Others Installing Stable Diffusion AI on Win10/11
- Thread starter Deleted member 1121028
- Start date
Emad is an idiot and is just trying to save face in the name of his business venture. Tools like this, especially open-sourced and freely available, are continuing the pattern of breaking the entire concept of copyright apart at the seams. It is unethical to expect copyrights of 100+ years or more to continue to be the norm, while the demand for public domain is greater than ever.
Even now, YouTube is forced to emulate TikTok's model of not giving a fuck about popular music copyrights for its shorts. Of course, TikTok itself is just Chinese spyware and ByteDance does it because China allows them to, but at least some good has come from it. If the corpos had their way, Captain Picard would have to pay for copyright royalties every time he ordered his "tea, earl grey, hot" from the replicator. So, I welcome whatever brings us one step closer to a utopia where we're sharing our resources and not fiercely competing for them.
The creative world has a new tool to make their jobs easier, but the intense oversaturation of the industry is going to force people to re-think why they should bother jumping into an ocean of tens of millions of others if they don't have their own creativity to bring to the world. We're going to need universal basic income to replace the lost jobs in market that is already stretched as thin as possible. People need to join an industry because they want to and are good at it, not because they have to to survive.
Alright, philosophical arguments aside, has anybody found a good database of Textual Inversion concepts for sexual content? I know I could make them myself, but I'm sure thousands of other people are just doing the same thing, so I figured there was a database somewhere.
In the meantime, I'll share what I have on the subject:
*
* The
* Big ol' database of
There's a few subreddits with adult SD images out there (AdultDiffusion, sdnsfw), some with prompts and descriptions of what they are doing.
In the meantime, I'll share what I have on the subject:
*
You must be registered to see the links
* The
You must be registered to see the links
for
You must be registered to see the links
, an offshoot model from SD designed for anime modeling, using tens of thousands of images from Boora and its tagging system* Big ol' database of
You must be registered to see the links
. If you want SD to identify something or create it in a specific style, this is how you get there. I want a NSFW version of this, but at least some of the styles like H R Giger, Joe Madureira, "retro girl", and Sakimi Chan to get you in a more lewd direction. Just a prompt like "<sakimi-style> woman" already gives you great results.There's a few subreddits with adult SD images out there (AdultDiffusion, sdnsfw), some with prompts and descriptions of what they are doing.
- Jun 10, 2017
- 10,145
- 14,829
In most of the world, copyrights last either 50 years after the creation, or 50 years after the death of the creator. It's only in the USA that they change the Law each time the firsts Disney reach their limit date.
People don't want public domain content, they want free content, it's not exactly the same thing due to the radically different motive behind the twos.
They aren't doing it because of TikTok. It's the Law, short are too short and the music they use fall into fair use.
Yeah, it totally remove competition, there's absolutely none between Opera, Firefox and Chrome.
As if. AI generators (whatever they generate) will just lead to more people jumping on the market in hope they'll get their share of big money. Not only they'll not anymore be limited by their lack of creativity, but they'll also have way less to invest since the only tools they'll need will be free. So it's all benefit for them, and if it don't works they'll lost absolutely nothing, not even their time since the AI will do 90% of the work for them.
Well, Mark Twain started it. But, I don't think Disney is going to be able to get away with another extension again. Congress isn't in a good enough shape to push something like that, though, and the public backlash will be much stronger, now that people are aware of what happened in the past.
Also, 50 years after the death of the creator is still too long, IMO.
Depends on the people. The general public wants free music, and they've basically got it with music streaming services everywhere. It's not technically free, but it's a helluva lot better than spending $15 for a single CD. But, the general public isn't really going to be messing with Stable Diffusion. They are just going to reap the benefits of it.
YouTubers, streamers, indie game developers, and other small-time creatives want public domain and royalty-free content. They want to be able to use artistic resources to put out some video without having to deal with YouTube's shitty copyright tagging system. Not everybody has an entire team of artists, musicians, marketeers, etc. to create the content themselves.
Fair Use is only a US concept. And if you think YouTube, as a company, cares about fair use, they don't. Otherwise, they wouldn't be saddling their own creators with their copyright tagging system every time a video fairly uses some audio/video content. A system that has already been gamed by the corporate creators with dedicated third-parties that process DMCA claims in discriminatory ways. ("Oh, you have an negative opinion about our multi-million dollar movie... well, I guess we're not going to grant you fair use rights to use footage for the movie unless you spend hundreds of thousands of dollars in legal fees.")
One costs money. Another is used as a vehicle to push their ad revenue. And a third is developed by a nonprofit and puts it out for free as open-source software.
Not sure how that's a comparison.
Well, I actually agree with Emad's first paragraph in that tweet. Not everybody can create art, and this opens the door for infinite possibilities for people who couldn't create before.
But, also, like all technological progress, this shakes up the industry enough to put some people out of a job. For example, I can create any sort of artwork I want with a little bit of time and effort. Why would I go to some overpriced art fair to purchase a painting or framed photo, when I can generate whatever I want here, print it on a good color printer, and frame it on my wall?
I can only hope that in the next five years, it can create porn of such quality that you can just dial up whatever types you want, and it'll generate it for you. And put a dent in the unethical aspects of that industry. I'm sure AI porn won't completely take over the real stuff, but it'll be enough to reduce the footprint.
- Jun 10, 2017
- 10,145
- 14,829
US$ 15 isn't much for owning your copy of something and being able to hear it wherever and whenever you want.
And the guys who created those musics wants to earn something from it. For one group/singer that earn millions thanks to their creation, there's nine that have to go to works every day, even after a night of rehearsal or concert.
Among those nine, one will be lucky and have a one time hit, but it will not change much his life, just make it better. Ask google what happened to those stars from the 80's/90's that you haven't heard about since their big hit. Most of them are now back to their daily job.
Why should those who are the less creative be also the ones who benefits the most from the creativity of others ?
The notion of fair use is defined by the Berne convention regarding the protection of artistic works, and was already part of the original 1886 text. All the 176 signatory countries have it includes in their own Law.
They don't have to care about it or not, it's the Law, they have no choice but to apply it in the 176 countries who signed the Berne convention, period.
What isn't Youtube responsibility, and can be perfectly argued against in justice. It's not because no one, youtube included, want to starts a judicial war against a major movie company, despite being sure that they would won the case, that what those companies do is legal.
And it even less mean that youtube don't care, nor respect, fair use. They are a private company streaming content made by third parties, they don't have so substitute themselves to justice.
In the end, all those words when you could have just wrote a single sentence: I don't know shit about copyright and legal matters, but I'll still know that you are wrong.
I don't think you know what I do and don't know and I'm just going to end this (previously civil) argument right now.
Sorry I brought it up. Let's talk about how to use this amazing technology instead.
- Feb 16, 2021
- 101
- 297
You want to convert 3D garbage into usable 2D images?
simple: buy half a dozen NVIDIA Tesla A100, build a small server that can run the cards, next pick your favourite porn star that closely matches what you are trying to create, download the highest quality videos of her and break them into images, 60k danbooru... lol try 10-20 mil frames of porno, next run
You must be registered to see the links
run that program with 4-8k face samples of your character then just create the word prompts and bingo bango you are good to go!As far as copyright?
Since the first charcoal painting was smeared on a cave wall, humans have been copying each other; this AI just simplifies the process, what happens when the infant AI turns into an adult? Do you think you are still going to be able to distinguish what is being copied? At best we are going witness a new golden age of animation, at worst it's a toy we lose interest in.
Last edited:
No, I was curious if there was already a porn version of
You must be registered to see the links
somewhere on some Discord server or Mega link.Why? That's extreme overkill.
Also extreme overkill.
I'm not trying to "re-create my favorite porn star". Besides, that kind of data is likely already in the Stable Diffusion 1.4 model.
I'm just looking at conceptual data,
You must be registered to see the links
. Sex positions and the like.Just... no.
From Textual Inversions own website:
From the top guide about it: Download 3-4 images from here and save them in a directory. This will be our training data.
From another guide: about 3 to 5 images for best results (model may not converge if you use too many number of images)
DreamBooth is already old news. If it's not based on Stable Diffusion, it's old as shit. Even news from two weeks ago is old as shit. The whole concept is evolving in real time.
Hell, I tried like three different GUIs for SD before I realized that everything was converging on AUTOMATIC1111's version.
- Feb 16, 2021
- 101
- 297
The issue is the data is not available, and what it takes to create that data is where you fall short of understanding what is required to be able to create CONSISTENT images
but hey you keep waiting for that mega link
You must be registered to see the links
, and I doubt even with DreamBooth you need such overkill. At most, people have been putting 10-20 images in a TI concept. Not 10-20 million.
- Feb 16, 2021
- 101
- 297
You don't have permission to view the spoiler content.
Log in or register now.
wrong color, loss of texture, and looks more like a van with a camper attachment then it does an RV anymore.
now let's say i wanted jennifer lawrence to blowbang mickey mouse and crew while she shoves dildo in the shape of hulks fist up her ass, let's say i wanted to create multiple images for a comic, do you think training the current AI with 10-20 images is going to create a viable result? what's the requirement for even just training a simple face swap...
- You can now run this on a GPU with 24GB of VRAM (e.g. 3090). Training will be slower, and you'll need to be sure this is the only program running.
Last edited:
For that sort of multi-piece scene, you're better off with inpainting and combining each one. And you'll probably need to play around with the
Also, it's all a matter of how high-quality you want to get. If you want to train a model with every possible picture of Jennifer Lawrence, go for it. Just know that people have been able to pull off
You must be registered to see the links
. Even those are trained with anywhere from 200 to 200,000 images.Also, it's all a matter of how high-quality you want to get. If you want to train a model with every possible picture of Jennifer Lawrence, go for it. Just know that people have been able to pull off
You must be registered to see the links
, with a LOT less effort.Crowdsourcing, as always. How do you think all 607 of those other embeddings came about?
Hmmm.... getting a little more streamlined over time.
"Textual Inversion just got added to the AUTOMATIC1111 webui repo"
BTW, I've also found torrents for a
Other useful tools added to the webUI since some of my last testing, you can select the model you want to uses (ckpt file) under settings, and there's a tool to merge checkpoints as well.
"Textual Inversion just got added to the AUTOMATIC1111 webui repo"
You must be registered to see the links
BTW, I've also found torrents for a
You must be registered to see the links
and a
You must be registered to see the links
, but I can't find details of where they're actually sourced from, I assume one of them is 150k images from rule34. Let me know if I shouldn't post those magnets.Other useful tools added to the webUI since some of my last testing, you can select the model you want to uses (ckpt file) under settings, and there's a tool to merge checkpoints as well.
Unstable Diffusion Discord is helpful, and AUTOMATIC chats there on the regular.
You must be registered to see the links
- Sep 12, 2021
- 510
- 741
I'm toying with Stable Diffusion (website only, as my potato has only 2Gb Vram) and it's tickling my interest.
I'm just aiming at placeholder backgrounds for the game I'm developing (the team artist is light years from reaching the first chapter and I noticed that the people I'm asking to review the the game read only the prologue, that's the part with pictures). So nothing too problematic.
My gripe is: how do you people turn those 512x512 pictures into a widescreen? The way SD deals with field organization is very, very squared, the only solution I can see to transform the images in 16:9 is to do a lot of work in Photoshop, that would kind of defeat the purpose.
I'm just aiming at placeholder backgrounds for the game I'm developing (the team artist is light years from reaching the first chapter and I noticed that the people I'm asking to review the the game read only the prologue, that's the part with pictures). So nothing too problematic.
My gripe is: how do you people turn those 512x512 pictures into a widescreen? The way SD deals with field organization is very, very squared, the only solution I can see to transform the images in 16:9 is to do a lot of work in Photoshop, that would kind of defeat the purpose.
Recommendation last month was to use 512 x 288 then upscale. There may be other tricks now, but the training is based on 512x512, so you either keep one edge at 512 and the other lower, or i think you can do 910 x 512 and use the high res fix. Time to play with resolutions i guess.
- Sep 12, 2021
- 510
- 741
Thanks. Upscaling isn't a problem, but the online version has no size option.
Since this afternoon I was alone at my workplace I quickly installed the UI version on my workstation and did some experiments at 512x256, but there I couldn't get decent results for some reason, even using the same prompts. Doh. I guess I'll have to learn how to use outpainting scripts, instead.
There lot of stuff going on and I really don't have time to track everything .
But it's a good article (imho) between copywright/fund/academy/liability :
Visual artists really get fucked here as they don't have strong protection as music industry has. Choice to go open source is not as innocent as it may seem. StableD uses Laion and scrap from Shopify, PinInterest, GettyImage, FineArt. AiNovel/Waifu also use Laion and fine tune via 56k images from Danbooru. Meta (Facebook) scrap from Shutterstock/Youtube.
Seems risky for company to expose themselves to such potential law suits. Not sure how things gonna turn.
.But it's a good article (imho) between copywright/fund/academy/liability :
You must be registered to see the links
Visual artists really get fucked here as they don't have strong protection as music industry has. Choice to go open source is not as innocent as it may seem. StableD uses Laion and scrap from Shopify, PinInterest, GettyImage, FineArt. AiNovel/Waifu also use Laion and fine tune via 56k images from Danbooru. Meta (Facebook) scrap from Shutterstock/Youtube.
Seems risky for company to expose themselves to such potential law suits. Not sure how things gonna turn.
- Jun 10, 2017
- 10,145
- 14,829
Hmm, there's a beautiful debate regarding Law that hide behind this.
Technically speaking, what they are doing isn't different from a painter who pass his days in museum, looking at what other artists did. It will stay in his memory and become a part of his inspiration. And since the images are directly available, those websites can be seen as a modern form of museum. Therefore, from this point of view, what they are doing is legal.
But in the same time, an AI don't see. Therefore they need to download the images in order to proceed them, what mean that they aren't anymore just looking at them, but effectively using them ; plus many websites forbid this. Therefore, from this point of view, what they are doing is illegal.
But even us humans can't see the images from a websites without them being downloaded by our browser. We don't have direct, nor easy, access to the downloaded files, but for a while the images will be on our own hard drive. What then make their own downloading not so illegal.
But they aren't using a browser, but an AI can't use a browser, but an AI can have a browsing ability, but I highly doubt that they cared to do this, using a scrapper instead. And probably many others "but" that I don't think about right now.
In the end, after a long debate between experts, the judge would have to decide if the fact that the scrapper and the AI are two independent software, and probably located on different computers, prevent them to be seen as a single entity or not. This in the same way that we are one entity with the browser we use ; one being the user (AI/Human) and the second being the needed tool (browser/scrapper)
Single entity case: an AI need a computer to exist, and can only use scrappers software to browse the net. They are a single entity, and their use of those websites do not differ from the way a painter use a museum. It's legal.
But if the judge consider that they are two entities, then it would be illegal.
If I had to judge right now, I would ask only one question: Is the AI controlling the scrapper-like, or are humans starting it, then providing the images to the AI ?
If the AI is controlling the scrapper-like, it's legal ; an AI can't do other way to "look at what other are doing". But if a human have to intervene, then it's illegal ; they are using arts/photo without the consent of their rights owners.
All this being said, it's only the right owners that can sue them, and I doubt that those who have the money for that have the time, while those who can have the time probably don't have the money.