that you can do Bladerunner-esque zooming in to any detail.
I have seen some really decent images come out of the newer Midjourney models, but I think a lot of the difference you're seeing is in who is using it and who is posting images.
I think it's mostly fine-tuning. Midjourney essentially seems to be doing something similar to what people are trying to do by adding "by Greg Rutkowski, Artgerm, Alphonse Mucha" to their prompts using SD (but probably a lot more complicated) which also makes its Outputs seem to all have a specific recognizable "look". It's a Black Box with the only way to interact with it being Discord. There's no way to know what happens with the prompt on the Backend and you'll obviously not get the same result trying to use it 1:1 like he did (you wouldn't anyway, since it's a different model). You'll have to come up with these "Optimizations" by yourself with SD, but that also means you have a lot more fine control over the resulting Output.
Wot?
Wot?
No. At best, they're probably pulling from highly rated images in certain datasets for their finetunes, as Dexter suggested. Maybe even using their own users' most upscaled photos along with user text inputs to improve their training as well.
Stable Diffusion also ran via Discord during their beta. No agreement was needed, just community-aided content moderation, which Midjourney also does. Various Stable Diffusion users have set up their own Discord bots. The Codex could as well, if we wanted.
