It sounds like you’re proposing that someone should create their own fork. That’s always an option, and I encourage everyone to. But what else is the proposal here? Forcibly move the repo? It doesn’t really make sense.
It’s funny seeing “the code is a mess” presented as a reason why a repo is problematic. It’s code. We all have strong feelings about code. Everyone has their own way of doing things with different engineering and product tradeoffs. In this case, automatic was right for pumping out features as quickly as possible and getting the first mover advantage.
So, concretely, what do you feel should be done?
I also feel like people haven’t read the actual TOS:
> We do not allow sexually themed or suggestive content that serves little or no purpose other than to solicit an erotic or shocking response, particularly where that content is amplified by its placement in profiles or other social contexts. This includes: ...Pornographic content
https://docs.github.com/en/site-policy/acceptable-use-polici...
There’s a larger issue here, far behind automatic and this one repo. Those links clearly were not “little to no purpose other than to solicit an erotic or shocking response.”
> So, concretely, what do you feel should be done?
I'd rather see a more professionally maintained competitor, like InvokeAI, surpass.
To explain myself, these are specifically my pain points with Auto's repo:
1) No stable releases or versioning. This makes it difficult to use, as any given pull of the repo could be in a broken state. A better alternative is to, at the very least, tag known working commits.
2) Lack of testing; this is related to above. Even basic tests that would stop things like LDSR from randomly breaking would be a huge boost in usability.
3) Easier to use UI. This isn't a huge thing; Auto's UI is fine, though some better guides and documentation would be nice. But InvokeAI's UI and default settings are much better, in my opinion.
4) Lack of an official Dockerfile. It's nicer to have things in one place, especially to keep them in lockstep with code changes.
5) Association with racism.
When I said "professionally maintained" I didn't mean to imply maintained by a business or some other sort of gatekeeping. Though I can see how it read that way, and I appreciate your comment on it.
Instead I just mean things like the above, which other projects are getting right. They just don't yet have the feature set of Auto's project.
It'd be one thing if the dude was just prickly to work with or something, but it's unprofessional behavior by even the lowest standards.
-
InvokeAI on the other hand is being more open with their development, and making good effort towards having some sane licensing setup.
I'd say even just separating the project from personal accounts with racially charged game mods goes a long way for professionalism, but maybe you'd say that's gate keeping?
UI is subjective but it's not hard to call InvokeAI's more polished since again, the bar is so low with A1111's interface
> It sounds like you’re proposing that someone should create their own fork. That’s always an option, and I encourage everyone to.
That is actually not an option as there is no license on the first-party code in this repository (the only licenses I've seen being of a few third-party dependencies listed in html/licenses.html), so it isn't even clear to me you can legally clone it ;P... forking it would seem extremely problematic.
> If you set your pages and repositories to be viewed publicly, you grant each User of GitHub a nonexclusive, worldwide license to use, display, and perform Your Content through the GitHub Service and to reproduce Your Content solely on GitHub as permitted through GitHub's functionality (for example, through forking). You may grant further rights if you adopt a license. If you are uploading Content you did not create or own, you are responsible for ensuring that the Content you upload is licensed under terms that grant these permissions to other GitHub Users.
https://docs.github.com/en/site-policy/github-terms/github-t...
So while you can "fork" it, it's not forking it in the FOSS sense.
Edit: As far as I can tell, the repo maintainers are far too busy writing code and creating to care what anyone does with the code. It will change 10 times in the next 8 hours.
> You're under no obligation to choose a license. However, without a license, the default copyright laws apply, meaning that you retain all rights to your source code and no one may reproduce, distribute, or create derivative works from your work.
Yes, it does note immediately below that "if you publish your source code in a public repository on GitHub, according to the Terms of Service, other users of GitHub.com have the right to view and fork your repository". But that doesn't mean it's open source. You probably can't/shouldn't create derivative works with your forks. And you definitely wouldn't be allowed to just use this code in a commercial product, for example.
[1]: https://docs.github.com/en/repositories/managing-your-reposi...
Irrelevant. You're quoting docs.github.com, which is neither authoritative (docs are written to GitHub's documentation standards, not by their legal team), nor is it always up-to-date or well informed for the non-legal matters where it ventures to comment about the state of GitHub's products/services.
> that doesn't mean it's open source
Strawman. The question is "can you fork/copy?", not "is it open source?". The answer to the former is "absolutely yes, unequivocally". Anyone saying otherwise is wrong.
I like how you realize exactly what they're proposing, but proceed build a strawman that the comment didn't propose so you can forcefully tear it down.
In general it doesn't seem likely that in the long term AUTOMATIC1111's project will sfay the defacto SD interface.
We're already seeing more polished efforts like InvokeAI and various desktop clients, the field just moves so quickly that of course, the least structured interface with little to no interest in licensing concerns and the like is able to integrate new features very quickly.
But there will be points where things stabilize, if only temporarily, and that's where we should see alternatives start to make ground.
I meant to explore the logical possibilities, but in hindsight that does sound cheap. Apologies. For what it’s worth, all of the recent AI art deplatforming has me a bit jumpy. It did seem like a possibility that people might call for certain repos to be taken down, but I meant to phrase it as more of a question.
In fact, my comment reads as a cross-examination, which I’m disappointed by. I’ll try harder to write more thoughtfully.
As for automatic, no doubt he’ll be surpassed someday. But ML is one of the few remaining “Wild West” areas of code, before everything has been productized and polished. I hope he gives the competition a good run.
Same. It's been a thrill for me as a programmer to finally have an augmentation that allows me to do art and express myself creatively. That's perhaps why I'm so adamant on calling out the racist associations of Auto's repo. The last thing AI art needs is more bad press. But it's also why I was careful to point out that I wasn't criticizing Auto's work, as I'm not for removing code except under extreme circumstances. I use the project every day. I just want something better.
I also don't particularly have strong feelings about Automatic1111 as a person, since they're quite young from what I understand and I was also edgy when I was younger.
Perfectly suitable for a project about AI content generation. Every related submission also has at least one comment talking about the legality of it.
Or a "PeacefulProtests" rimworld mod that "adds George Floyd" as a storyteller. It also adds an all-black tribal faction called "peaceful protestors" (with a BLM reference in the source code) that has a high chance of addiction, adds a status effect called "socioeconomic status" that disables all work abilities and creates recurring raids by the aforementioned faction with in-game messages announcing them as peaceful protests and friendly visitors and even showing their names in blue (i.e. friendly).
No matter your political leanings, it should be obvious what the joke in this mod is supposed to be. Combined with the "WhitesOnly" mod this certainly gives an impression.
This has been bugging me for a while. The AI community is morally and ethically bankrupt. Whether it be OpenAI scraping the shit out of FOSS code so they could sell it back to us, Clearview AI scraping people's Facebook photos to train law enforcement tools, or whatever. It's all a "someone else's data" play. At least Stability AI released their model weights to the public that they scraped data from.
The whole field is being coopted by grifters, too: you see plenty of videos on how to make "passive income" by reselling low-effort AI art on print-on-demand sites. Yes, it is possible to make actual art with AI, but you need to actually spend time with inpainting and prompt engineering to make it work. This grift is not that. It is literally just "type prompt, spam result, receive bacon". Think like how Amazon marketplace sellers turned Amazon into a shittier flea market.
The underlying technology is hilariously exclusionary, too. No we don't have training code for our "open" AI, and if you had it you couldn't use it because you need a rack full of A100s. Let us also not forget that neural networks are basically proprietary software's final form: perfectly functional piles of weights with no extant source code, no way to explain what they're calculating, and basically no reverse-engineering tools. Even in the Actually Libre AI case where we have model weights and reproducible training code, we still can't verify that an AI is doing what we told it to do[1].
I'm not exactly a believer that technology is cursed with the values of the people who created it. You can definitely wrestle good technology out of the hands of bad people. But the converse is definitely true: people who have terribly shit values will curse the technology that they use. Or, more specifically, they will gravitate to whatever they can bend to fit what they want. And it definitely feels like AI art is attracting exactly the kinds of people who don't create, only consume.
[1] AI safety experts call this the inner alignment problem.
I wouldn't go so far as to say the AI/crypto communities are full of white supremacists or nazis but there's certainly a cultural underpinning there that is more welcoming to them than in other communities. That said, it's not exactly like tech spaces in general are doing a good job at keeping them out, especially ones celebrating "apoliticality" to a fault by regurgitating hacker culture myths about meritocracy.
But yeah, I strongly believe that you can't separate the means from the ends. Human development is ultimately about people interacting more than technology. You can only introduce a technology designed with bad intentions to a community without poisoning it only when that community is innoculated against it and maintains its vigilance. But some technologies simply aren't worth salvaging because they're manifestations of the values they were designed with.
Luckily I think that applies more to crypto than to AI, where the damage it causes is largely a consequence of the economical system we exist in rather than the tech itself. Namely, creators not being compensated for the creations that feed into the AI and jobs being lost to cheaper products generated by the AI.
[0]: Noteworthy case in point being that many widely known NFTs are procedurally generated using templates created by artists who typically do not continue to benefit from the NFT sales themselves and that there is a massive problem of art from sites like Devian Art ending up as NFTs without the creators' knowledge or involvement.
[1]: e.g. https://cointelegraph.com/news/new-video-revives-debate-over...
Crypto is a technology cul-de-sac: it consists entirely of solutions to problems that only exist because of it's odd choice of problem domain. i.e. if you weren't trying to make a payment system for organized crime, then you wouldn't need to deal with Sybil-resistance. So it effectively counters my belief that you can wrestle good tech out of bad hands, because it's not actually good tech.
I recall seeing something on Mastodon a while ago that I don't quite remember the source of, but I can paraphrase as "left-wingers tend to create social spaces and art, while right-wingers tend to consume and destroy them". While I absolutely despise this sort of one-dimensional thinking, there is a kernel of truth to this. Or, in other words, Hitler failed art school for a reason.
Art is, first and foremost, a language used to communicate ideas. People on the far-right[0] do not have a whole lot to say. When listening, they tend to get caught up in superficialities and ignore the underlying idea. Their goal is not to speak, but to prevent speech[1]. So naturally they would be attracted to technologies that let them "gussy up" their nonsense into something that "looks" artistic.
This even goes for the not explicitly far-right stuff, too. After all, to them, the purpose of art is to look pretty, not to say things. That's why there's such an odd focus on overdone cinematic aesthetics in AI art - because it "looks pretty". There's tweets of people saying that artists should just give up and submit to the machines because Stable Diffusion has already "perfected the female form". What they meant is that it generates sexier porn and panders to their tastes more. The AI does not communicate ideas with its art, and if it did, they would reject it and delete it.
[0] I am specifically talking about the right-authoritarian corner of the political compass. Right-libertarians can put down their torches and pitchforks now.
[1] This is also why I do not shed tears for them when they get banned from online spaces.
There's so much icky stuff going on in the generative art community and no one knows just how deep that rabbit hole goes.
E.g.: https://github.com/Kahsolt/stable-diffusion-webui-sonar
(See bottom of page).
To be fair, there is substantial interest in using things like Stable Diffusion to generate pornography. There are several projects for hentai, etc.
There have also been large threads on /g/ teaching people how to generate hypernerworks with suspiciously young looking subjects since the beginning. The "icky stuff" has a disturbing footprint.
Wanting to making pornography is perfectly normal and healthy; that's not something I consider icky. When I quoted "icky stuff" I thought we were talking about the generation of underage subjects.
Have you seen ANY legal age subjects? Subjectively, how much fraction of anime pictures do you consider to be portraying individuals able to consent?
I mean, if there was an NN code that is returning 0.0120000 for 85% in inference data but ranging evenly between 0.0 to 1.0 in training, intuition shouldn't be that it's the reality that is broken.
But it kind of is depending on your definition of icky. The fact that they've trained the models to be extremely proficient forgers should raise all sorts of red flags. I get the concept of homage and every great artist borrows/steals, blah blah, but as these things continue to get better, the lines between legit artist work and forgery will only blur to unknowable. Digital artists will now need to know how to cryptographically sign their work. There have been tools for this that most ignore because of pricing. (and no i'm not talking about NFT nonsense)
It's in my self interest to not see icky things.
It's like going to an artists twitter and then complaining that their art style is garbage and that they should change it. No one is forcing you to consume that person's content.
Not all content on the internet needs to appeal to everyone even if everyone is on the internet.
No one is forcing him to use GitHub.
They'd rather lose a few Automatic1111's than have schoolteachers telling students "Don't go on GitHub, it's not safe" (or having school administrations ban its use).
It's not a "normie" take that we shouldn't sexualize children, even if they are animated.
I was expecting some of the horrible stuff I’ve seen on 4chan, not something you might see in a mainstream (bad) anime.
If anything, it looks like the AI cleaned it up and made it more presentable.
Well as many have said years ago, private companies like GitHub can censor anyone at anytime and point at the ToS no matter how vauge, opaque it is. They just have to give you notice, and they can still do it again to anyone else.
There is nothing shocking about this. What is really amusing is that people continue to expect a centralized version control management service to not enforce 'censorship' on their own services and products and when they do, everyone here is throwing a tantrum about it.
There is always an option to self-host your own GitLab, Gitea instance if you don't like what GitHub is doing.
That said, I hope this whole debacle has shown a light on the potentially problematic nature of having the Automatic1111/stable-diffusion-webui project be the go-to UI for using Stable Diffusion. Not just because of the viewpoints implied by Automatic1111's other repos. Automatic1111/stable-diffusion-webui is a mess of a codebase and while it has the largest plethora of features and plugins in the StableDiffusion ecosystem, it's also one of the hardest tools to use. InvokeAI has a much friendlier UI, for comparison. Plus Automatic1111/stable-diffusion-webui has no formally published license, making its codebase a tangled web of legal nightmares.
None of that is to degenerate the work of Automatic1111 or the hundreds of contributors to his project. And I'm for the most part not a fan of "censoring" code. But I hope we can agree it's not ideal to have such an important repo side by side with a "WhitesOnly" rimworld mod.