tater Posted August 23, 2022 Share Posted August 23, 2022 (edited) Been messing with a few of these, and I thought it should probably get a thread. https://www.midjourney.com/home/ https://openai.com/blog/dall-e/ https://stability.ai/blog/stable-diffusion-public-release Those are text prompts to images... incredibly fun to play with. Midjourney: Prompt was "the hydrogen sonata" A Photoshop plugin I am thinking of getting does an astounding job at recovering faces/details in images. Gigapixel AI: https://www.topazlabs.com/gigapixel-ai My grandfather, circa 1918. I think the AI did a decent job, the one on the right is shockingly close to what I saw in the mirror at his age. To be clear, left is a scan of an old family photo, right is what gigapixel did in a couple seconds. Edited March 30, 2023 by tater Quote Link to comment Share on other sites More sharing options...
Geonovast Posted August 23, 2022 Share Posted August 23, 2022 11 hours ago, tater said: Those are text prompts to images... incredibly fun to play with. Had a fun run with craiyon recently. My pfp is shamelessly taken from one of its results. It's certainly not the best one out there, but it's really fun since most of its drawings look like the scribblings of a hyperactive 6 year old whose psych meds stopped working. I don't remember the prompt for this one, but one of my favorites. Since craiyon gives rather small pictures, I also ran this one through an AI upscaler. Quote Link to comment Share on other sites More sharing options...
tater Posted August 23, 2022 Author Share Posted August 23, 2022 Midjourney is pretty "artsy" in output, DALL-E seems more photo-like, though I have yet to make anything that I like with it, limited tries so far. I have not messes with Stable Diffusion yet. (one of the midjourney variations of "the hydrogen sonata") Quote Link to comment Share on other sites More sharing options...
tater Posted August 23, 2022 Author Share Posted August 23, 2022 This prompt was "there are rocks in those clouds": Quote Link to comment Share on other sites More sharing options...
tater Posted August 25, 2022 Author Share Posted August 25, 2022 Quote Link to comment Share on other sites More sharing options...
DDE Posted September 4, 2022 Share Posted September 4, 2022 https://www.vice.com/en/article/bvmvqm/an-ai-generated-artwork-won-first-place-at-a-state-fair-fine-arts-competition-and-artists-are-liquided Quote Link to comment Share on other sites More sharing options...
tater Posted September 5, 2022 Author Share Posted September 5, 2022 @Geonovast noticed your icon, and have many minutes of CPU to burn on midjourney... Quote Link to comment Share on other sites More sharing options...
JoeSchmuckatelli Posted September 6, 2022 Share Posted September 6, 2022 AI generated... Won art competition. https://www.google.com/amp/s/amp.cnn.com/cnn/2022/09/03/tech/ai-art-fair-winner-controversy/index.html Quote Link to comment Share on other sites More sharing options...
tater Posted September 6, 2022 Author Share Posted September 6, 2022 Quote Link to comment Share on other sites More sharing options...
kerbiloid Posted September 6, 2022 Share Posted September 6, 2022 Soon the neuronets will be ecranizing books and make personally optimized movies based on the user preferences. And generate videoclips on the fly based on the music and the user preferences. This will immerse himans into the virtually changing augmented reality. Quote Link to comment Share on other sites More sharing options...
DDE Posted September 7, 2022 Share Posted September 7, 2022 On 9/6/2022 at 10:30 PM, kerbiloid said: Soon the neuronets will be ecranizing books and make personally optimized movies based on the user preferences. To be fair, with the kind of adaptations we've geen getting, they can't doo much worse. The real threat is recursive apocalypse: neuronets being trained on material predominantly from other neuronets. Quote Link to comment Share on other sites More sharing options...
tater Posted September 7, 2022 Author Share Posted September 7, 2022 The visual stuff is getting better week by week. This will be interesting to watch. The ability to label content will be very useful. Ie: label a character you generate, and that data includes a seed so that you can then ask for that same character in another image. Say "Luke" render him in 3/4 view wearing a robe, in a desert with 2 moons. Then render in an orange flight suit with helmet. You could storyboard a movie, or completely render a comic or graphic novel with just prompts. The next big thing would be to have the engine not look just at labeled 2d images, but have more labeled 3d model data. Game engines might be good for this (UE5?). The idea is for it to start to be able to deal with objects in 3-space. You can ask for buildings, and in some cases they look OK until to pay close attention, then they make no sense. The ability to have it generate a 3d building—inside and out—would be pretty amazing. Quote Link to comment Share on other sites More sharing options...
tater Posted September 8, 2022 Author Share Posted September 8, 2022 Quote Link to comment Share on other sites More sharing options...
DDE Posted September 8, 2022 Share Posted September 8, 2022 11 hours ago, tater said: The ability to have it generate a 3d building—inside and out—would be pretty amazing. I think they'll basically have to do what they did for text - have to teach the AI to understand the connection between design elements of each and every object you want to emulate. Right now... remember how in Solaris, the first time the girl "spawns", her dress is just a continuous tube, and the buttons are non-functional? That's what the AI is doing with buildings. Quote Link to comment Share on other sites More sharing options...
JoeSchmuckatelli Posted September 8, 2022 Share Posted September 8, 2022 The more I look at these, the more fascinated I get. I think the criticism is misplaced - because there is definitely a fusion between the ai generated image and the human artist who curates the image(s). Some of the compilations I've seen are better than others. This is good This is meh Maybe I'm just drawn to the art style of one more so than another (first is Mind journey, second is StableDiffusion) - but I think the first Hotel California video is more evocative. I would not limit the folks who are doing this to being merely curators. Just as you can call an animator or videographer an artist - the art is in selection of the image and using it/them in telling a story. This shows a bit of the process - b/c the artist chose to present multiple images rather than selecting only one. @tater - this is a fascinating rabbit hole! Quote Link to comment Share on other sites More sharing options...
tater Posted September 8, 2022 Author Share Posted September 8, 2022 6 minutes ago, JoeSchmuckatelli said: - this is a fascinating rabbit hole! Right? Quote Link to comment Share on other sites More sharing options...
DDE Posted September 23, 2022 Share Posted September 23, 2022 Alex Andreev having his fun with one of the more basic modes of Midjourney Quote Link to comment Share on other sites More sharing options...
tater Posted September 28, 2022 Author Share Posted September 28, 2022 Quote Link to comment Share on other sites More sharing options...
kerbiloid Posted September 30, 2022 Share Posted September 30, 2022 The future universal human salute gesture. Spoiler Spoiler Because the Matrix still can't into fingers properly. "I feel something weird here... Is it neuronet?" "Yes, it is. Look, it's confused by the number of fingers." Spoiler So, when you meet a new person, first ask him her them it to show its hands. Just to be sure that it' s real rather than a deep fake. Quote Link to comment Share on other sites More sharing options...
tater Posted October 1, 2022 Author Share Posted October 1, 2022 Quote Link to comment Share on other sites More sharing options...
tater Posted October 12, 2022 Author Share Posted October 12, 2022 Quote Link to comment Share on other sites More sharing options...
DDE Posted October 13, 2022 Share Posted October 13, 2022 Stupid question: why can't many of these (e.g. Stable Diffusion) be run on a consumer laptop at an appropriate glacial speed? Why is there a really tall minimum system requirement to neural nets? Quote Link to comment Share on other sites More sharing options...
DDE Posted October 14, 2022 Share Posted October 14, 2022 Inspiration to prior question: NovelAI, whose original product is a GPT text 'game', have their own version of Stable Diffusion as well. And in theory it's basically another Waifu Diffusion. But, except for the frequent issues with eyes - it seems to use anime pupils on normal-proportioned faces - I haven't had too much trouble taking it in a wholly different direction. Spoiler Quote Link to comment Share on other sites More sharing options...
Rutabaga22 Posted October 14, 2022 Share Posted October 14, 2022 I love midjourney. I did afterlife as a promt and got some kind of beautiful fantasy landscape. Quote Link to comment Share on other sites More sharing options...
tater Posted October 16, 2022 Author Share Posted October 16, 2022 On 10/14/2022 at 2:17 PM, Rutabaga22 said: I love midjourney. It has a style all its own, I agree. Quote Link to comment Share on other sites More sharing options...
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.