|
TheWorldsaStage posted:https://twitter.com/DeviantArt/status/1591113199218487300?t=Kmn-_vplmBbAExKupfX3tw&s=19 "we are using stable diffusion which was trained on all of your stolen art, but we promise not to steal any more going forward"
|
# ? Nov 11, 2022 19:47 |
|
|
# ? May 27, 2024 05:45 |
|
it still won't make me a convincing bologna trombone but at least it's pretty good at pianos
|
# ? Nov 11, 2022 19:55 |
|
I've spent an hour on another attempt to successfully install Automatic1111 and it's impressive how utterly dogshit its installation process is. This might be the first time I got pissed enough to want to break a coder's hands before I even launched their software.
|
# ? Nov 11, 2022 20:04 |
|
Megazver posted:I've spent an hour on another attempt to successfully install Automatic1111 and it's impressive how utterly dogshit its installation process is. This might be the first time I got pissed enough to want to break a coder's hands before I even launched their software. Reposting this, maybe you're making it harder than it needs to be? quote:Requirements: Windows 10 or 11 and an nVidia video card with at least 6 GB of VRAM (10xx series or better should work) I've used these steps to install SD a couple of times on different Windows machines with no issues. WhiteHowler fucked around with this message at 06:29 on Nov 13, 2022 |
# ? Nov 11, 2022 20:25 |
|
Rutibex posted:"we are using stable diffusion which was trained on all of your stolen art, but we promise not to steal any more going forward" Note that there's still no legal precedent on whether this counts as fair use, so it may or may not be legal for a big tech giant like e.g. Yandex to ignore those user preferences and scrape all that art for training purposes anyways. Megazver posted:I've spent an hour on another attempt to successfully install Automatic1111 and it's impressive how utterly dogshit its installation process is. This might be the first time I got pissed enough to want to break a coder's hands before I even launched their software. IIRC the installation process is just "have the right prereqs installed and then run one batch file and it'll automatically grab everything else". It was probably a lot smoother for me since I already had most of them from doing other programming stuff though. There are already some "professionally packaged" programs that Just Work(tm) without any of that computer janitoring though, like Diffusion Bee (sadly mac only). e: WhiteHowler posted:Reposting this, maybe you're making it harder than it needs to be? Some of these steps are unnecessary. GFPGAN is optional and iirc it can download it on demand now, and the cuda toolkit isn't needed either. Also if you want to update your install in the future you are better off installing git first and then doing a git clone instead of just downloading the source .zip, since that way you can use git pull to update to the latest version without redownloading the whole thing. RPATDO_LAMD fucked around with this message at 20:36 on Nov 11, 2022 |
# ? Nov 11, 2022 20:31 |
|
AARD VARKMAN posted:it still won't make me a convincing bologna trombone but at least it's pretty good at pianos i made this one earlier today "democracy made out of ham"
|
# ? Nov 11, 2022 20:39 |
|
RPATDO_LAMD posted:Note that there's still no legal precedent on whether this counts as fair use, so it may or may not be legal for a big tech giant like e.g. Yandex to ignore those user preferences and scrape all that art for training purposes anyways. I'm curious to see how this plays out I once saw someone describe encryption as, taking a chicken nugget, then trying to recreate the original living chicken based on the end result nugget AI training doesn't work exactly the same but you would have similar difficulty generating the original images the model was trained on based on the final model Presumably if an artist is listed in the model you could make a claim that their rights were infringed but without the model training log files you'd have difficulty proving specific images were used Maybe you could copyright your entire style but given how direct and indirect collaboration is key to the artistic process that seems unlikely
|
# ? Nov 11, 2022 20:46 |
|
Can you do "democracy if it were a BLT sandwich"
|
# ? Nov 11, 2022 20:47 |
|
Also apparently the opt out is for third party only, and it's unenforceable, at least that's what Twitter artists are saying of the fine print
|
# ? Nov 11, 2022 20:48 |
|
RPATDO_LAMD posted:Some of these steps are unnecessary. GFPGAN is optional and iirc it can download it on demand now, and the cuda toolkit isn't needed either. GFPGAN is optional but strongly recommended. I didn't know it would download on demand now though, I'll update. Cuda Toolkit isn't needed? I thought it was a requirement. quote:Also if you want to update your install in the future you are better off installing git first and then doing a git clone instead of just downloading the source .zip, since that way you can use git pull to update to the latest version without redownloading the whole thing. I intentionally wrote this for people who don't understand and/or don't want to monkey around with git.
|
# ? Nov 11, 2022 20:54 |
|
CUDA is required but it should automatically grab the version of pytorch that comes bundled with cuda 11.3. Unless your machine already has pytorch installed but no CUDA. Fair point on the git thing
|
# ? Nov 11, 2022 21:06 |
|
RPATDO_LAMD posted:IIRC the installation process is just "have the right prereqs installed and then run one batch file and it'll automatically grab everything else". Yeah, that was the issue. I am going to rant a bit and I'd like to state in advance that I am grateful to you for posting that and trying to help, this rant is not aimed at you: I've already figured out and been using the cmdr2 build of SD for a while, but I wanted to see if A1111 had any extra functionality I wanted. So, I already had Python. It was just a newer Python version and A1111 errors out when you have the temerity of offering it anything other than specifically Python 3.10.6, but also can't be arsed to just include that one specific version in its install. Cmdr2 SD just downloads and installs the Python version it needs when you install it. I also had Git which I downloaded A1111 with, it was just bitch-rear end lame-o noobface GitHub Desktop Git and not the cool h4xx0r command prompt Git, which is an insult to A1111. I needed to install that one specifically. Cmdr2 SD, again, just updates on its own whenever you run it, I don't know if it has its own Git or not. Admittedly, I didn't have CUDA. Cmdr2 SD, again, just installed it for itself on its own, as far as I can tell. But I didn't have any issues with it either way. Finally, apparently my specific GPU has issues with SD unless you add some extra arguments when you run it. You can find them in one of the Issues in the A1111 repo. Cmdr2 SD just detects the GPU models and adds those arguments automatically. I managed to wrangle all of this and make A1111 work, but it was a huge pain in the rear end. It does have a couple of features that the other UI currently lacks, so I guess I'm gonna keep fiddling with it.
|
# ? Nov 11, 2022 21:49 |
|
Hadlock posted:Maybe you could copyright your entire style[...]. Like maaaaybe I could see an artist trademarking their style and then trying to sue the producer of an AI, arguing that it's confusing trade dress. That's normally to prevent confusion by not allowing one manufacturer to package goods in a way that would lead consumers to believe that it's a different manufacturer's product. But some software producers have successfully won lawsuits based on this due to (the court decided) confusing similarities between the "look and feel" of their product(s) and the infringing product(s). That all said, I think it's a pretty big stretch to argue that you're putting stuff out in the marketplace/on the internet/whatever and that grants you protection against anyone looking at what you're doing, learning from it, and then producing similar stuff. And, I mean, we've collectively wandered a long way away from what copyright was originally intended to accomplish, but that thing right there—enrichment of the public domain by the production of a lot of works—is much closer to it than the "make the most money for the copyright holders for as long as possible" thing it's become. I mean, I totally get the idea of human artists wanting some protection against this sort of thing, same as every other worker wanting some protection from being replaced by automation. I'm absolutely down for that in the abstract. I just don't think copyright (or other existing intellectual property protections) are the tool to get there.
|
# ? Nov 11, 2022 21:51 |
|
Hadlock posted:Can you do "democracy if it were a BLT sandwich" i wasn't able to get BLT sandwhich to be democracy, it just doesn't have any interest "democracy if it were a BLT sandwich" hotdog however has some very american ideas "democracy if it were a hotdog sandwich" but most interesting of all, Orange has some strong feelings about the working class! "democracy if it were a orange" bonus content: "the smurfs waging war" "superman vs ryu" "teletubbies vs carnage" "mario vs sonic" "wario optical illusion" "Junji ito drawn by junji ito drawn by john burgerman drawn by junji ito" "a machine gun made by gnomes" "warning bells ringing through the multiverse" "your boss"
|
# ? Nov 11, 2022 22:35 |
|
Rutibex posted:i wasn't able to get BLT sandwhich to be democracy, it just doesn't have any interest these are loving incredible. real mad magazine vibes.
|
# ? Nov 11, 2022 22:54 |
|
Megazver posted:Auto1111 sucks to install but I must have missed part of the prerequisites, I'm just not sure what and don't wanna keep trying poo poo that doesn't work. I did find this - https://camenduru.itch.io/stable-diffusion-webui - which I was able to just unzip and run, works great but it's a month old now and I don't know how to update it without breaking it.
|
# ? Nov 11, 2022 23:02 |
|
Rutibex posted:but most interesting of all, Orange has some strong feelings about the working class! Pretty interesting to see the difference between BLT and Hotdog. I bet hotdogs have a lot more pop culture references than the lowly BLT All great images Can you do the above image in img->img with prompt "Bernie Sanders as statue of liberty, holding an orange"
|
# ? Nov 12, 2022 01:01 |
|
Hadlock posted:Pretty interesting to see the difference between BLT and Hotdog. I bet hotdogs have a lot more pop culture references than the lowly BLT
|
# ? Nov 12, 2022 01:27 |
|
SubG posted:That all said, I think it's a pretty big stretch to argue that you're putting stuff out in the marketplace/on the internet/whatever and that grants you protection against anyone looking at what you're doing, learning from it, and then producing similar stuff. And, I mean, we've collectively wandered a long way away from what copyright was originally intended to accomplish, but that thing right there—enrichment of the public domain by the production of a lot of works—is much closer to it than the "make the most money for the copyright holders for as long as possible" thing it's become. The Monkey's Paw Curls Steps: 32, Sampler: Euler a, CFG scale: 7.5, Seed: 1220674975, Size: 512x512, Model hash: 4470c325 https://ipwatchdog.com/2022/11/01/us-copyright-office-backtracks-registration-partially-ai-generated-work/ I've been following the case of "Zarya Of The Dawn' because I don't know the legal status of the images I generate on my own home computer. It seems the registration was incomplete and that by all means when fixed should go through and be granted, but now I have to pause everything I was thinking of doing while I track this all down. How silly of me to think "Your work is under copyright protection the moment it is created and fixed in a tangible form that it is perceptible either directly or with the aid of a machine or device." meant I could actually own anything at all. Rutibex: "Throng" is a power word, add that to your spellbook.
|
# ? Nov 12, 2022 04:14 |
|
User was banned for this post. Illustrated by Greg Rutkowski, Fantasy, Concept Art, Splash Art, Hyperdetailed, Trending on ArtStation, Intricately detailed, Dynamic lighting, triadic colors, volumetric lighting, 8k resolution. I guess the weird red drum by her crotch is supposed to be goatse? MechanicalTomPetty fucked around with this message at 04:20 on Nov 12, 2022 |
# ? Nov 12, 2022 04:18 |
|
I believe that is the physical manifestation of the ban hammer, for the mods who have trouble finding it
|
# ? Nov 12, 2022 04:23 |
|
KwegiboHB posted:
You might not like it but this is a very good outcome. You absolutely don't want judges applying broad copyright protections to AI generated art. We don't want to accidentally grant the company that owns the AI training weights copyright to every single thing it spits out. If broad copyright protections are going to exist they will favor big corporations, not individuals. quote:Rutibex: "Throng" is a power word, add that to your spellbook.
|
# ? Nov 12, 2022 11:01 |
|
I have discovered a very rich vein of RNG. If you ask Midjourney v4 for "a diagram with many different X" it will spit out a huge variety of examples. This is a productivity hack as it lets you generate hundreds of examples instead of 4 at a time. You can then isolate the ones you like and feed them back into MJ v4 to make them nice Its really quite amazing how these things are presented. The "diagrams" seem to be context sensitive. So a diagram with Rings will have the rings arranged into a ring. The most interesting images come from "a diagram with many different people". This produces very unique results every time: Some objects are better than others. Shoes, cars, crystals, and boats are extremely good and the AI makes a lot of coherent designs: With some items like "potions, magic items, coins, flags or guns" MJ v4 goes a bit overboard with creativity and you get very odd designs. But there is still a lot of usable coherent stuff.I love the weird terrible guns For more abstract things like eyes or occult symbols its very good indeed! You can generate bulk waifus Quite good at creating fantasy monsters and new D&D races!
|
# ? Nov 12, 2022 17:11 |
|
Rutibex, holy shiiiit nice prompt! In other news https://twitter.com/DeviantArt/status/1591239714098806785?t=agSi8mE_t0V5qq969X0-NQ&s=19 I'm still really unclear about whether all DA art was and will still be used for their private checkpoint because they keep saying 'third parties' and even then ???
|
# ? Nov 12, 2022 17:37 |
|
I made some fun portraits of music icons. Uh, sorry Dolly. That's what the AI kept giving me, even emphasizing "close-up".
|
# ? Nov 12, 2022 22:28 |
|
Finally got this installed and it's pretty fun! "dieselpunk [your building here] with oil pipes and trees with glowing reflections on water, cel shaded, intricately detailed, thick black sharp lines, 2d, sharp focus, cinematic, moebius!!!, ghibli!!"
|
# ? Nov 12, 2022 22:53 |
|
New embedding training method: "dreamartist" is supposed to work to train effective embeddings using just a single source image https://github.com/7eu7d7/DreamArtist-stable-diffusion https://github.com/7eu7d7/DreamArtist-sd-webui-extension Sadly I don't have enough VRAM to train embeddings so I can't try it out but it sounds pretty cool
|
# ? Nov 12, 2022 23:58 |
|
man, ai art news are a rollercoaster
|
# ? Nov 13, 2022 00:07 |
|
RPATDO_LAMD posted:New embedding training method: "dreamartist" is supposed to work to train effective embeddings using just a single source image Wow, that looks pretty impressive, how much vram do they expect me to have to give this a try? The git repo just mentions that the reconstruction feature can be turned off on setups with lower vram, but they don't offer any numbers.
|
# ? Nov 13, 2022 00:41 |
|
It probably uses the same VRAM as "normal" textual inversion / embedding training. The internet says that requires a minimum of about 12gb, which is like a 3080 or better. Although you could also do it on a paid service like colab or runpod for a few dollars. RPATDO_LAMD fucked around with this message at 00:59 on Nov 13, 2022 |
# ? Nov 13, 2022 00:54 |
|
An example of how easy it is to guide MJ v4 with a basic lineart drawing: prompt: https://i.imgur.com/LkKk29h.png ruined library on a haunted spacestation --v 4 output: I upscaled two of those, then remixed one, inserting the other in as the image prompt in this prompt: https://i.imgur.com/KX8Qgy4.png Eldritch-haunted space station. Vivid psychedelic haunting. Astral horrors. --v 4 And I ended up with these: I could fix the bright white spot by adding color to the initial sketch (should probably be done anyway, black lines on white gets interpreted as pencil sketches or whatever quite often) but they are both pretty much iterated versions of my initial sketch. deep dish peat moss fucked around with this message at 02:48 on Nov 13, 2022 |
# ? Nov 13, 2022 01:15 |
|
The best part is you really do not even need to know how to draw. You could half-rear end mspaint something and get results like this. So you can render out your kids' drawings or whatever: By Studio Ghibli: as a Gritty Graphic Novel:
|
# ? Nov 13, 2022 02:49 |
|
deep dish peat moss posted:And I ended up with these: Reminds me of that one building in Rome ... Pantheon. It's perfect deep dish peat moss posted:as a Gritty Graphic Novel: Amazing
|
# ? Nov 13, 2022 02:59 |
|
WhiteHowler posted:Reposting this, maybe you're making it harder than it needs to be? I went through all these instructions which are way the gently caress more erudite than they sound for someone with zero python (or code outside of web dev stuff) experience. There was no option in the Python installer related to windows environment path or anything like that, so I'm assuming that means just installing it in the /windows/ directory? But anyway when I try to launch it I get an error about being unable to install Torch because the referenced version doesn't exist: code:
I went to pytorch.org and tried to install it through the python console using their instructions but I think I need some kind of additional installer package? Because things like CONDA and PIP are not recognized syntax in the version of Python I just installed from the link in the quoted post. deep dish peat moss fucked around with this message at 04:43 on Nov 13, 2022 |
# ? Nov 13, 2022 04:41 |
|
deep dish peat moss posted:There was no option in the Python installer related to windows environment path or anything like that, so I'm assuming that means just installing it in the /windows/ directory? No! Do not!
|
# ? Nov 13, 2022 04:47 |
|
But it specifically says to in that post
|
# ? Nov 13, 2022 04:48 |
|
deep dish peat moss posted:But it specifically says to in that post when you install python, there is a checkbox that says 'Add Python to PATH' check that checkbox do not install it to your windows directory
|
# ? Nov 13, 2022 04:49 |
|
Okay but when I check that and install Python NOT in the windows directory, trying to launch SD immediately gives an error that says "No Python at '"C:\windows\python\python.exe'" And when Python is installed to the Windows directory it gives the error in my other post!
|
# ? Nov 13, 2022 04:53 |
|
https://superuser.com/questions/143119/how-do-i-add-python-to-the-windows-path PATH is a system environment variable Basically it is a list of folders for windows to look for .exe files in. Whenever a program says "hey windows please run python.exe", Windows checks every folder listed in the PATH variable to see if it can find python.exe, and either runs the first one it finds or errors out if it can't find it. You have to add the python install folder to that list (make sure to just add it and not replace the whole list)
|
# ? Nov 13, 2022 04:55 |
|
|
# ? May 27, 2024 05:45 |
|
RPATDO_LAMD posted:https://superuser.com/questions/143119/how-do-i-add-python-to-the-windows-path This has been set up correctly since the beginning, it appears that part of the problem is that the version of SD WebUI linked in the post I quoted tries to run pip.exe from its own /scripts/ directory, not the one in the Python directory. And that does not include pip.exe However SD (the version linked in that post) does not run without Python installed in /windows/ and I don't know what the deal is with that? I can probably fix it by sticking pip.exe in stablediffusion's local /scripts/ directory since that's where it's trying to run it from but it would need to be in the windows directory to do that and why shouldn't it be there?
|
# ? Nov 13, 2022 05:05 |