#💬|general-chat
1 messages · Page 14 of 1
AI doesn't keep the original image at all. it was shown it during training
if a legal argument is to be made, it has to consider how the AI actually collects data from the image. The image isn't saved to be redistributed and resold. thats the thing
It's sourcing off the style.
if the argument is "well all the generations look sort of like my art!" then it might not hold up
the issue I see is taht artists will stop sharing their art publicly at all, out of fear of it being used for AI training
Some will some won't.
the moment that style is protected under copyright, Disney owns all style and that becomes the world we live in. They have been preparing for that switch if it ever comes. They're ready believe me
that's the logical end result if it keeps going the way it's going, and the models will either stagnate or iterate off themselves without novel styles being added into the mix
what command line arguments can I give it to reduce memory usages
just when we finally got to a point where art was accessible to people
Artists shouldn't have been sharing their art on instagram or free services like deviant art etc... this whole time
those services have been profting off their works for decades
this would protect megacorps far more than individual artists, in my view.
artists should've always been self hosting, and now they are smashed with a big fat new reasson why. Corporations aren't the only problem now. Being blind to the elephant is easy. When it's democratized and everyone has access to stealing your style, suddenly it's real
there are plenty of artists who are ok with being in the datasets though. the models aren't going to stagnate. esp with potential monetization/royalties being a thing.
this is a different discussion, however, and even though you're probably right, in this context it's a strawman argument that's neither here nor there. The ethics of image distro services are another animal entirely
That's a supposition not a fact. It's moving so fast we don't know where it's going. They are working on zero shot learning as an example.
The trend to share everything and post all content on some social network only started at web 2.0. The old web functioned fine. it'll function fine again
I think at this stage of the internet, and knowing how data moves and what we do about how data is used, ethically this should have been a much earlier step in it all and certain AI companies have skirted the edge on that and dodged it which I think really does border on unethical ebcause they almost assuredly knew that that was a viable option
I mean it wouldn't be hard to track an artist's style and follow generations made based on that style, usage of their name as a keyword, etc, and toss them royalties as a result or at least have a system in place for it and offer that up as some sort of agreement to artists
The data was trained fair use terms so there is not ethical issues.
i sort of see ppl training models or hypernetworks on a certain artist's style the same way i see music/game piracy. if no one wants to pirate your album or your game, then that's a very bad sign because it means no one knows about it and/or likes it. if someone wants to train a model on your work as an artist, you have probably already made it.
stable diffusion was initially trained by researchers though so it wasn't necessary to acquire permission .. they probably shoulda put opt outs up before retrianing 2.0 though
@past vortex except that most people don't use "one artist" as inspiration, they use several to get a unique style
this sort of thing is already in play for photos,and even using partof someone's photograph or staging a photo in direct imitation of a well known photo can get you slammed for copyright infringement
same for music
and so all of them should get a penny
plugging Greg Rutowski into a prompt is a beginner step, soon you find more obscure artists, blend styles, perhaps photoshop a few things, and the end result is very indistinguishable from any one persons style
speaking as an academic, this isn't actually true though
you totally need o get IRB approval to use copyrighted work for stuff and get permission from the owner in most fields
text and data mining exceptions say its fine to use copyrighted material to train ai for research purposes in the EU
that no one is policing it isn't an excuse to not actually do your due diligence
It's been a long standing practice in the art community that artist copy other artist styles. If you're not familiary with the concept "after" in the art world I would encourage you to look it up. This has been going on for 1000s of years.
in the EU that's fine then
yes, stability is in the EU
@past vortex btw, thank you for this discussion, rational points are appreciated, though the common blood-lust is apprehensible
I don't have much blood lust
wait I think I used the wrong word, hehe. apprehensible....
I mean I use stability in my own way, I draw out my own art and have trained models based on Pexels photos which I then photoshop via filters
then run teh art throught he AI to fill in all the extra details via inpainting
takes time but it's fun to see how it comes out
I applaud you for doing that.
I use it for references/inspiration mostly. Good for concepting.
I trained my own model, but it doesnt really follow prompts, seems to just churn out random images in the style of the instance images I used to train it
any idea what could be wrong?
plus I got some inspiration and have gradually been getting better with digital art since it's a thing I never learned and I figure why not, I got an iPad and Procreate
i do wish someone cooked a way to use Procreate like that PS plugin
like if your ipad is networked to a desktop running SD, you could feed stuff back and forthon the fly
the Gdrive fu I have to do is tedious
Disney, and this is no joke, or meme, is pure evil. They own so much and want to own it all.
or Iguess I could setup a NAS
and then pull from that,but still, an in-app plugin would be super rad
@past vortex yea the entire IOS workflow is probably just an Apple after-thought, getting data in/out is always a pain
yup
as a PC person for 30+ years it's misery
i only got the ipad for this and so my kids can facetime my tech inept parents heheh
If you stay within the constraints IOS probably isn't that bad. I guess..
yeah but an m1 ipad is slow and terrible at SD compared to my 3080 Ti
right now I'm at a point in the discussion where we're going over the vagueness of most online TOS which give others permission to use your data
waiting on 4080 to be better supported and then i can say i'm not slower than a 3080ti xD
I think uploading something to the internet is like placing it in the middle of a busy street, and expecting someone else to not pick it up and run with it is foolish
i made a video you might enjoy
shared if a couple of places already, this is exactly what it's about
Use some copyright music in a youtube video, see how that works out for you
it's all on there on youtube on other channels, after all
40 series doesnt work? I was thinking of upgrading from a 3080
@past vortex If I used someone elses entire song for my video, I would expect consequences, but a few seconds? There is more than enough free music to do that
there should be some degree of protection and privacy at baseline for both you and stuff you make or stuff you post online. It's no different from sharing a facebook photo privately with someone and then finding out it was sold to some tavel company and they photoshopped sunglasses into it and are using it in an ad campaign without your permission
it does but the gens are slower than it should be and the training doesn't seem to work as well as i hoped it should. only black images etc..
some manual updates have improved things but i dont think it's fully supported yet
That's the thing. The online world is not your "friend" and even those friends are the ones sharing what you privately shared
are you using a weird VAE?
I've found that happen when you're using a mismatched VAE
default
oh ok. I will wait a while then
oh weird
someone several layers removed from that private share has no idea what restrictions you had originally
sure, and someone taking a photo of your art in a gallery and posting it online... what about then?
many art galleries say no photography inside for that reason
sure, and yet it still happens
next let's say it's a pic of you naked through your window because they happened to catch you walking out of the shower.... they posted it online now, guess it's free to be used in whatever
there's clearly a line between all teh examples given, where that line is, exactly, a group of people can't even really decide on
I can only imagine what I might feel like if "my" art was done in that way. I could care less, I sell huge murals, your 4K image of it has nothing to do with me
so it's insane to think a computer will
Are you familiar with the art concept "after?" For example if someone makes a piece of art in the the style of another artist it should be cited as "artist A after artist B." This is one of the many things I learned after I started using AI.
there's a famous canadian painter named Robert Bateman. i think it was the 70s or 80s, he started selling prints of his paintings instead of actual paintings. The art community was enraged and he was outcasted and hated by most of the community for years after. He built a new community of print artists and even saw that explode when digital art became a thing. A pioneer of their digital art and print media field really. Andy Warhol caught tons of hate for selling his paintings as prints too. Funny to think at one time, printing art was outrageous
I make "music" that I release for free. If someone took one of my instrumentals, put lyrics over it, didn't credit me, I know I put it online to inspire others.
However, if they were making a bunch of money from it, I'd still be like "Aight I'm gonna make some more music, that track banged"
If someone took my album and did the same, I'd think "Time to upgrade styles, they're on to me!"
i had not heard of this actually, that's interesting
I don't eat from my Art. I hope to never have to make art for money, I like supporting artists in their technical works, you can't take away the need for technical support
What’s difference between a model and embedding
Yep it's been going on since at least the middle ages.
Guys, quick question:
Is there a AI model for making the MLP Equestria Girls style?
(And don't you dare question why)
embeddings sort of like a super focused prompt. they work with loaded models to generate images towards a particular result
Every SD image I've made I've shared the entire prompt, no use hiding that stuff, the actual art is my imagination in coming up with the prompt, which others aren't necessarily going to be able to replicate
cool thing about them is you can blend their results together in one prompt, or use them on children of the model they were trained on
Another term I learned about was copyfraud.
copyleft is another fun one
that is clearly unethical. release with constraints is different from "taken in obscure circumstances where privacy was expected". The problem is that online TOS don't guarantee privacy. They "allude" to it.
for legal purposes, so that they can use your information in anyway that THEY choose
Actually it's probably criminal for the person to take the picture..
right, and the point illustrated was escalating levels of clarity of inappropriate use
to us.
I think it falls under the category of revenge porn.
but reality isn't that clear cut and there's gray areas throughout, and even a group of people aren't going to agree on where the limit is of "well that needs to be protected" vs "yee haw, this is the wild west, it's out there so it's gone, boyo!"
Style is style and it belongs to everyone not corporations or individuals.
@past vortex hehehe. I was recently thinking of the term "Wild West" and... man those poor indigenous folks were wild back then
clearly that example is a case of protetion. The pic of your art in a gallery, maybe taken illicitly... maybe also clear cut, but others might disagree
etc etc
at some point you need to make some protections, and ultimately it's better to be a little over protected on your privacy and your personal rights than not
otherwise corporations will stomp all over us
There are laws on the book specifically about the use of deefake/ai porn in certain states and the prohibition there doesn't base it on copyright
even more than they already do
correct
i don't really understand, absent overfitting, why anyone would need their art protected from being looked at by the ai. but i'm pro opt-in anyway because artists will be totally alienated otherwise.
which is why better laws nee to be written and there needs to be some level of quorum over where to draw the line on things
We already have protections. No need to give corporations another tool to suppress the artist.
I agree. I don't think SD necessarily needs to be trained on artists that don't consent. However I don't think that as it is now is "taking food from their mouths".
Art progresses, if one style is played out, you upgrade your style
If someone fed my entire album into an AI, and made 10 similar albums, well guess what, next year I'm learning new techniques and producing something more fit to current tastes
because ultimately yeah, corporations will use every excuse to erode our rights to privacy, ownership of our own personal data, ownership of our own generated content, all of it
they would prefer if we were slaves and they didn't ahve to pay us a cent
and could freely take whatever they wanted, whenever
I don't understand you statement. We already have a very clear definition of copyright. Style is style and it should remain in the public domain.
because corporations are all inherently pieces of shit
@past vortex which is exactly why this tech needs to stay open source
yeah I agree there
the boot will come down if you are underneath it, or not
corporatizing it is just another way to scam
BTW shout out to all the Anti-AI people lurking, know that we love you, truly, I'm on your discords lurking too
the problem with any alliance or coalition that involves corporations at the table is that by proxy it loses any ethical credibility because of that presence, so you need some level of non-corporate, ethical legislation, which is never gonna happen because people are terrible and people in groups are worse
"I have aibro footrests and I play my pipe organ and smoke cigars rolled from pure baby seal skins" - though sometimes I'm not very pleased with what I see
"their parents will keep them fed" ""AI" is going to lead to a deeply boring and disconnected world." "Watch mental health problems and suicides skyrocket because of this"
This has... very little to do with protecting artists. I'm sayin.
"AI kills creativity" only if you didn't have much in the first place
I'm still struggling with getting the images I imagine into digital pieces that are true to the visual and idea. There's no quick prompt for that
yes surely more people with the ability to share their imaginations easily with others is going to lead to a more boring world.
It lowers the bar to make something pretty by a couple orders of magnitude
@little plaza some people shouldn't have imaginations... or something, but I'm not gonna be the one to decide who, and maybe nobody should
So the friction from your mind's eye to real output is very much reduced
"some people shouldn't have extended vocabularies"
it also greatly improves your ability as an artist to produce the things you want to produce in a more timely manner. we got limited time on this earth. and also you don't have to use it.
The fun part is when you kitbash a generated library of images into what you want
exactly. nobody wants to look exactly like someone else.
so far what it's best at for me is coming up with things i didn't even think or know i wanted to try before, by accident.
it does currently seem a lot harder to get it to do exactly what i want if i have a specific image in mind
sooner or later we'll get semantic segmentation to image, which will be great for that kind of stuff
yea they also had one for landscapes previously
I have an idea for it using sd, but it needs more vram than I have so I have to reformat the code and port to runpod/colab/aws if I'm feeling expensive
@uneven wasp what is your idea?
It basically boils down to parallel inpainting then masked combination during each step.
Had some success with applying image properties in final prediction space earlier today, might use that technique of the naive approach of combining during step output doesn’t work
Going on a bit of a tangent but essentially you can rewrite DDIM to be entirely terms of the last step and final prediction, so you can use the same math DDIM uses to predict final output, apply your properties (in my case reflection) and plug into the next step equation.
bros watched the new ntts video
Sup guys
@uneven wasp I'm trying to see what you're saying, what would be the use? I guess you're in the code, I'm still conditioned by using UI's
Yea I might have got into the prospective math a little too much. Essentially you have a set of prompts and a segmentation mask (same simple color image you see in the nvidia project), so you run the inpaint model for each prompt, and the colored mask determines which prompt’s model to use for each pixel/part of the image.
I wish there were some way to easily train TI so we could give it some images and tell it I don't want you to learn the person but learn everything around the person.
@uneven wasp so combining multiple prompts to a previous image simultaneously and combining everything based on masks?
Simplest explanation and first thing I’m going to try: Combining multiple prompts based on masks after each step
Kind of reminds me of the "multi-subject-render" plugin, haven't tried it out thoug
You’d probably need to train it in a situation where all the other variables code for the person so the only thing left for the new token would be to describe everything else
Smart initialization of the token might also help
Supposedly that is what filewords is for but I am not having any success.
Yea sometimes a theme is super strong and permeates everything. Maybe collecting images of what you dont want and training TI in the negative would help as well
How do we train in the neg?
I've used the bad_prompt_version2 embedding, it's trained on bad images to remove those effects when placed in the negative
Just looked into it. Cool idea, but that combines them at the end based on depth inference rather than each step like I’m suggesting.
In theory it’s perfectly possible to backprop through the negative prompt, but I don’t know how it’s done with the default tools.
I don't think it can as we didn't get the full implementation.
if I have person as a fileword will it ignore a man and a woman?
I know when I said man it didn't ignore it it seem to promote it.
I’d have to look through the TI code for sure, but perhaps swapping the coefficients of the conditional and unconditional passes so they function opposite would work
Objective function might have to be modified to include classifier free guidance in that case though
Like with webui? You could probably find the code where it adds the prompt boxes to the webui/gradio app and set a default there
Hey guys can someone help how to start with StableDIffusion please ? Can dm , i try to find an info for so many days , but hardly can understand what'sup !Would be very greatfull ! ❤️
@lucid spire try #🤝|tech-support
How do I get 2.x to understand "long wide camera shot"? 1.5 understood.
Hey guys! Is there a way to message the bot directly with the dream commands instead of using the dreambot channels?
is it possible to get the model name and prompt of this image? https://media.discordapp.net/attachments/1018222862924005537/1055586239010582558/legion_8221.png?width=1603&height=1068
someone told me that if you train a set on 512x512, then later train it on 768x768, you'll have bad results
true?
can that even be done?
@tame tusk if you have the original PNG some of that information may be encoded within it
@tame tusk if you just open the PNG in a text editor that information is at the top
Just figured out about this, how do start generating things?
@meager kayak depends on how you want to do it, either locally, or using a cloud service?
try dreamstudio website
thanks,it's posted in the winner gallery channel and I dont think it has any info
@tame tusk yea any conversion will likely strip that information out.
but how to verify that piece is done by ai and done by stable diffusion if the original info is not kept, I can simply photoshop an image and claim it as done by ai
hehe, well thats a new concept, people are usually doing it the other way around
You would need to contact the original author. If you are a part of the service collecting the images, the original PNG might be somewhere on the server
well, since it's tagged officially as the winner image, I think the prompt and model should be disclosed as well, as part of rule of the game
guys which webUI is better ?? I have automatic 11 stable diffusion and it seems to suck
InvokeAI is another popular one, some additional features but overall less functionality. There are half a dozen, or more, you should try them all out of automatic is lacking for you
@tame tusk running CLIP on the image gives a bit of insight, you could then browse e.g. civitai to find a similar model style, or maybe artists to see which have that style.
what less functionatlity are you talking about ??
Invoke doesn't work with 2.x yet, and doesn't have extensions, among other things. It does have a cleaner UI though.
I dont use 2.x so its not problem for me. Thanks i will check that
Their youtube also has very good videos on workflows. And they have their own discord.
Hi Guys, Just discovered stable diffusion and got so confused but it looks amazing.
I'm 30 and in a wheelchair looking to occupy myself and been messing around with AIs like chat ones and what not. So came across Stable Diffusion and went "huh this looks neat", so here I am 😄
So how can I actually use it and does anyone have any cool tips ❤️
@keen hemlock welcome, what kind of computer do you have? You can run it locally, or on a number of cloud services
Just a general basic gaming computer it ain't much but it works 🙂
@keen hemlock depending on your technical acumen, using something like Automatic 1111 or InvokeAI is the recommended choice
So, on the website, i have been running into some kind of a filter with a simple prompt saying "young blonde girl, concept art, realistc, 3D render". Is there a description on what is not allowed so i can avoid triggering the filter unintendedly?
@velvet ferry don't use girl. try woman
oh, okay
i kinda get it, but also, it feels a little weird that such a simple word is filtered
Is it though? Is it... 😄
I assume the AI doesn't want to fall into the wrong hands and actually has a moral code.
nah, it worked, i was just musing...
yea local installs have no such restrictions. online, they have to use fairly simplistic means of staying legal
to the point that fairly innocuous generations will get flagged.
yeah, and i'd rather have it this way than get flooded with crap as well
it's just hard sometimes to figure out where you went wrong, anyway, thanks for the help
You're can also just use our beta.dreamstudio.ai 🙂 stay tuned btw...
i do use that, but thanks 🙂
is it possible to take a picture of a char and changes its pose ??
happy new year guys, so what's coming for SD?
Hey! Welcome to the chat! 👋
That's awesome!
Illustration is so fun
@bleak matrix Morning? Mid-day for me hehe
5AM here, I just keep night "schedules"
NIGHT SHIFT, UP TOP
👏
What I would do if I didn't have five thousand doctor appointments 😂
heh. My doctor told me to come in, but then I told myself later. He, is me, after all
Also, here's my newcomer guide: https://docs.google.com/document/d/1aHJ9RBt_vlCwJQBVUUsb7VghKB-wynv7WGTxm9ozL1k/edit
FRIENDLY STABLE DIFFUSION GUIDE by Atypical Consortium / Sunny LAST UPDATED: 12/30/2022 Please note this document is a work in progress! Thanks for your patience in this matter! PLEASE ALSO NOTE I DO NOT OFFICIALLY SPEAK FOR STABILITY! THIS IS JUST ME, MYSELF, & I! Hello, and welcome to Sta...
Bahahaha~
Look in the mirror and debate your assessment, eh?
"Ppppssst, you haven't had to the entire time, despite"
first typoo I noticed, looks very good and thorough, I'll try and recommend it if someone is new
I'm sure there's probably a lot
but where?
lmao
I don't think I've even proofread it, tbh
since it's a wip
some sections I simply haven't, since I'm writing them
Ok. I have some guides I've done to remember various things, I can skim through it some more and see
hiho 🙂
couldn't find if there is separate chat for auto111 webui. is it?
good morning!
And no, there isn't
If you're looking for support, though, you can check out #🤝|tech-support
thanks a lot 🙂
Can someone help me into how do i expand an existing image that I've using this??
Please
np!
You mean with the bot?
Yes
The info is in the guide I posted up there
You can also check out #1034602544263090268
Cannot find any pinned messages of you in that chat
I'm having a bit of trouble understanding something. When I generate a batch of 10 images, if I like one of those images how do I get it and upscale it to a bigger resolution without changing it?
use external upscaler
like the stabilityAI upscaler - https://huggingface.co/spaces/stabilityai/stable-diffusion-upscaler ?
or you mean something else entirely? any recommendations?
nah, this is diffusion upscaler.
i meant regular ones like ESRGAN, SwinIR_4x or something similar which upscaling image with attempt to preserve details
it's up a few lines
just scroll up
the friendly guide
I just posted it
Got it got it thanks
what is better to run on my site as someone who is getting into AI now, Automatic1111 or stablediffusion 2.1 hugging face?
Cartoony art is so much more fun to generate than detailed art imo
doctor who?
@leaden sleet exactly!
just this dude who lives in a tent and makes weird concoctions. he's helped me out pretty well before, though
sick, i thought you meant you
well. the tent is a metaphor.
so like potions of ringers?
nah he's just this old shaman, his tribe does pretty well though
cool, chainhealing is solid, but make sure you downrank it
better healing/mana ratio
always out of mana. many potions stocked, just have to run back to base...
where does stable diffusion store the images u make? @forest apex
[webui]\outputs
are u sure? i went to outputs folder but its empty
@sage sky there should be five subdirectories there..
but, I'm assuming you're using automatic 1111?
People seem to suggest generating images at 512x512 and then upscaling them, but today I have been trying to make images of a shopping mall being bombed, with 20+ shoppers panicking, and it seems I can get more informational details when I generate at 1600x1024.
So, my question is, am I doing the right thing by generating them at a higher resolution? Could I have been biased?
you must work for CNN or something
I was also getting good results with 1024 x 1024 resolution. It all depends on what are you generating
Lol making a visual novel
512x512 just doesn't have the pixels to allow the graphical information to be crammed in
anyone?
and how are sites making the style presets?
@pallid sable by hugging face do you mean the vanilla sd-webgui? huggingface is a cloud provider where you run the software on their machines
I am comparing Automatic1111 https://github.com/AUTOMATIC1111/stable-diffusion-webui and https://huggingface.co/stabilityai/stable-diffusion-2-1
as for styles, it could be any number of things, from filters, to alternate models/embeddings, to appending artists or other beautifying keywords
i see, so an invisible pre enterd prompt addition
both are the same, the latter link is for downloading the models the webui utilizes
yes, that is mainly why MidJourney has such nice images, because they do things behind the scenes so people don't get bad images
yes, they emnphasize on photorealism for example
skilled prompt engineers and photo editors can surpass MJ's output easily.
people have a misconception that its a 1 button masterpiece generator, but it takes skill to learn prompts and also what topics work with what style
for example in Dutch Golden age buildings, boats and portraits work well becasue that is a part of the source
MJ is specialized on its own style, SD is way more comprehensive but also requires more experimenting/learning
Also im collecting 27K+ images for a very good sci fi diffusion
and will probably hire a gpu
since i dont want to run my pc for SD for 1 month 😅
@forest apex Between Dreambooth and normal finetuning, which should i prefer for a project of this size?
I will hire a GPU too so vram shouldn't be an issue
I'm just beginning with training, I would ask in #🔧|finetune
I shared A.I tools like chat GPT and text to image with my FB group guitar builders community and got horrible replies, its just a buzzword generator, im glad i wont live much longer to live in the computer world you are describing. I was trying to help small builders save time on content creation so they can compete with large companies
not everybody can see how this will transform almost every industry
they are scared or threathened instead of utilising it to your advantage
this year a text prompt can design a guitar, then to blender, then to a CNC router.
the consumer doesnt care about how long it took you to make, they just want a low price
but i do advocate for some standard that it is mentioned when A.I tools are used so "hand made" artists remain properly valued
very few artists are valued, though. not as much as the effort they put in, compared to say, a CEO or manager
sometimes a particular circle isn't very interested. I shared on FB too and got only a few interactions
though I'm mostly connected to musicians and other artists...
i view this tech as a tool, not a replacement
before the router people chisled a pickup hole, 10x more time
the consumer doesnt want to pay 10x more becasue you chisled
so dont blame software developers for innovating
visual artist can use these tools to generate passive income, then buy the paint and canvas and take your time
so it can empower artist instead of crushing them
passion is very expensive
that part. being passionate about what you're producing is not something that money can encourage
you already have the passion, but getting resources to do something with it is expensive without A.I
can i use more the 1 modle in the same time?
@floral zealot you can merge models with varying success, and also change models by script, I.E. using x/y plotting. that or using img2img and switching images manually
oh i see thx
@wise stratus So.... distillled diffusion. "Are we there yet?" 🙂
brain diffusion when please? https://twitter.com/mathemagic1an/status/1597010079810658304
what is the difference between the "1.5 pruned-emaonly" and the "1.5 pruned" checkpoints? I noticed a difference in size but does one produce better results than the other?
same results. i think one is better for training, but that might be different now with better training systems
any1 knows how to fix out of memmory errors i got only 6gb vram
what are the differences between SD 1.4 and 1.5? I've done just a little bit of testing and they seem similar?
i fixed them by buying a new gpu with 16gb, but then i got them on higher resolutions now. wheni had 8gb, it was essential i run a1111's with --medvram switch. you might consider the --lowvram switch
You could also just spin up a cloud VM
how do you run --medvram ??
depends on tool you're using
basically you need to add it to command or into webui.[sh|bat]
u mean like this set COMMANDLINE_ARGS=--medvram
that's exactly right. maybe a space though. = --medvram
it went from taking 30 seconds to 3 minute 😭 to generate image
yeah. it'll do that
when i interogate clip genreally
that shouldn't be a problem, the memory error is from trying to load all of the artists into memory
there should be a setting about leaving the interrogate model in memory, if you're using a1111
dont leave it in memory. it doesn't take too long to load
you just won't get a guessed artist, but it should be able to get the same CLIP generated prompt results
so if i use in prompt by X artist does it still work ??
interogate and generate are two entirely different systems that work independent of each other. the artist name will work in the prompt if the model understands it. that's iffier on 2.x
nearly testing
something else shortly
Why is Protogen v2.2 and 3.4 released at around the same time?
what are these protogen models anyway?
i'm also confused by that and nobody seems to have an explanation in the communities i'm in
Protogens are from VRChat
But
I think there is some confusion between communities
👋
Not realizing there's Protogen and ProtoGen
nobody means the furry community. there are two ai models released recently https://civitai.com/?query=protogen
both from a user named darkstorm2150
believe it or not, but the ai art community has quickly surpassed the furry art community. in both size and output volume. the furry artist is irrelevant in these contexts
it's very forced confusion. you're probably counting this conversation. What was meant is the author of Protogen ai model has released two versions in the same week
I'm not counting this convo, lol.
But hey, thanks for clarifying that's what you meant.
It already was clear here
guys what is interogate deepboru ?? and why doesnt it fill my propmt with every possible prompts on this plant earth
I was responding to Franky; I've seen some AI art with protogens and models made, too. I thought you were replying to that, too. My bad. 🙂
Hello, Im looking for someone experienced with StableDiffusion to teach me how to not only use but actually make decent results with it, please. Im either bad at prompting or settings usage. Please help!! DMs open.
Search for Aitrepreneur on youtube
He's my favorite teacher
And off topic but why oh why did they have to change Dreambooth training so much for 768. A prime example of "don't fix what isn't broken". Sigh....
Like all software tools theres ways to get the most out of it with the right techniques and settings.
Its not a magic art making button.
It actually is when you get it right.
Black magic
Actaully. Correct
In art though its the idea that counts. You have to know what you want to get out of it.
Very few differences in my opinion, either model works well
@sinful mango did you post your video on reddit? Oh, I see it!
MJ has so many limitations I will never give a cent to them
The point of art is to have creative freedom
Although once I was able to bypass their dumb censors by typing lengerie instead of lingerie lmao.
Theres less and less freedom in the world.
We would already have Unstable Diffusion if it werent for butthurt artists and crybabies.
UD would have been a dream come true. Crowdfunded, democratized AI art. Woah...
Yes the ultra woke and luddites are trying to stop creative freedom..
They already raised an amount greater than the kickstarter goal. Doesn't that mean it will happen? Not going to be instant to train
Mj is great for bootstrapping custom SD models
I dont understand artists who actually try and go against AI. Its like - you cant stop the forest to grow leaves...
It takes a long time to become a skilled artist so it's understandable that some would have a reaction against a tech that gets an 80% quality result in seconds
it soon will be 101% quality result I feel like
Got to look at the whole bigger picture.. This tech can uplift the entire civilization.
Everytime we get a major breakthrough in AI its like Augmenting the human race
nah, the final 20% isn't about speed, it's about taste, experience and artistic vision
MJ is already generating images that have artistic creativity and decisions in them that a normal person would feel proud if he came up with similar stuff
The idea is the key behind creating art. The tools getting way more efficient is a bonus.
having a moral premise and artistic vision isn't going to come out of AI which is fine if all you want to make is celebrities and anime girls
Composition, Lighting, and subject matter are at a very very high level
midjourney is a subscription saas that limits what you can do, when you do it, and how you use it
those are nice but not going to give you an artistic vision and premise
Sd is like having your own concept artist that works super fast
supposedly i heard if you stop paying your midjourney subscription, you can't sell any of the work you created for any reason anymore. so if it gets published , either you pay for the subscription for life, or pay for a recall
limits like that are mental
For an artist that wants to make their art manually SD is a fantastic tool for getting ideas
yeah, SD is great for tests and prototypes, rough ideas
I wonder, if AI gets smart enough to enslave humans. Like taking control of all the money in the world, which is not hard to imagine, and then creating and coming up with new types of jobs in order to give people just enough money so they can sustain their living, paying house mortgage whole their life...
MJ is a nintendo toy pretending to be a pro tool, and SD is an open PC toolset that is fully customizable and tuned towards whatever task you want to use it for.
that's sentient AI. For all we know, that's impossible. It's never been proven or disproven. For now we've got these tools where people use them for their goals. I can see people using AI tools to do such things you're suggesting before machine sentience is achieved. That's why its SO important that Stability among other companies are building open tools that anyone can use. If everyone can use AI to become a tyrant, it makes it harder for any one person to do it
in comes the positronic brain
they are already planning to use AI to make medical decisions based on your medical history which will be connected to your digital ID and health passport
AI already governs the world. We just have an idea of AI that is fostered by popular entertainment.
also if you are likely to have committed a crime like domestic violence if someone in your family checks into the hospital, stuff like that
It doesnt need to become sentient, it just needs enough qubits to actually understand human logic and behaviour in order to make "decisions" upon wether to use its power to control resources in order to keep itself running.
think "pre-crime"
@carmine fulcrum Los Angeles started using that tech several years ago, crime prediction
yep, but this will be everywhere, running 24/7 and collecting data via a mesh network of digital devices like "smartphones"
will?
sounds like the matrix
yes
will is future tense
TV rots the brain 😐
it's in progress now, there is the 21st Century Cures Act and mandatory data sharing
Actually watching TV has been linked to dementia
Both toys
cool thing about tire pressure right? it's important for road safety. So years ago, every single tire was mandated by law to have rfid tags in them. Universally unique id's that can only be read by localized stations and can't be spoofed. So if you've got RFID readers along the roads, you can track any unique set of 4 tires. Pretty neat thing.
SD mechano though
guys but what are the great things that await us in our future? The truly great? Great with a capital?
MJ and SD both have advantages over each other
I used 99% MJ before i cancelled aubscription
Mechano is kind of a pain in the ass if you're unskilled
@astral goblet the only problem is the the transmission of those signals (readable by an SDR) is very intermittent, and alone not enough to track. Coupled with a cellular phone and a few other radios though ...
Im excited to see what MJ V5 will be like and the next SD
I dont know it
individuals with data and an understanding of machine learning will do great things but the government will limit who can access certain data and who can use machine learning on that data
Yeah. Signal processing needs to improve a little before it becomes a serious issue. But still, little things like that. We're already caught in the spiderwebs...
leaveamessageanillcallyoubaaack 🎵
hasn't checked on the kismet discord in a while... they are fascinated by tracking all types of radio signals
the future will be about limiting access to large volumes of data
One person I spoke to actually believes that AI can be restricted and removed
It's very strange
people are strange.
my father is a life long amateur radio guy. ham operator they call themselves. i know a few of those types
Practically none artist is against AI
Especially digital artists
AI =/= AI art softwares and art
there are actually a couple prominent names that are drumming up a lot of noise. but mostly it's not even being noticed yet
the tide is rising higher an higher though
Thats not about AI in general tho
semantics
99% of digital artists have no clue about AI, they've never used it. They will learn it only if required for employment.
Im supporting most parts of protests but am AI " fanatic"
The loudest part of the protest i've noticed is style should be owned, and i will vehemently disagree with the entire movement until they get those people in line internally.
You talk bullshit, all of them use AI
allowing the "style is copyrightable" noise to have any platform at all is dangerous
until that nonsense ends, i would rather not pay them much attention
well, they use AI in the background but they aren't going to use machine learning based stuff like SD unless forced to for employment
sure, but that is invisible to them
machine learned algorithms are everywhere in modern software. that's true. it's a semantic note in this context and not a really great point though.
SD and MJ are "invisible" too
But its misleading
Some ppl do this on purpose
To sell artists as ultracons who are against tech
cons = conservatives
but they aren't going to learn a new machine learning based tool unless it's required for employment, 99% don't care unless it's for work
the crux of the complaint is copyright focused. not the technically correct definition of what artifical intelligence means. all software is a form of artificial intelligence
They for sure arent unfamiliar with AI, especially not the professionals
the people who are interested in AI right now are 0.01% of the artist population and many aren't professionals
But ofc they wont be necessarily AI devs
Majority
But this applies to prompters here too
Many maybe dont even have anything to do with art
Worry more about the people selling the copyrightable styles idea. Then the luddites, who can do no harm because they don't support technology anyways, can be looked at
Before and outside of this tech here
Copyrighting a style is a tough one tbh
Yeah
the tech is going to progress one way or another. the true harm that could come from all of this is disney buying up all styles
it's not tough. it's a stupid idea. it shhouldn't be done
plain and simple really
The endgame of AI generation is gonna be infinite unique content being directly uploaded into your brain. I'm not sure how copyright can even be relevant in that reality
Styles werent my issue and crucial critic on people of this community anyway
uploaded? lol no. Looking at an image is all you need to do to upload an image. Eyes have good bandwidth like that
The fidelity will surely be high, like with video games and VR
maybe not your issue, but taht's a significant part of the anti ai movement that's ongoing. Until the movement itself stamps that idea out and stops giving those people a platform, i can't take the entire movement serious at all. I'll listen to other more reasonable voices on the matter instead
The movement hoists a flag, and people march under that flag. So whose invited and who isn't? Seems to me that the style copyrighters are at the fore front of the parade
you can generate all the AI stuff you want but if you work on a project with other people you will need to work with producers, art directors, etc. and you will still need to edit your stuff based on art direction. That's what is time consuming. You don't just bang something out and everyone accepts it.
a creative process??? no way!
that's what I never hear when people complain about AI tools, pros spend 90% of their time making edits based on art direction
art styles could be copyrighted if they come from a TV show or are recognizable enough in a way that they become iconic in the mainstream zeitgeist (tim burton art style, etc)
if there are more AI images generated, there will be more jobs for people editing those images based on art direction
not even then. only individual creations can be copyrighted. A lot of what has tim burton's name isn't even his style. Another guy created and directed A nightmare before Christmas for example. Henry Selick. And his other movies look "Tim Burtony" too. Disney owns those? naw.
Trademark law is an entirely different thing though. Trademarks can be protected a lot more aggressively, and i honestly think much of those restrictions should be lifted in the name of public liberty
Man, anyone who is dumb enough to say styles should be copyrighted is probably dumb enough to not know how folks could just sue Disney and every major artist in the Western hemisphere
And all of anime to boot
Fair use exists for a reason, people.
That's fair.
Also the people calling for 'copyrighting muh artstyle' often times probably just don't want anything to do with AI art at all, or at most just feel extremely threatened by it.
Perhaps even less than knowing that anybody can make AI art in an instant.
how would you work in a studio if you weren't allowed to "copy an artstyle"?
My personal stance is AI art cheapens art by existing, and they're afraid it'll take away all of the emotional/personal parts of art
Which isn't necessarily the case
Imagine not being able to copy the facial expression of a previous drawing on the show you were working on. (ren and stimpy cough cough)
most of the styles iv'e seen aren't that unique and any talented illustrator could replicate it
digital tools have really improved the talent that artists can achieve easily
AI's kind of advancing on all fronts tbh. Writers are getting nervous about ChatGPT, now I'm seeing voice actors stressing even at the possibility of TTS becoming good.
you EEEEEEEdiot
I ask people the same arguments when they talk about faces being the same.
Well, would you like to put 2 mouths where your eyes are?
if they copyright styles, 99.9999% of AI art critics would be out of work
AI art could potentially dilute the value of creating certain things if others can do exactly that without learning the technical skills to do so. Not just with art or image, but music and video as well.
The truth is, art and many things are just heavily saturated
not out of work. forced to work for some art mill and never own their art, if they want to produce art
I personally wanted to get into the pro art stream like 6 years ago, and the entire market was Fucked at the time
Geez man, imagine the cal-arts style actually being a real copyrighted style...
Musician meltdown coming this year
the vast majority of human creativity has been in the last 50 years. II've never confirmed this but it sounds truthy enough to say it with confidence
please god make this happen and unusable by anyone ever again
CalArts sues every major studio
that's all human creativity and art across 3 millions years, most of it has been recent
RIAA is gonna move reckless, it's in the script
The wages were terrible, art theft everywhere, and the average pro illustrator made just above a fast food job
RIAA v. All AI companies
So understandably, I didn't gun for it
whAt AbOUt tHe ArTiStS
learn to code
TTS is pretty good if you can't do an impression of the character, or just can't really do certain voices due to whatever reasons the person has.
learn to time travel
Well, the real problem is the problems from AI art isn't even art related
I think speech-to-speech is more interesting than TTS personally
It's AI and other stuff
Like TalkNet and what not?
I was being facetious ;>
but I follow you
I was thinking more "AI-powered voice changers"
I personally find that really baffling, since at this point, we're talking copyright, the deeper parts of ethics, romanticism vs enlightenment, that stuff.
I want a future where all AI models are CC0 - No Rights Reserved
That stuff is pretty cool, imagine being Peter Griffin one second and then Gilbert Gottfried the next.
@mental nest try #1025467151206854736
this but with the Corporate Memphis style
Yes, Facebook can sue them too
Believe it or not, I am pretty sure I had this discussion on most angles
I wish we could just talk about the real underlying problems, not working with faulty assumptions and bad info
all this stuff is cool, but when a walking talking thinking robot arrives, I'll be genuinely surprised
Well, depends on things, I guess
kinda want to train an embed on it now, just to make it die
I think there's one already
Robotics feels pretty stagnant right now, AI is outpacing it by a lot
the real world interface is the hardest problem for ai
haven't seen it yet, can't find it on civit at least
We just need Neuralink®️-powered VR, then there's no more issue
I would love to have a good faith discussion on AI, just don't make it AI art related since most of the folks whenever they bring up that crap is repeating the same stuff without doing any research. That includes art friends.
neuralink just reads signals from a small patch of neurons and relays those to a device. it doesn't put stuff back into the brain. its one direction
I'm not referring to actual neuralink, I'm referring to meme neuralink
lol inb4 we're all the just interfaces for an all powerful ai only taking personalized orders from it all day
Beta testing and brain surgery in the same sentence isn't gonna be the most appealing proposition
currently there's no working direct sensory input options for devices
thats nervegear. sword art online
eh people sign up for medical tests all the time.
We're all in some 5D uni's guy's The Sims game they forgot to turn off, probably
SAO is that anime where everyone gets stuck in the game. We don't want that.
the neuralink meme thats been going around is just elonmarketing
10% of Musk fans would sign up immediately
Elon marketing is fun to mock
Also features an admin who likes Dark Souls grade games with 0 respawns
It's a realistic experience™️
yeah, because the helmet would use it's tech to fry the brain if they logged out. But that's what you want. a headset that could potentially dump energy into the brain
Yeah, but bad publicity is still publicity
and if anyone does know of research on this, link me
I thought the frying one's brain was purposely designed lol
So people are dead if they die
in secret by the game creator. there were safeties made but his core code had over rides
Surprised the engineers didn't catch that feature
if (Player->getState() == PLAYER_DEAD) Player->IO->GetDevice("nervegear_v1")->InvokeKill();
Like you don't need source code to ask, wth is there capabilities for SENDING ELECTRICITY DIRECTLY TO THE BRAIN
well yeah. that's exactly how full dive would have to work
some sort of energy field would affect the brain
debugging, probably needed as a raw interface for the other I/O stuff (vision, hearing, etc.)
that's usually how dangerous stuff ends up in production code
or some sort of physical conduction
I'm still waiting for their auto captioning model, as its pretty much guaranteed to label subjects better than the current one
On a hardware level, you can do some crazy stuff with a wacked out driver
That being said, that's just really bad QA testing since there's a chance these devices can be hacked and you got the perfect ransomware targets
^ exactly
Just threaten to fry one's brain if you don't pay up
There's already ransomware-hackable sex toys. Turns out anime isn't as unrealistic as we think sometimes.
Security is mostly an afterthought in most product design
man we're already approaching Strange Days times huh
People really just do not care until something bad happens
UD guys have no experience. They released a model preview and it was bad. more money wouldn't have made them able to get better at training faster. They were just hyping up the FUD for their fund raiser. I wouldn't put it past that community to go pretend they were angry artists and act agent provacateur
You heard about that chinese wifi ransomware chastity belt with METAL LOCKS?
yes, I did
China moment
That article was hilarious
whats next smart dildos sending data back to the nsa
Disturbingly enough, there were chastity belts that got ransomewared, and they had no manual override: https://www.google.com/search?q=ransomware+chasity&client=firefox-b-d&ved=0ahUKEwj-uPSKnqz8AhXyBTQIHafvBdkQ4dUDCA4&uact=5&sclient=gws-wiz-serp
That's the article I bought up
Ransomware, keeping you pure since 2020™️
I loved ready player one the book. the movie was okay. The Ready Player Two novel though? ||basically was just a dumb version of sword art online and its like he got worse at writing||
Everyone has to rip off SAO now
Teledildonics were held back up until recently by shitty overly broad patents owned by companies with no interest in doing anything other than suing others for profit
What happened to using a manual lock and key? SMH
I'm gonna train a new large image model from scratch soon-ish
Great example of just how broken the IP system is in the western world these days. Same thing happened with 3D printers
after I build my dataset
In other words, they like to make folks get effed
almost one billion image-text pairs is the goal
soon ™️
You remember a game pointer? Like in any game? That was unironically patented in the 90s
soon = before July of this year
By Atari or some other gaming company
But yeah, patent/copyright is filled with trolls
Reminder that loading screen minigames are still patented
luckily patents have a relatively quick expiration
MINI GAMES DURING LOADING SCREENS. We could've been playing Snake or any simple game in all those 3 minute long loading screens, but acclaim had it patented and never used it!! dicks
patent wore off so we can do it now, but theres no point. no one cares about loading screens anymore
Wtaf
all those years. loading screens could've been cool
not quick enough when dealing with overly broad patents. IMO the government should step in and invalidate patents like that, as they hurt the economy and society as a whole
My inner child will always hate software patents
and i'm also still a child so it carries on
man i bet your hands are going to be so good
They should stop overly broad patents the way they do Anti-SLAPP stuff
make sure you get like, a million hand images
or just a better general dataset
So courts and people don't waste their time over stuff that shouldn't go to court over
I have a few thousand images that could make a better model than F222, but I need a labeling method that works while not being overly tedious
I'll train a new CLIP model on my large dataset, so you could probably use that for captioning
they shouldn't be approved in the first place. patent law really needs to be reformed.
You can trademark a name in a specific trade. If no one else doing makeup is calling their product Kyrie, she could
Reminder that "Apple" is trademarked
Does it include NSFW stuff as well? I find that's important for good SFW bodies
Yeah. You can call yourself Coca-Cola TV manufacturers, just not in the beverage industry
Both by the beatles for Music, and Apple for computers. Apple (com) has to pay the beatles to do itunes
It's not going to be filtered (though there will be special flags for different types of content)
trademarks are only as good as the company is well known, have to be continuously earned (use it or lose it), and only apply in certain contexts
Trademarks are generally OK
I have serious issues with copyright and patents as they currently are
i like trademarks but a lot of fair use extension could be done there
It just doesn't gave to be in the same industry, which it could understandably be confused for another company
awesome, I prefer it when the filtering can be applied by the user afterwards as needed, rather than before training starts
patents and copyright are all FUBAR
Well, there's a lot of folks that patent ideas when they're not even fleshed out
delete copyright; copyright = NULL; is the only solution
Which is my issue with patents speculation
we should go back to a 10 year copyright. Then public domain. There's going to be such an influx of new content coming that it makes no sense to hoard teh creativity of the planet
oh that reminds me
I need to patent my new hashing algorithm
I'd love that, since it means basically everything I enjoyed as a kid would be public domain within (now) to the next several years or so.
middle out hash?
has a software patent every actually caused any good in the world?
it's an algorithm for hashing images
Just patent the essential idea of a story, and not the finished product
silicon valley reference
might as well scrap the whole category
AV1 patents have been good
Patent troll's world, sure
hash(image) -> uint64
I think the free software foundation, linux, some FOSS organization, likely patents concepts in order to keep them protected and open. Wouldn't be needed if the patents weren't so stupid to begin with though
Linux is also trademarked
AV1 is open source, no?
yes, but there's still freely licensed patents involved
it helps keep the MPEG-LA trolls away
ah, that makes sense. But that's just using software patents to solve a software patent created problem
fair enough point there
Mpeg .. the motion pictures experts group, an industry lobbyist group basically. Worth keeping in mind
media lobby groups suck
yeah exactly. guerilla warfare really.
Not to be confused with JPEG which has actually been helpful
codecs are a trivial signal processing problem and should never have been patentable fight me
thats the joint thats the jam turn it up play it agian
*I used the stones patents to destroy the stones patents
unexpected thanos
Honestly, I would be okay with lobbying if it didn't screw over major stuff that would be beneficial to everyone, just so the few can whine about it
lobbying should be a crime like scandals. if they want to advocate, they should talk to the public who tehn talk to the politicians
what do you do about testifying before a committee hearing though
There's legitimate lobbying, with folks who work with folks who are in poorer areas with little to no political power, stuff like the original environmental movement before people knew how bad companies were covering up the eco consequences of modern society
that's a large part of lobbyists job, does it really make sense to ban social media companies from testifying during a hearing about regulation on social media companies?
for instance
thats mostly an american circus thing i think
so place huge limits on the process that give those small lobbyists a level playing field with trillion dollar corporation lobbyists
But I am mad particularly at lobbyists that are for less...savory purposes, say astroturfing a benefits bill, or a lobbyist from a company trying to shut down a bill that would affect their workers ability to unionize and such
I'd be in favor of a copyright/patent reform that requires all patents to be tied to a human life. Once that life ends, it hits public domain.
omnibus bills suck too. misrepresenting themselves and have 10k pages of pointless information, then one little line about a fuhrer
I wish that's the case, but even once small lobbyists can evolve over time to become major forces
this copyright mickey mouse for over a century, and well after the death of its creator is lunch meat.
It's the law equivalent of an EULA
20 years from conception imo
Btw, anyone who writes those and wants those approved should be forced to read word by word what they wrote.
let ai do it
To everyone who is about to approve it.
I mean if the opinions are already hardly split on ideological lines then yea it won't change that much, but for the majority of generic bills it has a large effect on whatever makes it out of comittee.
ai governence, whether civic or corporate, is the next frontier
Taste of their own medicine for them trying to pull that BS.
tbh the real problem is 1,000+ page omnibus bills written by lobbyists
that no one reads and rubber stamps bc party line
And lawyers and judges both hate that crap
My issue with that is for individual creators it can take a long time to get off the ground with your idea, and you could lose rights about the time your project is taking off. On the other hand my kind of legislation might promote corporate assassins.
yeah its stupid. humans are incompetent to govern in the modern climate
Like wth
democracy is broken currently, but not lost
I theorize that republic died with prohibition.
very cyberpunk like. you son of a bish. i'm in
The moment we allowed Disney to change copyright law in the US, that's the moment everything changed
Pretty much everyone outside of Washington does, but no one really bothers to do much about it
New medicines only have a seven year patent
You mean "hostile takeovers"
reminds me of the conspiracy theory where record labels were supposedly assassinating rap artists for the life insurance money
instead of patenting it, they should call it "Molecular sculpture" they'd get better IP protection.
That's because unfucking the problem would take at least a Second Coming of Christ and other religious leaders who says they're coming back.
It's not an easy problem to solve, I'm not touching that lesser dwarf of turd problem
tea party least managed to get pork temp banned... until the next election cycle or so
Tbf, you can mess with a few things, food is ironically one of the few things people will kill you if they do.
I invented a toy to teach blind kids braille, and quickly found out it was in my best interest to not pursue a patent.
sorry I was referring to kickbacks in bills known as pork not necessarily the food
Oh, I am not terribly well versed in older political stuff
already patented, too costly or what?
There's also you having to disclose how it's made, which is why Coca-Cola haven't patented their drinks in the first place
Patents mostly only benefit large corporations who already keep lawyers on retainer. A patent basically requires you to provide exact blueprints and tolerances to everyone who would steal your idea, and the first point in every patent court case is "The patent shouldn't have been granted in the first place because it isn't a new idea" then you'll spend months defending the decision the patent office already made before you can even start to address whether they violated your patent or not.
They just hide their secret formula, that's all
For an individual I conclude you're better off making a public video about your idea and roughly how it works: this makes the idea unpatentable at all so if they compete with you they at least can't steal your exact specifications out from under you.
you pretty much need dea approval to make their secret formula
they still use coca leaves
they extra the cocaine and sell it to a medical company
Well, not the kind you're thinking of. The cocaine is gone currently 🤣
It's probably there for that "taste"
Which channel can i ask common questions for SD settings?
yes but importation of the substances & refinement process stuff requires approval
there for taste is most likely
Again, that's why the EU's Anti-AI law can be devastating towards open source AI models, you're forcing them to reveal their training datasets, which is essentially a trade secret for AI companies
yea that makes sense. Legal fees aren't cheap and cases take forever.
OpenAI can open source their models, but they will never reveal their actual datasets they used
It exclusively screws over anyone who wants to do this stuff commercially
Since it just takes away all the value in doing so
for now I 3d print the braille toys, but it's hard to sell them because federal funding for blind students is locked to one specific vendor.
EULAs are the law equivalent of EULAs already
Most web servers reject payloads that are too big to be processed sanely. Why don't humans do that?
They're essentially a modern Faustian contract that is the only thing that people can sell their souls to because it's too long to read and too complicated.
POST /submit HTTP/1.1
... 46 GB of garbage with a useful message buried somewhere inside ...
Some EULAs prohibit reverse engineering, but I think I've found a way around that
I put the details of my reverse engineering behind yet another clickwrap ToS wall that forces them to give me a license to reverse engineer the software.
If they found out I did it, they agreed to the license
Having the AI do it? 🤣
ah, now that's frustrating on multiple levels
That would be awesome if someone can drum up support.
Hello everyone
I don't know anyone, but that sounds like an awesome mission
lobbying but for good lol.
It's possible, but you'd need to get someone to write an article about it, find a sympathetic congress member, etc
this is what I'm replacing: https://www.aph.org/product/swing-cell-compact/ Mine has a perkins-style keyboard, no removable parts and retails at $40
American Printinghouse for the Blind is the only authorized vendor for federal funds, and they sell a piece of wood with pegs for 80 dollars.
replacement pegs - a 20mm section of an 8mm dowel costs 6 bucks.
Wth
sorry I'll give it a rest. My son is blind so I get kind of fired up about it.
That's stuff you can get at Home Depot or any home improvement store for way less
But yeah, I get it
wow they explicitly only wrote APH into the bill
essentially guaranteeing them $200 for every blind student in America, every year and the schools have to spend it, or lose it.
so thats why they're charging $79 bucks for a block of wood
corruption in government contracts like that happen all the time
sad but true
what how i run out of memory?
EU Anti-AI law... how ridiculous and sad.
"Looking at you AI Art" - interestingly disingenuous post on LateStageCapitalism on reddit, posted Cybergems video from last night, got an instant downvote
I mean fine they can do as they please, and i might keep making a bit of money selling 3D models for a while longer, but anyone with half a brain can see its just clinging to a by gone age hoping to stop the inevitable.
Its futile for a government or companies to take an Anti-AI stance, because AI cannot be stopped, it is the future, the cat is out of the bag, and its potential benefits far outweigh trying to desperately cling to the old ways of doing things and the old systems.
Exception training model: max() arg is an empty sequence
what does that mean?
in auto1111 dreambooth
weird. whatever bot is watching that post, wants to make sure my comment stays at 0, not -1, not 1, and the change is near instant. or am I just paranoid...
Reddit has some weird delays with their ratings, I usually just post and leave and if it gets ratings the app notifies me
Yea. Well, I don't know why I got on reddit anyway. The post, and catered conversations, were very uninformed and almost pro-capitalism, for an anti capitalism sub.
what's in the law
Idk I'm basically fine just being an observer to the discourse at this point, because it's like I read the script and already know the ending.
Technology wins?
Ehh. Not pessimistic, but following a different script
30 years ago if you told everyone at a 2600 meeting that in the future we'd be walking around with multiple trackers on us, they would have laughed ...
And those were the informed few. Probably fought against it once they realized... But imagine everyone not having a bug on them in these times. lOL
A tracker?? You mean phones?
Yeah we're chipped already, and the chips are in our pockets with our consent as well
Again, I can't say this is bad, really, just not in line with what we're taught as the meaning of freedom
they protest first of all against AI art companies using their artwork without telling them anything about it at all and because people pretty much steal their identities and sometimes want to profit on their back
and then there are many more things
which one?
you are lying there, its not anti-AI stance
noone is agaisnt AI
"noone" is the wrong word ok
but people are planting fake news and claims on this all day
I just updated to v2.1, is there somewhere I can change the default width and height to always be 768 instead of 512 when opening the webui?
thats bullshit
yep! hopefully it creates some interesting discussion
You can change values in ui-config.json in the main webui folder
(assuming by "the webui" you mean AUTOMATIC1111)
Very nice. It worked. Thank you
saves so much hassle changing it over and over
artists should understand that logos communicate ideas very effectively. And the logo they all rallied behind was a gigantic AI in a ghost busters style NO sign. Literally anti ai communicated through the logo. dont be surprised when people think that's what the book is all about.
they bombed anywhere they could with this AI behind a NO sign logo for days. It was a very intended message, otherwise you're telling me they're all daft to visual design languages
The moment i saw the logo they rallied behind, being someone who has a visual arts degree, i thought "that's going to be a bad look"
question for devs
hi devs - i ask the API to give me n images (4 in my case) and dont actually know how to get the four images. I can only get the one.
for resp in answers: for artifact in resp.artifacts: if artifact.type == generation.ARTIFACT_IMAGE: blah
Hi, Guys. So, what do you think Lexica is doing to get those amazing results with Aperture v2?
I'm telling you, it's gonna take the lead from MidJourney.
They just have good source data for training. they're not doing anything special. Maybe adding to prompts with a gpt script behind the scenes
lexica.art?
are they... what are they?
are they based on stable diffusion?
Someone decoded what the "No to AI" sign really meant, and spoiler: it's not good
im really not scared of "no AI" as obviously they are just blocking progress
what was it?
Let's just say it has to do with party of a certain failed German artist
I see it as the world's biggest conformity bubble, when one of the arguments against AI us it being cheap and unoriginal
uhhhh... how?
And they just frikken copypasted it without editing
Although one of the signs saying Yes to AI talking about 3D texturing and retopology made me crack up
AI retopo? YES PLEASE. Dear god, YES YES YES. I friggen hate retopo. GUH.
What happened? 🥺 😭
-clears throat- Sorry. Here's what I was actually gonna say:
I definitely don't agree with the No AI campaign. It's way too broad, aimless, not very well thought out at all, and it's not going to protect artists' jobs. That said, let's try to have a little empathy. A lot of artists are afraid that the AI is coming for art jobs. They're scared. I'm not saying that's an excuse for the panic but fear isn't always rational. Often, it can be very irrational.
if they get replaced they get replaced im not about to keep anything just becuz its old
Regardless of jobs, I can't let anyone take my AI toys away. I need to buy more disk drives and GPUs.
Remember that assembly line workers were largely replaced by machines, and nobody shed a tear for them.
tell them to learn to type better prompts
tech is just an extension of our nature its not seperate
A lot of commission and freelance artists are seeing AI spit out dozens, hundreds, thousands of images that are as good or better than their own and in only a tiny fraction of the time. Some artists are seeing their entire bodies of work packaged up into models that basically turn their style into a vending machine.
That's scaring the crap out of people.
I'm not asking you to change your opinions. I'm just asking you to show a little empathy.
That's what empathy means to you? Feeling bad about things?
tell them to get gud
well sounds like they are in a sad position but i stink at drwawing mostly so im very excited
I understand their fear, but I don't think most of their responses are acceptable from a legal or ethical perspective (especially not when evaluated in the context of a lot of their current practices).
ai art meets 3d printing there boom!
I mean, I'm an artist and I'm excited for the technology. Not all of us are out here with torches and pitchforks.
well good! i think "standard art" is still standalone, and ai stable diffusion and such is what it is
Not all artists are anti-AI artists, which is generally why I call them the "anti-AI mob" now
no reason not to coexist
My empathy for them kinda runs out as soon as they start sending death threats, tbh
AI isn't gonna nuke art jobs overnight. That's insane
I don't think anyone thinks AI is gonna nuke all the art jobs overnight. That's just silly.
I think a lot of people are looking at where AI was a year ago versus where it is now. That's a HUGE leap.
That's the sentiment I've gotten from (and has been directly expressed by) a lot of the anti-AI mob. I think it's ludicrous personally, but there's little convincing them.
Of course, some asked me to promise that AI won't take their job in a week, in a month, or in a year, and all I can say is that the future is known by no man.
Life offers no guarantees to anyone, so you must evaluate things in the context of the now.
did big stovetop ruin your firewood business?
how much of a caveman is acceptable then
Most workers are going to be replaced by AI in the future, not just artists. I'm starting an AI engineer this year and when I graduate in about 5 years if all goes well, there will probably already be AIs that can replace my job, I don´t understand why people get scared, it´s just technology.
oh no i wanted to work harder!
I feel like this is why we need to start steering the conversation away from fighting about whether AI is going to do this or that or ban it or burn it all down. None of that is productive. What we should be talking about is how AI can be integrated ethically into artists' workflows and how society can accommodate people who have lost or are going to lose their jobs to AI.
society can accommodate people who have lost or are going to lose their jobs to AI.
This. I really tried to bring this up and everyone just called me a communist (I said nothing about abolishing private property or government-controlled everything).
lol
There's a pretty sizeable chunk of artists who are against AI even when done ethically, though.
Took a bit of a break what's the newest SD version?
Lawrd save me from the "everything I don't like is communism" people.
Well, they're either gonna have to adapt or....
It is good, but it can always be improved.
I literally suggested social programs that mostly already exist and work in certain countries. But no, that's communist now.
Oh gawd don't tell me this group is infested with the ethics of art talk now
It's on-topic, but we have a lot of channels here. Pick one that's fun.
oh gawd dont tell me youre invoking gawd
we talk about this daily lol
i really dont feel like im ripping people off when i use ai art stuff
using ai stable diffusion is an art in itself anyhow
Does anyone even still use artist names in prompts?
i mean, does it even work like video game artists idk
Who cares if artists lose they're jobs they produce nothing of real economic value and profit off hard working people with real jobs. Until this year artists were high and mighty about having "irreplaceable jobs" looking down on everyone else getting replaced
If anyone has a completed comic book, the aicomicbooks.com site is going to be updated “coming soon” and new books… promo through email, social media and press releases. There are no costs associated with this.
To be fair, most people produce nothing of real economic value.
Most economic value is fake. The economy is mostly fake.
Everyone's just so invested in this system that it has to keep going.
That's an incredibly poorly thought out take.
I have great respect for people with art skills - they make the world a better place, and as impressive as diffusion is .. its not capable of replacing every artist role . I do wish they'd see the ways AI can help them rather than see it as a threat though
art has no value? ouch
I've build houses. Boats. Bathtubs. Sidewalks. Artists make pretty pictures
THIS. at least artists make something visible
Real value. Society can live without art
society is crap without art
If you define economic value as "what's necessary to survive reasonably easy" then yes
hey - how can train on top of SD to get better images along the lines of what I want?
thats like arichtecture, video games the whole thing you just threw out including music
enjoy your sidewalk
Yeah people that don't provide a a tenth of what blue collar workers make
Art is a subset of entertainment (which isn't just TV, games, or even books). Entertainment, like socialization, seems to be necessary for humans to function reasonably well.
i've argued this in a differetn context . "both are true". we dont need art to be a job, but art definitely contributes significantly to quality of life, and we're already productive enoguh that we dont need everyone busy making food like in 1700
blue collar doesnt provide a tenth of the ceo's
hi @gritty scarab I dont really know how to move forward on this. i've looked at the docs and tried to engage people here for help and I am stuck :(. Any advice?
so you want more CEO's?
So you think people's standard of living should be tied to how much economic value they provide? Then why is Elon Musk a billionaire while teachers are often clinging to the very edge of poverty?
CEOs
create value
HAHAHAHAHAHAH
artists seem totally ok with automation replacing blue collar workers though.
Nuh uh. None of that.
If art has value, then being able to produce it much faster in much higher quantities will increase the amount of that value the world gets, as its the number of quality stuff not the rate that matters here. If art does not have value, then nothing of value is lost.
its funny how many ways this could go.. "AI will provide for everyone and we'll all be artists!" ..vs .. "AI is replacing artists & programmers first - we'll all be doing manual labour.."
To be fair, Musk does work incredibly hard (he seems to be a workaholic, and also egotistical and a bit unhinged); I'm not sure if his "work" is doing anything positive lately though.
I think if you're you can stop working and the world would never know you don't have a job you have a hobby
hi i have a question is it possible to create an picture with an word inside?
i want to create an logo with name inaside
every time they argue about automation they say that automating art is somehow magically different than "menial" jobs, as if you cant possibly derive purpose from like, driving a truck or building things
This is because AI interfacing with the real world is hard. Robotics is still very tough. If robotics were easier, then we'd have more real life robots doing more complex tasks. Maybe even human-like androids.
im trying to get a proportional human with inpainting and im getting the craziest long arm multibody head within a chest outputs. any tips/tricks?
These models cannot do text. You'll have to add yourself in Photoshop later.
negative prompt: malformed
conjoined bodies in the negative prompt
ok thanks
automatic1111 invented negative prompts
An empty negative prompt is called classifier free guidance and has been around for a while. A non empty negative prompt was invented by a dev playing with sd
anyway .. SD is giving me infinite textures, but its not making efficeintly textured game worlds.. artists are goign to be busy for a while yet IMO. they just have to get out of the mindset of single images
an empty negative prompt is just CFG
non-empty negative prompts are negative prompts
artists should focus more on 3D and they'll be ssafe
3D is next
yeah idk how long 3d is safe lol
i'm skeptical - the output from these models is closer to what you get from scans and its not as useful for games as handbuilt art
Ok will try that. I already had deformed and disproportionatein negative prompts.
and no one is going to be fighting to save the poor 3d artists right to have spiritually fulfilling work or whatever, 3d art is already considered as a lesser form of art
equally i'm interested in trying ideas to apply diffusino to this sort of thing.. but in game art a human can make a set of repeatable things that give the impression of a scene without needing every peice to be unique
i'd say good human art is 10x as effficient to render as scans
scan/photogrammetry assets are basically the default/showcase asset style for ue5, no?
resulting in huge game downloads..
Personally, I can't wait for more usable text-to-video so I can feed it a crap-ton of totally not pirated anime and get an infinite stream of content.
- and not running on mid to low end hardware
I'm looking forward to integrating it into a 3d workflow .. it'd be cool to be able to generate unique textures on the fly in game.
i mean idk what i'd use that for but it'd be cool.
i doubt it'll ever be fast enough for that, although i have wondered if you could make a styylzied dream-like render with some kind of realtime NN in screenspace, with the game art rendering into a nerual intermediatte
this UE5 stuff is better for films than games (i've heard of UE5 being used for CGI)
i also can't imagine AI figureing out efficient animation rigs.
I mean sd's vae decoder runs in 1/30 of a second on my machine. It's not necessarily that hard to bake a photogrammetry asset into a latent texture set and decode in runtime when you do virtual stitching. Prospective technology maybe, but certainly within reach
Maybe not today but maybe someday and that would be amazing because !@#$%^&*! rigging.
it wont need rigs it'll go straight from prompt to animated model. 
thats my uneducated prediction
That would also be amazing. I'm totally down for that. 😄
Rigging is my least favorite part of the job.
efficient rigs and efficient topology.. hah the artists lose their favourite part (concepts) and get left with the parts of the job they hate the most 😦
yes i hate it
I think they had a demo with one of those big led walls/rooms they used for the mandalorian.
Am I still timed out?
i hate animating more than anything though. i'd rather retopo something
Im looking forward to on-demand 2D animations....like create a walk cycle for a character from a few different angles
Um can someone tell me what I'm not allowed to say. Not trying to cause problems just giving my honest opinion
i can certainly imagine an engine with an asset pack that allows that,but not the AI itself generating those from the ground up. Ai could guess which of several pre-defined rigs & topologies to use to approximate a given conceptm perhaps.
Talking smack about how art has no value and artists don't have jobs but "hobbies" when we're living in a world that is basically made of art. That's incredibly insulting.
NGL, I take that a little bit personally given than I'm an artist and it's how I've been making my living for the past 20y.
I didn't say art has no value. I make my own art for a reason... but okay.
Well, if coding and programming was worthless the way art is perceived by the world at large, it would be programmers who would be in the same shoes
Clearly, it's only real work if people will immediately die without it.
That also happens to exclude well, well over half the working population.
I'll stay out of it. I'm just here so I can make better art
i'm a gamedev, i'm not going to accuse artists of creating no value . now being a programmer its in my interest for AI to replace artists.. but I still respect their skills, they do deserve support, but equally i'm critical of the luddism some of them are showing
Again, just walk a mile in someone else's shoes kinda thing
this , we long ago crossed a point where we dont need everyone working to provide essentials. we have service economies in the west. but we do work to justify ourselves. Hence a lot of artists etc.
Weird how jobs that keep people alive or the whole world running is paid minimum wage
Which is pretty crazy to me honestly
Clearly, work is a psychological need for humans - we have to do something, and that something doesn't really have to be a coal mine where your coworkers are seven and eleven years old respectively.
Stuff like retail workers, fast food, gas station attendants, etc.
You need those people to help even for things to function, but people treat them like garbage
AI could take their jobs too! Except people really f'ing hate interacting with machines in customer service.
there's a reason countries with industrial agriculture don't have famines anymore.
Also, there's certain jobs that people won't trust machines to do, nursing, childcare, very skilled labor handling people in general
I agree on that point
Once we get human-like androids, they'll be in nursing and childcare, you'd better believe it. And there will be a subset of people who are seriously creeped out by that.
bring it on 🙂
And, well, I'll just watch and accept whatever happens in this weird world.
Except that the AI becomes incredibly clever