AI Rendered images using only a text string
Added 2022-06-04 17:24:37 +0000 UTCHey everyone, just want to give you an update. Returned from a vacation and now that our ranch has sold we bought a new place and we'll be moving soon which always throws a kink into things.
So, one of the more interesting things I've run across lately is this new idea that you can use AI to create amazing images. Here are a few of the ones that I've managed to squeak out with very little training and just a few hours of playing around.
The AI I'm using is called midjourney and unfortunately it's still in beta and only available by invite. You can register for a free invite by going to https://www.midjourney.com/
It's interesting how they provided you an interface. First off you need a Discord account and then you log into their server, very much like you might log in to our server ( http://cw1.me/discord ) and then they have a little bot (which is really just a macro) that you type in something like:
https://media.discordapp.net/attachments/982310672299876412/982427422588825630/Custom-Fiberglass-Aquarium-built-for-the-Tennessee-Aquarium.jpg , many people looking at a gigantic indoor saltwater aquarium full of fish and corral and a huge octopus, Ashley wood, ridley Scott, hyperealistic, maximum detail, photorealistic, cinematic lighting, --w 2096 --h 1080
The first part of that is just a URL to an optional seed image and in this case it's an image of an aquarium I found online. And the rest, is just a description of what you want the picture to be. Once the bot receives your request, it will generate 4 low resolution versions of what the AI thinks you're looking for. Then you have the ability to choose any of them and get variations on it, or you can upscale it which will create a high resolution image with a lot more detail.
I'm talking with developers about what it takes to create something like this inside of blender, and you just never know. I certainly would like to combine this with synth and see what it generates!
I'll be sure if let you know when this thing comes out of beta or if there are other viable alternatives. It's pretty GPU intensive so I suspect you need at least a 3090 to run it with any sort of efficiency. Of course, the midjourney version does all the processing in the cloud so there is no GPU needed on your end.
Stay tuned!
Comments
Yeah, I've been digging deep into this stuff and am now mostly all Disco Diffusion. I think I'll spend a little time talking about this technology and what it means for our industry.
Chipp Walters
2022-06-15 06:39:53 +0000 UTCYeah, I've been digging deep into this stuff and am now mostly all Disco Diffusion. I think I'll spend a little time talking about this technology and what it means for our industry.
Chipp Walters
2022-06-15 06:39:19 +0000 UTCI truly like the surreal output, it's almost dreamlike. The possibilities seem endless with the right phrasing. I also subscribed to Quick-Eyed Sky's page, as it teased some other interesting images done with AI. I spent about 2 hours following the rabbit holes, so I hope to try this stuff out somewhere along the line. For my profession, I use rather mundane tools like Maya/Vray and Unreal Engine, but the artist in me would love to try this out.
John Brennan
2022-06-15 04:09:47 +0000 UTCYes, on the $10/Mon subscription. Not private.
Chipp Walters
2022-06-08 13:37:26 +0000 UTCVery impressive! Btw, are you on a subscription plan, are these images shown in the feed, or are you paying for a private subscription? I'm trying to understand how this works.
Anton Grandert
2022-06-08 07:59:18 +0000 UTCRight--or the use of assets in a scene? (or the use of professionally mixed paints that you didn't create yourself for a painting). I agree that if the art is just "I want a butterfly eating the Chrysler building" and it turns out super cool the creative work or effort is almost all on the AI side. But assuming this gets used to augment processes I think it's wonderful.
Marco Chacon
2022-06-05 02:28:24 +0000 UTCTruly amazing! Just imagine if this AI model is sold within a render farm like this: https://blogs.gre.ac.uk/designsupport/renderfarm-machine/
Alejandro Tejada
2022-06-05 01:04:00 +0000 UTCImpressive and scary!
Romain Heck
2022-06-04 22:34:56 +0000 UTCI dunno but to me this is just another tool, much like a 3d rendering engine like Cycles. I mean you can feel good about a Cycles rendering, correct? Even if you didn't paint any of the pixels. My guess is it will take some time to figure out how best to use this sort of thing.
Chipp Walters
2022-06-04 22:25:48 +0000 UTCWow! If something like this comes to Synth... mind blowing!
Joe Maron
2022-06-04 20:30:03 +0000 UTCI get it but I do not like the idea of AI creating art because what ownership can you feel good about with this program doing all the work for you. Yes it looks very cool and there is a lot to think about with this type of programming. This will continue to make great artwork in the near future questionable if done by AI or artist.
Jon Moberly Creative 3D
2022-06-04 20:14:34 +0000 UTCVery interesting... Thank you! I have submitted for an invite and will see if I can get in to try some things...
Justin Jacobs
2022-06-04 19:21:00 +0000 UTCI used your exact terms but omitted "escort" as it's a banned word (no sexy stuff) and got these images: https://filedn.com/lLMW4jXsJqxXkRYjd1UCoKL/attack02.png https://filedn.com/lLMW4jXsJqxXkRYjd1UCoKL/attack01.png
Chipp Walters
2022-06-04 18:41:36 +0000 UTCSends back the image, no 3D involved. It uses a server farm loaded with 3090s to create the image. You don't need a 3090 to use it, but if they make a desktop version, which doesn't use a server, then it would use the local 3090-- not sure if that's in the works ATM.
Chipp Walters
2022-06-04 18:32:31 +0000 UTCMy God, this is incredible. I *so* hope they let me in.
Marco Chacon
2022-06-04 17:44:23 +0000 UTCSo the AI sends you back a jpg image or a 3D scene? because you wouldn’t need a 3090 for a jpg if the AI returns a jpg image…. I could totally see needing one for a 3D scene that the AI sends back and you have to render out.. very intrigued by this… I assume since it’s beta something like B17 bomber under attack, World War II, P51 Mustang escort, hyper realistic etc. probably wouldn’t create a result that would be as good as a more general description like the aquarium. The images are amazing though and I wonder how it’s programmed to combine those things.. does the AI scour the web looking for images and them composites them together? Great find!
Justin Jacobs
2022-06-04 17:43:43 +0000 UTChaha... I dunno if you have read the book Life 3.0 ... looks like it's getting real (the part where the AI becomes selfconscious and needs to expand and therefore needs to buy more cloud storage – starts to think about ways to make money ... it then produces scripts / movies nd starts a streaming service like Netflix. With the help of its AI and market research it becomes quickly the best service there is.)
Jay Dee
2022-06-04 17:36:32 +0000 UTC
