Oh! A new one is finally up. It's nice to see actual progress again, the art of slop stagnated into piss filter hell for a year. Though ever since companies stopped releasing their models publicly, local genning has been halted. Are there any local video models? It's all web apps and corporations you have to work through.

Can't wait until the Internet is completely fake and unusable.

>>1823956 Twitter already is, 99% of all posts are just bots. it;s a view of whats to come.

>>1823956 The post-truth world is going to have a way different effect on the ZOG's control than they think it will. Once fake videos indistinguishable from reality begin to proliferate, no one will trust a single bit of propaganda they create anymore.

>>1824015 That goes both ways, sadly.

That pickleball video has a weird origin. It's from some scam group that's trying to promote their own pump and dump scheme? It's very memestock adjacent. >>1823956 To be honest, we've been progressing towards this path since the photograph.

>>1824015 People will either not believe anything or they'll relegate truth checking to some authority figure since they can't trust their own eyes anymore. Normalfags will probably do the latter.

>>1824015 >post-truth world All that shit sounds scarier than it really is. What it's really going to do is just reset the clock on public information. You won't be able to trust anything you don't see with your own eyes, and which happens in your immediate area. For everything else, just pick a source you like and run with it. Poor Kentucky farmers 150 years ago had no idea if what they were reading in the papers was true, and no way of verifying it. People made all sorts of wild claims, and you just believed them or you didn't. Hell, even 40 years ago, your options for verifying information was limited and large media institutions had nearly a monopolistic control on the flow of information and culture. AI will probably end up leading to a renaissance of these legacy institutions, who by their sheer size and reach will be seen as having the connections and the influence to get to the "truth". We had a brief peek behind the curtain thanks to the Internet and the free and open flow of communication, but that time is almost over. We're going back to the status quo. At least, unless there's some way to figure out definitively whether or not something is AI generated no matter how good it gets. It's a long-shot, but these things do happen. I mean, they've been able to fake DNA evidence for over a decade now after once being considered the gold standard of evidence. In large part it still maintains that gold standard because they discovered patterns of discrepancies which occur between naturally formed and lab created DNA profiles. Maybe there are also patterns in the way that AI media is generated which can be detected, even when they appear indistinguishable to real recordings to the naked eye.

>>1824043 >since the photograph Yeah sure, whatever the fuck you say. You going to tell me fairies aren't real either?

>>1824179 You peaked my curiosity, how were these pictures made? It doesn't look like plain photoshop (although it could very well be), was there some kind of technique back in the day that allowed for things like this? This is interesting...

New thread? Good timing, I grew whatever balls needed to try out Forge Couple. I imagine like AI itself, actually making anything elaborate will take some heavy experimentation in regards to which prompts go to which mask but if I ever feel patient I may know where to go if I want certain compositions to be just right. Or close to right. Any and all pointers are heartily welcome! So far all I got are https://github.com/Haoming02/sd-forge-couple?tab=readme-ov-file and https://desuarchive.org/g/thread/106240788/#106261868 Primary concerns is keeping to a singular art style, possible LoRA limitations (using two radically different character ones caused 3D to overpower 2D), how strong the global mask is, and God knows what else. I'll get back to practicing inpainting Soon™.

>>1824178 >150 years ago The age of objectivity began with the printing press, but it ends now. And yes, "resetting the clock" to the death of objectivity is itself the terror.

>>1824196 I believe they were early attempts at double exposure photography. It was a lot harder back before you could just manually wind your film back a single cell.

AI has ruined regular porn for me but sometimes I get too lazy to prompt. There should be a plugin that randomly generates prompts for you after you set some guidelines for it or something.

>>1825057 Soon there won't be any porn. Man i can't wait when we just blow up venus out of orbit. So many days, so much meta, so many happenings, religious, constant fight against chinese ai, mafia fights between cults, exposure of evils of cabal and their soothsaying pipe bombs, fall of the pajeets with disproof of absolute, change of overton window, death of liberal way that just recently seemed like a scary weimar future, and anon... anon still can't stop being horny. All that while some ai program is ddosing psy-ops daily through real humans, instead of bots, psy-op farms of psyop-ed people.

OK enough with the doomposting fearmongering. Post cute pics - the true essence of AI!

>>1824196 There were techniques, such as painting over film, like the early colored movies(yes, they had to hand-paint each frame) or how they did in the first Kin Kong movie, where they would project the people in such a way as to appear small, then film the "screen". Or maybe they just did some cut-outs and placed them in the frame to be photographed. Or it could have been forced perspective, like they did in the first LotR Movie, though I am not sure about this last one.

>>1823904 Can it be hosted locally? what are the requisites?

>>1823904 >make a new AI General >specifically about an iPhone-only app with niche use & only a handful of examples gathered from elsewhere >zero ties to the previous thread

>>1824196 The second one's double exposure, you can tell by how translucent the fairy looks. Not sure about the first one. >>1825979 Slop OP for slop news. I hate how dead self-hosted is and how cucked SaaS is. Even Tensor.art got Mastercarded, so you can't use it to prompt tits. >>1825846 >Hosted locally >2025 This fucking guy

I'd love to do local gens but man I do not have the money for a PC that can.

>>1826019 It's not even that computers couldn't do modern local gen, it's that the major players stopped releasing anything to the public years ago. You'd never see something like Stable Diffusion released today.

>>1824178 Have you been living under a rock? People abandoned reality a long time ago in favor of turning "fake it till you make it" into an ontology. Just look at who's in the white house. People already do believe anything they see, especially when they want to believe it. That ship sailed a long time ago.

>>1823904 There's an ai board for this.

>Go to the other board to die slowly like the rest No thanks.

I order 30 AI midna impregration videos. Move it, Chop chop,

>>1829350 >Don't use the other boards >"Why is /v/ the only board with traffic?" Self fulfilling prophecy

>>1829419 >/v/ >traffic

>>1823904 Too bad they gatekept with content violation. I wonder if someone already downloaded the model

>>1824022 Is this real?

>>1829605 You know with myself being new into imageboards, its really a shame that it seems like most traffic (atleast on 8chan) is geared towards the fringe topics where people specifically come here running away from other websites, while 'normal' topics are dead in comparison. Wish it was more lively across the board.

>>1829419 >scream into the void >NO YOU DON'T UNDERSTAND YOU JUST HAVE TO SCREAM INTO THE VOID EVEN MORE lol, go away idiot

>>1829720 4cuck faggots had a chance to leave their shitty fed-infested censorship-ridden website but a large majority of them decided they preferred the weight of the boot to the freedom of 8chan.

>>1829762 "Be the change you want to see in the world", nigger.

https://civitai.com/models/1901521 Pony V7 is finally out, but currently on-site gen only. Supposedly will be up for DL in a few days once everything is converted to proper formats and such. Like all base models, you want to wait for checkpoints.

>>1830156 It has such an "AI art" look. That mashup of everything on ArtStation. It's a shame so much slop looks like this, because it's perfectly capable of more diverse styles.

>>1830161 Literally the baseline of a brand new model with no fine tuning and only minimal style prompts.

>>1829239 >>1829419 >There's an ai board for this. /ais/ is for old /v/ ai threads. I guess it makes sense if you count their hentai diffusion general. They all type in bean though. >>1829720 >Wish it was more lively across the board. I am stubborn about using different boards and sites even, but me too.

I'm curious if ai has gotten better at replicating a style with fewer pics now?

>>1830374 How few pics are you looking at? Flux only needed like a dozen for a pretty broad one. Even Illustrious can get by with 30 something if you're only trying for a narrow focus (e.g., character art on white background, portraits) instead of wanting to be able to do anything. I've done more experimenting with styles and found "only" 100 something does decently well for some things

>>1830405 got any examples of stuff you've trained on specific artists using those numbers?

>>1830417 All of these should list their dataset quantity or have it on the page for the Pony original (which should be linked). If one is missing tell me so I can fix that. https://civitai.com/user/NanashiAnon/models?tag=style&baseModels=Illustrious Fallout Talking Heads and Dawn of Souls style both only use 38 images, but work well for their narrow focus. LucasArts style used 54 images and draws some scenes perfectly, but has no idea how to do others (I think this one is also highly impacted by dataset contents. It skews heavily male, and the females present are either butch bikers, a kid, or a woman in a spacesuit). Arcanum portrait's 76 is similar (plenty of female portraits, but data is low quality and I had to correct auto-label a lot on them), but I think it's because I don't know a good checkpoint for the semi-realistic style. 104 for the Carmen Sandiego style (once I fixed the aspect ratio of the TD) and 126 for Code Lyoko (once I tagged the eyes) both worked pretty well. By contrast, Golden Sun official art didn't work too well with only 32 pics and 11 of those non-character pics, but it still does something things relatively acceptable.

>>1830497 I'll be honest anon i don't know enough of what you're talking about so you'd have to break it down a bit more, which image is trained on who and what?

>>1830562 The pages for the models with more details on their training data is on the linked page.

>>1823904 Some funny Sora2 videos that I found online.

>>1830735 ah, are those models you trained personally?

>>1830750 >garfield has two tails What this garfield was training for? To fly and carry blue hedgehogs?

Can Sora2 make image to video?

>>1830750 #4 just made me realize how easy it'll be to fake game leaks. Especially if somebody puts effort into it, just make an AI video of gameplay, then put it on a screen and record that.

>>1831963 Sherma is a guy, though.

>>1830800 Yes

>>1831986 Hornet made his asshole pregnant

so did someone manage to find unrestricted Sora 2? Honestly we're going to see some good anime slop soon.

>>1831380 Yes, if you feed it character references as the first image. I managed to create videos of some beloved OTPs of mine, but I like to edit them with fitting voices so I'll be sharing one for now, Sora and Moroha at a festival! And after that, a bonus Mew video! >>1833999 Kek my sides. Nonetheless I'm having fun with Sora 2 even despite all the restrictions, it already created clips of my dreams that I'm loving a lot. Local version would be a dream to have.

>>1834884 If avariable, how much would it cost? And I guess there's nothing locally hosted equivalent. To make Shadowrun videos or any other silly ideas made by my very poor drawings.

>>1831986 >>1832078 (⚆ᗝ⚆) I did not know. Horny blindness.

https://civitai.com/models/2024752/il-cynthia-aurora-cindy-vortex-jimmy-neutron-jimmy-timmy-power-hour-style?modelVersionId=2291524 Cindy Vortex, but it's only the Jimmy Timmy Power Hour version. I wonder what she'd thinks of Timantha.

>>1836021 A ton of people draw the knight and Hornet fucking, despite being siblings, so I guess a lot of people just make porn of the characters with zero context.

>>1836165 >I guess a lot of people just make porn of the characters with zero context Wasn't exposing such the original intent behind Bridget?

>>1836165 >I guess a lot of people just make porn of the characters with zero context. Touhou, Blue Archive, and Genshin Impact in a nutshell.

Isn't sora 2 only available on apple right now? fucking miss me with that gay shit also im surprised nobody has genned lewd stuff with it yet, once you can do that a lot of porn "artists" are going to be out of jobs

>>1836393 >a lot of porn "artists" are going to be out of jobs That is a narrative I haven't heard in ages, what's up with that? I remember everyone panicking about AI like it was seriously going to take artists out of their money... and somehow I am still getting paid to draw.

>>1836427 Well technically people have been putting money into people who gen ai now instead of porn artists. Seen it floating around so while you may be safe it's not like it's not happening at all

>>1836393 Given that all image generation models do is steal actual art from actual humans, artists won't be out of anything any time soon.

>>1836433 >putting money into people who gen ai now instead of porn artists. Is that really the case, though? Cause the way I've seen it, people who pay for Human-generated art and people who pay for AI-generated art seem to be different pools that don't quite relate to each other, so neither of them are being radically affected by each other. I mean, I know first hand that the average commissioner and the average proompter are two different individuals with different priorities. ...then again I'm talking as a SFW artist that takes SFW commissions, so the brain of a coomer and perceive and pursuit art art is beyond me. >Seen it floating around so while you may be safe it's not like it's not happening at all Oh, for sure, I just mention myself being safe because, judging by the scenario presented by many lunatic peers of mine, me being safe at all is a major contradiction, specially when I'm no one particularly special and I have nothing that should guarantee my survival lol.

>>1836445 How much do you need before the ai can create anything smutty though down to majority accuracy? it's an interesting thing, we're on the threshhold of just some star trek type shit where you put an artist in and boom there's a copy of their style and it's getting closer and closer >>1836454 For what it's worth, i don't want anyone going out of work who genuinely needs the money to survive, that would depress me so i'm kind of like stuck in the middle on that because i know a shit ton of third worlders need that smut money to just have some food for the week.

The talk of there being an AI bubble has boiled over into mainstream news. https://archive.ph/rwqDX What I find interesting is even Sam Altman is speculating there's a bubble going on. He apparently believes "silly start-ups" are getting crazy money from bad investors. AI spending will reach well over a trillion dollars before the year is out, a completely incomprehensible amount of money. A more worrying fact is that AI spending is the main thing holding the US economy up right now, America would actually be approaching recession if not for AI hype. https://archive.ph/xP9XH What confuses me is how any of this is supposed to make money. Everything I've seen of AI, and by that I mean the generative neural networks such as LLMs and TTV broadly plastered as "AI", have been toys; search engines at best. There's no obvious path to profitability. Any interrogation of this is vaguely hand waved away by the tech industry with "AI economy".

>>1837445 >What confuses me is how any of this is supposed to make money that's the fun part, it doesn't, exactly the same as youtube, reddit, and most sites as a mater of fact

>>1837445 >What confuses me is how any of this is supposed to make money. It's not. That doesn't matter. The purpose is to total control of access to all information, and therefore purposely restricting the range of thoughts and actions to only those deemed acceptable by the authorities.

>>1837535 Kojima was right..

>>1837445 Free quality testing using the public by seeing which images/videos the users seem to focus more on, which is used to refine the AI further toward successful results? It's possible that we'll eventually see these systems paywalled in the future, which then would bring in profit.

>>1837445 >What confuses me is how any of this is supposed to make money. Way to bring up flashbacks to the dotcom bubble. >>1837854 >It's possible that we'll eventually see these systems paywalled in the future, which then would bring in profit. NovelAI and most of ChatGPT already is, though I'm surprised there's no yearly option for the latter yet.

>>1837445 >What confuses me is how any of this is supposed to make money. They think they will be the 5% that got profit, or that they actually know how to use AI, unlike the vast majority, who didn't know how to use it. https://archive.ph/THJsc >MIT report: 95% of generative AI pilots at companies are failing >the vast majority stall, delivering little to no measurable impact on P&L

>>1837445 >What confuses me is how any of this is supposed to make money. Its not, the objective is to create infrastructure and condition the population into accepting AI before the actual AI in the 3033.

>>1837445 >What confuses me is how any of this is supposed to make money I believe you're starting to understand why economists are calling it a speculative bubble. The same thing happened in the early 2000s with the internet called the Dotcom Bubble, start-ups with three employees could call themselves something stupid like "Pets.com" and get a billion dollar evaluation.. Ask what any of these .com companies did or how they made money and you were flooded with buzzwords like "information superhighway" and "global village", but no actual answer. In my opinion, what's happening with generative AI is the public seeing a neat and mesmerizing technology and jumping to "THIS IS REVOLUTIONARY THIS IS GOING TO MAKE SO MUCH FUCKING MONEY", erroneously equating being alien and unprecedented with massive potential profit. This reasoning probably stems from the last few major revolutions in technology having the same initial wow factor, like smartphones, the internet, and even television. Initial exposure to generative AI is surreal and almost like sci-fi, ergo investors believe it's another paradigm shift. All generative AI is good for is porn, Indian sloptent, and disinformation.

>>1837445 >AI is a bubble Sure, but that doesn't mean it's going away or that it isn't going to take over pretty much every industry. The internet was a bubble too, and lots of people lost their asses when that bubble burst. Yet here we are. >There's no obvious path to profitability. Don't most of the big commercial AI providers charge a subscription? Or, like CoPilot, Microsoft will use that to leverage other products which do make them money. I could definitely see a future in which Microsoft pushes CoPilot tools on game developers to provide hooks into the games which allow for one click configuration to the most optimal game settings (as well as some level of troubleshooting problems) across a near infinite array of hardware configurations. Something to make Windows gaming as dead simple as console gaming to make the transition as easy as possible to casuals. Then restrict that feature to games bought in the Microsoft Store or via Gaypass.

>>1838388 While I think it's apt to compare this to the Dotcom Bubble, I don't think it's accurate to think AI will have remotely the same effect as the internet. It'll definitely stay around but not as the world-shifting technology it's being billed as. >subscriptions That sounds like it makes sense, right? Except the amount of investment in AI year to date is more than five times the entire subscription software market. It would take over 30 years of sustained AI subscription revenue to pay off the investments made this year alone.

>>1837445 AI is being very useful at boosting my code production. So much that at times it's almost all AI what I put out. From that side I can see that making money by boosting development, and there also a lot of other more simple uses like graphics for powerpoints, fast release of concepts, among others. Is it speculative? Yes, will many businesses fail? Possibly, but there's no denying that consequences will never be the same

>>1836165 >>1836384 I thought he was too cute to be a boy or something.

>>1836165 >I guess a lot of people just make porn of the characters with zero context. With Hornet and the Knight their relationship is pretty well known so I'd bet at least 90% of the porn is drawn knowing the context.

>>1830405 >Even Illustrious can get by with 30 something if you're only trying for a narrow focus (e.g., character art on white background, portraits) How many for backgrounds and full body?

(Will go live in less than 2 hours, posting now because I'm posting on the subject) The discussion on what's needed for a style and mentioning Flux made me decide to try doing a style in Flux. https://civitai.com/models/2028451/flux-golden-sun-official-art?modelVersionId=2295710 Disclaimer, I have no idea how to prompt for Flux, which will likely hurt the dataset and gens. >>1838750 You can do full body as a "narrow" style LoRA (in-fact, official game art is a large chunk of such), it's doing it on a background or doing "exotic" things when only trained on humans when not having enough data on that.

>>1838774 I'm torrenting an anime in preparation of making a LoRA that I'm surprised doesn't exist yet. Not so much considering a style LoRA but a character one. I'll either follow your guide or the other one*. Whichever doesn't outright require I finagle with some image editing software. Not like Photoshop's easy to pirate these days. I also hope to learn inpainting (or controlnet or Forge Couple) well enough to turn the funnels in pic related into Haros. *and to clarify from the previous thread, I was asking how the other guide's methods compares to your own? As in, is there anything that it does better or worse.

>>1838811 For most LoRAs you'll only need very basic Image editing consisting stuff like cropping and clearing stuff off plain backgrounds (e.g., text from white text bubbles or stray pieces of alternate views, like pic related demonstrates), and maybe some color correction. You can literally do that level of stuff with something like MS Paint or (for Linux) KolourPaint. Beyond that you should only need to be done if you're really desperate for data and working with a dataset that's effectively "every image of this character that exists", and that should only be the case for super minor characters (but I do like doing such characters). For more advanced editing, GIMP with a basic plugin set (I think it's "GIMP Resynthesizer"?) is plenty enough. As for the comparison of guides: I can't comment on the steps for setting up local training but for the rest most of it is either very similar, detailing the author's preferred method for data entry, or just telling how to use specific programs that creator uses. I think it really underestimates the power of just tagging "solo focus" (especially when the BG extras are opposite sex, don't overlap in tags and/or aren't drawn as a "real" character) instead of bothering to remove characters from BG. I use data with stray shoulders and legs all the time and it's not an issue. I also think "Don’t confuse dark scenes with night-time shots." isn't really needed if you tag "dark" and "night" as applicable.

>>1838900 (In case it wasn't clear, the black boxes I added to that pic are suggesting how you'd crop each view to its own image. After that cropping that, just select the extraneous bits like parts of other views or the name plate and hit Del and save as, then reload/ctrl+z to repeat for next view)

>>1836445 >muh stealing I've never seen the art that AI created for me anywhere else. >b-b-but it stole the style So does every human who gets inspired by someone else, which is 99% of them.

Requesting Trump and Haltmann sucking money out of anons pockets as Susie hands them all "Fell for it again" awards https://inv.nadeko.net/watch?v=5fTv9ThJg6U

>>1839047 >muh stealing You fundamentally do not understand how these things work, and thus your input is irrelevant and ignored. Fuck off, retard.

>>1839047 I mean, if you want to be like that... >I've never seen the art that AI created for me anywhere else. That makes zero sense when most AI image generators operate under base algorithms (LoRAs?) by fundamental. You're going to tell me your funny box learned how to do pretty pictures on a certain style by itself. It's not unique when it already borrows the processing from a database accessible by several other users. ...and that isn't even accounting the fact there're a lot of base models out there copying other artists' styles, like Rizdraws', or that one Ranfag who's name eludes me, I've even see some in the artsyle of Shigeshisa Nakaue. It is irrational to think people out there aren't making models of artists that already exists out there. >So does every human who gets inspired by someone else Yes, that is stealing and yes, it is a valid artistic practice... when you know it's different from plagiarizing, which is the proper word for this specific context in the AI discussion. And no, that is wrong for both humans and the AI to do, even worse for the latter because the plagiarist is not as simple as just an individual, but a whole ass database financially supported by hundreds of people who's end goals and moral principles can be easily put to question.

>>1836558 >For what it's worth, i don't want anyone going out of work who genuinely needs the money to survive, that would depress me That is a sentiment I can appreciate and commend. Cheers, anon. If it serves for anything to you, one thing I learned from hanging around commissioners and doing commissions is that there is more to the process than just getting your stuff drawn. Part of the reason people commission art is because of the patronage and the satisfaction of having another human hear your autism and do said autism justice, as well directing them in the process at all. There is also a bit of altruism and ego to it, some people just want to be supportive and say they're doing good for someone. There's an inherently human aspect to art that the AI simply cannot replicate, not because it needs more training or anything, but because what it lacks goes beyond computer logic, it's human sentimentalism and foolishness. As long as there are people out there that value that foolishness, art and artists will not go anywhere anytime soon.

>>1836445 >>1839420 >actual theft of actual humans where the actual art is still intact He's right though. Perhaps this is instead a failure of the english language since there's not a proper word for it. What you typed sounds retarded. The act of image diffusion with artificial intelligence is not stealing. >>1839447 Plagiarism doesn't fit either. Rizdraws did not and won't make this image thinking it was funny for the resulting image to be horny instead.

>>1836427 Its a shite narrative, is what it is. You draw something truly novel? Fantastic, you're ahead of AI by at least thirty years. The current and even future generation of artists is still secure, as long as the artist has a functioning brain, because AI fundamentally can't create anything wholly unique. It must be trained (and learn) off of something, and the best it can do, to make something new, is to mix and combine. A lot of porn addicted nigs get all uppity about AI being able to cater to their specific fetish. No, not really. You're getting worse quality (at best its the same) and the best you can say (and have said for the past half decade) is that "it's really getting there!" Off the top of my head for an example, Khyle has made some really funny work with various characters that AI just can't do. Regardless of your opinion the quality, I think parodists like him have a pretty safe space away from AI for awhile at least. >>1839447 >base algorithms (LoRAs?) by fundamental. Technically you didn't actually address his point. Yes, the base tags exist elsewhere. But a specific composition? I don't think there is many non-AI images of Saria flexing her wrinkly butthole for actual models to take advantage of. But there is images of bum wizardry and also images of Saria in various positions. Just because others can easily access the same toolbox doesn't mean there isn't some uniqueness present. Styles are more similar to what you're talking about. Except you can mix and match those, too. Funnily enough, AI has actually gained a somewhat original style. The overtly shiny AI slop style that looks like shit because its the result of a billion styles combining into one. >plagiarizing Reminds me of that dude who took a WIP image from a twitch streamer and "finished" it in AI. Tracing is poor professionalism, but its done so often even in successful business contexts that I can't take any argument with it seriously. Anyway, plagiarizing is not the proper word for this discussion since we're not dealing with art theft or tracing. AI art really just doesn't trace, unless you feed it an image, drag down the noise slider and let it run.

>>1839522 Would any anon hit that?