|
Comfy Fleece Sweater posted:I don’t disagree, it’s just the first AI text-to-3D I’ve seen in the wild. I wonder who else might be working on 3D AI. https://twitter.com/NJetchev/status/1571044012018245634
|
# ? Sep 24, 2022 19:13 |
|
|
# ? May 30, 2024 13:40 |
|
gently caress You And Diebold posted:I found that if I tried something that requested too much memory it would error out but never free up the memory it grabbed so I couldn't run anything else after that, had to quit out, restart it and try something a little less resource intensive. I had to actually close the browser and pretty much everything else to avoid running out of system RAM But GPU seemed to be clear, I don't think there's much else I could've done there.
|
# ? Sep 24, 2022 19:20 |
|
Forgot to post this on Monday but this is "a picture of the queen when she gate crashed my wedding party as she was staying in the same hotel!"
|
# ? Sep 24, 2022 19:27 |
|
https://twitter.com/runwayml/status/1568220303808991232 https://twitter.com/runwayml/status/1572210647336882177
|
# ? Sep 24, 2022 19:33 |
mobby_6kl posted:Yes I was! I still don't understand why it wasn't working but whatever, I created a new environment and it was fine. Eventually I got to the point where it runs out of memory. I think textual inversion needs a ton of vram, but there's an issue with some info about running it with less.
|
|
# ? Sep 24, 2022 20:24 |
gently caress You And Diebold posted:https://twitter.com/runwayml/status/1568220303808991232 The one-click tracking seems new, but the other video feels like they are awfully exaggerating since the "video" part of "text to video" is actually imported and they are just applying AI masking (which already exists, it's even available in the automatic repo) and powerpoint text on top of it.
|
|
# ? Sep 24, 2022 20:26 |
|
Boba Pearl posted:High-effort posts automatic1111 was remarkably straightforward to install and I'm blown away with how quickly the community has developed all these tools and scripts but does anyone have any inpainting advice? I'm wondering what the best method is to fix weird hands and whatnot - do you just tightly mask the hand, or include some of the surrounding area? What about the prompt, change it to something basic like "open hand waving" or whatever?
|
# ? Sep 24, 2022 20:32 |
|
I haven't messed with this too much, but one benefit of the Automatic1111 build is the inclusion of the negative prompt, so you can put stuff like "three arms" or "mutant hands" there to minimize that sort of stuff.lunar detritus posted:The one-click tracking seems new, but the other video feels like they are awfully exaggerating since the "video" part of "text to video" is actually imported and they are just applying AI masking (which already exists, it's even available in the automatic repo) and powerpoint text on top of it. The tracking also isn't particularly impressive either. I mean it's cool and better than what I used in After Effects years ago but not like mindblowing AI poo poo. lunar detritus posted:I think textual inversion needs a ton of vram, but there's an issue with some info about running it with less. Ooh thanks, I'll try it out. E: the resolution was already at 256 mobby_6kl fucked around with this message at 21:22 on Sep 24, 2022 |
# ? Sep 24, 2022 20:34 |
|
code:
|
# ? Sep 24, 2022 21:09 |
|
"Close up of shelves filled with many different kinds of objects of different shapes and sizes and colors, photo realistic, studio lighting, in the style of Brian Froud, Photorgraphy by andreas gursky" "a tower, 100 tower, -100"
|
# ? Sep 24, 2022 21:13 |
Comfy Fleece Sweater posted:I don’t disagree, it’s just the first AI text-to-3D I’ve seen in the wild. I wonder who else might be working on 3D AI. I've been playing with Disco Diffusion's animation to spin a camera and 'rotate-around-a-point'. Then feed the images to Meshroom (free pictures-to-model software) to attempt to pull out a model. I've gotten close, but no full model just yet.
|
|
# ? Sep 24, 2022 21:18 |
|
I did an experiment with upscaling video with Real-ESRGAN 4x plus. Results were not great. Original https://i.imgur.com/xik2Ool.mp4 Upscaling with Real-ESRGAN 4x plus Experiment https://i.imgur.com/P8uCdPQ.mp4
|
# ? Sep 24, 2022 21:25 |
|
Comfy Fleece Sweater posted:I was just getting into 3D modeling, now I feel like it’s just a matter of time before modelers get a kick in the balls like illustrators did Just wondering. How did material artists take it when digital art got to the point where it was hard to tell between a photo of real artwork and output from Photoshop? I know that when every new genre of physical art came out, everyone hated it and did not accept abstract, pointillism, and even realism for a good while in each case. As an example, German around the 1920s-1930s constantly made exhibits for "degenerate" art of not only Polish/Jews but also Germans who made art using such genres like abstract and anything not modernist. https://en.m.wikipedia.org/wiki/Degenerate_art posted:During the dictatorship of Adolf Hitler, German modernist art, including many works of internationally renowned artists, was removed from state-owned museums and banned in Nazi Germany on the grounds that such art was an "insult to German feeling", un-German, Freemasonic, Jewish, or Communist in nature
|
# ? Sep 24, 2022 22:33 |
|
EVIL Gibson posted:Just wondering. My brother-in-law is a moderately well-known illustrator with a couple of well-received graphic novels, a whole bunch of New Yorker covers, and other stuff, and he switched to Photoshop as soon as there was an Oil Paint filter. He loves the speed and freedom to try thirty different concepts and not waste materials doing it.
|
# ? Sep 25, 2022 00:28 |
|
making xbox 360 era shooter covers
|
# ? Sep 25, 2022 00:33 |
After a bit of loving around gently caress you Python! finally got that Automatic1111 local install working. Early days yet but it's pretty nice and even on my 6Gb 1660 non super from three or four years ago I can make 512x512 in 15 seconds. It's also just very nice to not be tied to credits or whatever. It's psychological to some extent but it feels significantly easier to just throw poo poo at it when it's "free".
|
|
# ? Sep 25, 2022 01:14 |
|
Dick And Balls, beautiful lighting, Greg Rutkowski
|
# ? Sep 25, 2022 01:37 |
|
goatse, his cheeks spread wide, ((((ring))))
|
# ? Sep 25, 2022 01:54 |
|
I got some beautiful Halloweeny images from SD but I used the Greg Rutkowski prompt and it feels like cheating E: I also learned my initial headache with ram issues with SD was because the mobile 3060ti has only 6gb of vram, not 8 TheWorldsaStage fucked around with this message at 02:05 on Sep 25, 2022 |
# ? Sep 25, 2022 01:57 |
|
Yosemite Sam Xbox Videogame cover The roughest, toughest, mean ol’ hombre to ever jump a claim.... motherFUCKER
|
# ? Sep 25, 2022 02:38 |
|
Comfy Fleece Sweater posted:Dick And Balls, beautiful lighting, Greg Rutkowski pixaal posted:goatse, his cheeks spread wide, ((((ring))))
|
# ? Sep 25, 2022 02:52 |
|
IShallRiseAgain posted:I did an experiment with upscaling video with Real-ESRGAN 4x plus. Results were not great. The trick is that the Photo GANs aren't temporally stable so you get shimmer and artifacting from that. You really have to use a recurrent model built for sequential data or video. The other catch there is that most models right now are built by starting with good video and crapping it up so they have ground truth. The problem being they, well, suck at crapping up video realistically. They almost universally just down-rez it and maybe, if they are feeling frisky, blur it a little bit. Real SD video is usually covered with all sorts of compression artifacts so even state of the art video upscalers usually eat poo poo if you don't do some work ahead of time to clean up the source as much as you can. I took your video and ran this an ESRGAN 1x_NMKD_dejpeg_Jaywreck3-Lite_320k pass (cleans up blocking and JPEG like artifacts, one of the better general purpose one of these I've found even for non-JPEG). Ran PaddleGAN's implementation of PPMS-VSR on the now much cleaner input to do the super-resolution and frame blending. Ran it through a downscale in Handbrake back down to 720p to get it back down to web postable size. Then went over to FFMPEG and muxed the original audio back onto the new video file. Finally I ran one last pass and remux with ESRGAN using 1x_Sayajin_DeJPEG_300k_G to color correct it. Even after all that the results are better but still only mediocre. Most noticeably on the shots where lots of small text is visible. Even in the original video those fine details are destroyed and the AI just doesn't have enough info to properly restore it so it still smudgy. https://i.imgur.com/Otu2lmk.mp4 Imgurs compression makes it look a little worse but its not much better than that raw unfortunately. Arguably at this resolution its kind of a moot point and just some back belocking/bebanding/CAS sharpening in Handbrake would give as good or better results. Edit: Also Imgur seems to have ignored the aspect ratio settings on that upload but I'm too lazy to fix it rn so you get stretched rear end widescreen I guess. Edit2: OK I lied it was bothering me so I fixed it. Objective Action fucked around with this message at 04:33 on Sep 25, 2022 |
# ? Sep 25, 2022 04:21 |
|
run it through LDSR and report back in a year when it's done
|
# ? Sep 25, 2022 04:27 |
|
Have some neat combat mechs in a cyberpunk city: "Combat Mech, tech, ruined cyberpunk city, twilight lighting , octane render, ultra detailed"
|
# ? Sep 25, 2022 05:38 |
|
I have to say, I've been using tech for decades, and this is the first time it seriously feels like actual magic Like it's going into another dimension and pulling images from some other universe. Or casting a spell, summoning illusions with words. Writing prompts into my black spellbook. I know how computers work, but I don't know how it does *this* (I've been watching some ML videos, and I get the concepts, but it still blows my mind when it makes a perfect image) ... so yeah, I really like it, particularly as someone who is not gifted in art but does enjoy art BTW If anyone has an M1 mac, they got it running on apple silicon (really alpha, but it works) https://diffusionbee.com/ It's not nearly as good as the Windows offerings yet, but promising
|
# ? Sep 25, 2022 05:55 |
|
Comfy Fleece Sweater posted:Like it's going into another dimension and pulling images from some other universe. This should be right up your alley then. Link for large picture. https://imgur.com/a/yIUnxEl I can't wait to feed this thing every single Heavy Metal comic.
|
# ? Sep 25, 2022 07:57 |
|
Regarding negative prompting: I've been wondering how it tags certain phrases. For instance, if you want better hands by negative prompting "bad hands", will it simply give you less hands because it reads the word "hands" as its own negative tag? I'm inclined to believe that this might be the case, because in the SD beta they showed how our prompts were being processed, and each word was read as a separate tag. I guess what I'm asking is, is there a single word for deformed hands in any language?
|
# ? Sep 25, 2022 08:16 |
|
KinkyJohn posted:Regarding negative prompting: I've been wondering how it tags certain phrases. For instance, if you want better hands by negative prompting "bad hands", will it simply give you less hands because it reads the word "hands" as its own negative tag? Ectrodactyly
|
# ? Sep 25, 2022 08:18 |
|
KinkyJohn posted:Regarding negative prompting: I've been wondering how it tags certain phrases. For instance, if you want better hands by negative prompting "bad hands", will it simply give you less hands because it reads the word "hands" as its own negative tag? deformed
|
# ? Sep 25, 2022 08:23 |
|
KinkyJohn posted:Regarding negative prompting: I've been wondering how it tags certain phrases. For instance, if you want better hands by negative prompting "bad hands", will it simply give you less hands because it reads the word "hands" as its own negative tag? Just like a real artist, the best way to not draw hosed up hands is not to draw them at all…
|
# ? Sep 25, 2022 08:23 |
|
I haven't actually done img2img until now since using it was awkward in the build I've been using. It's pretty interesting how much it does dictate the composition, but still allows a ton of variation in the subjects and little details.BARONS CYBER SKULL posted:Just like a real artist, the best way to not draw hosed up hands is not to draw them at all…
|
# ? Sep 25, 2022 08:34 |
|
I really want to see some Mars Volta lyrics in these because it's all gibberish anyway preferably from their first album deloused in the comatorium
|
# ? Sep 25, 2022 08:54 |
|
I tried to create The Mars Volta with "a photo of pretension incarnate wearing a helmet made of curly hair" but Dall-E didn't get what I was going for at all sadly
|
# ? Sep 25, 2022 09:02 |
|
The biggest properly tagged data goldmine of images surely belong to the stock image behemoths like Shutterstock. This makes me wonder what their future plans are in an ai world. They could make their own dataset and ai model to sell, but they will probably be outcompeted by true open source projects. You get 750 images per month on their subscription plan. So if there were a community of stock image users contributing whatever they downloaded each month into a database, you can quickly build a high quality and properly tagged stock image dataset.
|
# ? Sep 25, 2022 10:59 |
|
KinkyJohn posted:The biggest properly tagged data goldmine of images surely belong to the stock image behemoths like Shutterstock. This makes me wonder what their future plans are in an ai world. Getty Images banned AI art and they are making a stink about all of it being a violation of copyright so the technology should be banned or restricted to billionaires or something https://arstechnica.com/information-technology/2022/09/fearing-copyright-issues-getty-images-bans-ai-generated-artwork/
|
# ? Sep 25, 2022 11:08 |
|
Boba Pearl posted:I will write you one my friend. Any chance this could get added to the op?
|
# ? Sep 25, 2022 11:17 |
Corporations freaking out too late and screaming "AI must be illegal!" is going to be a common refrain. Digitization and automation aren't stopping. No one should be surprised when the leopard turns and eats their face. Unless we get great filtered the information age is here to stay.
|
|
# ? Sep 25, 2022 11:24 |
|
Rutibex posted:Getty Images banned AI art and they are making a stink about all of it being a violation of copyright so the technology should be banned or restricted to billionaires or something Progress is inevitable. Adapt or perish.
|
# ? Sep 25, 2022 11:24 |
|
aren't Getty Images pretty infamous for stealing poo poo? edit: https://www.latimes.com/business/hiltzik/la-fi-hiltzik-getty-copyright-20160729-snap-story.html
|
# ? Sep 25, 2022 11:33 |
|
|
# ? May 30, 2024 13:40 |
|
Yeap pretty much all of the stock image corps are screwed. But to be fair, gently caress them. They have hosed over photographers and artists for the longest time, thanks for paying me less than 10 cents royalties per image used They even crowdsource their image tagging by making you meticulously tag and name everthing yourself. Which is why it would be the sweetest twist of fate if the "crowd" sources back their tags to build a stock image model that will put them out of business
|
# ? Sep 25, 2022 11:49 |