「Generated content thread」
Anonymous
is it chill if I chill here
I think this is AI. It may be drawslop.
Maki!!!
some of these posts give me nostalgia
>>105390 #
>memetic warfare general
it might have been created on 8chan for all I know but it started in 2016 on /pol/ from what I can tell, and was revitalized when the AI memes took off in 2023
sadly Bing Dall-E, the only API really capable of doing actual memes, was quickly censored and lobotomized
and thus the threads were mostly abandoned with occasional resurgences when China would release something like the Hailuo Minimax which was responsible for lots of the early AI video slop memes prior to Sora 2
Fortunately, in this economy, generating anime girls is far more lucrative than trying to sway elections with memes
>>106175 #
Would you be willing to make another thread with a basic guide for local image gen?
>>106179 #
djb. I figure this board will get more interest if it has more than a dump thread on it.
>have demonic powers to copy skills
>learn everything there is to know about weapons
>become top marksman in the mortal realm
Finally, somewhere i can chill out. Seems A.I will always be a huge taboo, everywhere.
Oh yeah so, what's the vibe of this board? Any NSFW rules i should be made aware of? Should probably populate your OP so more people stop by.
>>113654 #
Admin-chan will become upset if /ai/ becomes blatantly pornographic.
OP was previously populated, but autism resulted in its deletion.
>>113654 #
Hi, I'm the board owner.
Just put a spoiler on anything overly pornographic or NSFW.If you don't, I will.
If it's utterly repulsive I'll probably just delete it.
Otherwise, have fun!
Had to reinstall my UI, lost all my prompts.
two hours later...
WE'RE BACK!
>think my gens are starting to look good
>head over to /ldg/
>realize am still a promptlet who can't use a computer
Wow there were a lot of details I missed in the OP image
I should have spent this much time tweaking the prompt sooner!
>>116604 #
>Though is there a reason the swastika's on her back?
Yes, it was the best 3d slop I could find in the /ldg/ thread I was stealing images from at the time
Is this thread for full gens only or do ebin mene edits also count?
>>116643 #
Post anything you want ITT, it's the general.we had some other threads here dedicated to specific things prior to the autism catastropheRIP
Alternatively though, if you're an expert on something AI related or want to dump anything specific feel free to make another thread if you please!
>>116654 #
NTA i was thinking about starting a 2D general, or a specific open source general, given there isn't one on the catalogue yet. I'm no experienced OP maker though.
I just realized this thread doesn't even say 'general' on it, hmmm
>>116688 #
I'm procrastinating or else I would have a better OP cooked up with resources and such by now, but seeing as this is a weebsite, the distinction for a 2d thread would probably be redundant seeing as it's going to be 99% of the content anyways.
>a singular for all of those
I might just ask admin to edit the the OP with more information since it's completely lacking
And I think the bump limit is like 7000 so I'm sure we really only need one 'diffusion' thread for now unless like I said you wanted to dump somethingmany anons complained on /g/ that there were too many specific AI threads that could have all fit into one, if you've been thereIMO you didn't have to spoiler that btw, as long as there's no nudity or literal obscenity>used to 4chan's strange nonsensical "rulesI'm generally going with their definition of NSFW (no nipples or crude sexual depictions) but the worst that will happen is I'll spoiler it myselfThe /homo/ board has somewhat strict policy on lewds so I want to offer a slightly more flexible standard over here
>spoilered half the post
fucking formatting
im from infinity boards if you cant tell..
>>116708 #
eh don't feel too bad, i'm used to my cuckchan addiction. i leave for a month or even months at a time but a.i news has me crawling back. + the social stigma behind a.i, kinda the only place i can shitpost/share. hoping to break it by hanging here permanently.
also not used to this live post chatting stuff, really cool (and scary).
>>116713 #
>stigma behind a.i
It's weird but the main selling point IMO is that if you ever needed to be super anonymous and untraceable you can generate your own memes and reaction images and even be an avatar fag without leading back to a paper trail in the archives or anything if it ever came down to that, at least regarding image generationlike if the jews ever crack down on trying to find all the most anti-Semitic 4chan users... you'd still be able to shitpost with impunity and dodge any profiles they may have built on you
>live post
also weird but you'll get used to it and it's fun especially when you're taking part in a dumb debate and you see exactly what they're typing before you finish sending your post LMAOit's also a pretty good way to detect bots or non-english speakers if we ever got attacked or raided>>116728 #>hoping to stick aroundYou found the /homu/ board right?That's currently where almost everyone poasts unless we're watching a movie>/homo/my bad it's not homos, it's homu as in homura, the site mascotanyways here's the main thread, you just have to change the theme to be able to read it!>>115066
>>116720 #
funny thing is i don't even mind the ligma around a.i so much, it just makes socializing more difficult. not that people weren't already making socialization impossible before.
which is another reason i'm feeling kinda fucking stupid for letting myself continue feeding on my 'chan addiction post summer 2025 after that embarrassing leak. It's like having my perspective on the state of 4chan in 2025 vindicated in real time, but also proving my own hypocrisies. Whoops! But anyway, alts are cool, hoping to stick around and watch these sites continue to be relatively niche and nice.
That's part of what i like about this livepost thing too, IF i do catch a strange post, i'll see it in realtime and start formulating my response for it.
>You found the /homo/ board right?
no and i don't think i'll even let myself click it, also i have no idea how to spoiler on her lololol
>using cuckchannel in the modern age
ﾇﾝﾃﾞ
lmao i unironically kept reading that as /homo/, yes i've been to homu. probably won't lurk most boards besides specific topics. but i'll check it out more in depth. >>116720 #
(by the way should i be keeping my post in live mode until we're fully done 1:1 talking or is there an edit function? boy it's strange being a newfag somewhere for once)
>>116744 #
i think i got the hang of it right away because there's that silly motherfucker over on /homu/ i instantly wanted to debate with and followed your post about how.
seriously cool system, very fun.
>I still feel like picrel thoughhowever
never compare kermie to the cuckchan down syndrome frog. though i guess it's funny i call apu down syndrome when this is literally someone's down syndrome kermit gen.
>fucked up the text spoiler formatting again
that's it I'm going to KMS myself
>>116734 #
There's no editing or code of conduct really so if you're unsure about something you can keep the post open (not sure how long) and open a new tab
there's also cancel posting, as in deleting everything once you message has been recieved
>frog post
EEEEK! this is sure to anger the oldchildren!
>>116734 #
The whole esoteric thing behind pepe memes and how it was used by glowies to generate support for Blump was actually quite disturbing to come to terms with when I arrived here and coincidentally discovered a schizo on youtube around the same time who explained what a huge psyop it was.
I still enjoy frogs though, they're cute and funny looking.
>>116757 #
>What's the schizo's channel
it's one of his earlier videos where he appeared to be recording audio with a phone but it's one of the craziest channels I've ever found
[YouTube] ???
>>116753 #
What's the schizo's channel name? I haven't looked into any of this in years, much of my knowledge drained out of my brain and got replaced with a.i coomerism since 2023.
One big thing i remember spooking me a little bit, was how eerily repetitive it all felt from big memes of the past. But these days, much of the commotion isn't even real people, just bots. Which lends further credence to the psyop theories. I mean it's LITERALLY manufactured you can't get more concrete than if the original founder of 4chan got hired by google.. :^)
>it's one of his earlier videos where he appeared to be recording audio with a phone but it's one of the craziest channels I've ever found
Cool thanks when my tame impala song finishes i'm putting this shit on right away. Its been too long since i went down a schizo rabbithole.
>It's the "trump may be the antichrist" rabbithole once again
holy fucking KINO this is a good one, thank you again.
Sorry for intruding.
>>116708 #
Here is the posting guide from the info panel, click the i in the top bar to see the rest. Also feel free to talk/praise/complain about the site on >>>/homu/. I check in here sometimes but not too often.
>>116753 #
>pepe memes and how it was used by glowies to generate support for Blump
I think part of it was making it easier for bad actors to fit in by creating a meme monoculture they controlled. If you have a big corporate pepe folder for shills to use to fit in and you astroturf normal threads with pepe OPs for long enough, eventually it looks organic when you see pepe OP threads that are posting ragebait or pushing propaganda narratives.
>>116821 #
Nail on the head there essentially. Not everything's bots or the same israelis making BBCposting threads. Sometimes it's hidden in plain sight.
Or it's a guy like Fuentes telling the young dudes to stop liking girls and start voting for Kamala Harris lmao.
>>116826 #
>marked image as spoiler
>couldn't continue my post
what did the forum mean by this? Oh well. was just gonna remark how it's funny i rarely ever gen anything in 2D anymore, and that new Anima model was my first time in years going nuts with trying different artist styles, and even a bit of mixing.
>>116821 #
>meme monoculture
It's just sad that we felt like we created the wave when it was really just hijacking and engineering by glownigs
>>116830 #
>new Anima model
It's not the best compared to some wizard-tier illustrious workflows but the fact you don't really need LoRas for anything is kind of refreshing
>>116830 #
>marked image as spoiler
>couldn't continue my post
I've never experienced that bug. Did you lose your connection?
>>116831 #
Pepe was suspicious to me from the moment I learned about the Mason's god Jahbulon. It has a frog head. Meme magic was real, but the kek stuff was blatantly astroturfed.
>>116833 #
>prevented me from typing more
Are you on a phone? Strange. Try spoilering again. Works fine for me.
>>116831 #
Given it's still early in training at v0.2, it's very impressive. Miles ahead of pony v7 or chroma. I hope it's our next step up from sdxl based models. the text encoder is very impressive.
>>116832 #
nah, it just prevented me from typing more.
>Are you on a phone?
no lol. it's no big deal though ill just remember to spoiler after i know i'm done with the post.
>>116835 #
that video was really good, though it was a definite reminder why i don't research this stuff anymore. It's pretty dark, definitely puts the head in a dark place.
>>116832 #
>Meme magic was real, but the kek stuff was blatantly astroturfed.
So it seems, so it seems.
Btw if you have time for some background noise this old Hidden Amuraka vid I posted here >>116753 # really goes into detail on it
>>116833 #
>no lol. it's no big deal though ill just remember to spoiler after i know i'm done with the post.
I'm the admin and sole dev, I'd like to know if it's reproducible because I'm not even sure how that could happen. I'll take a quick look at the spoiler code, I did change some of it recently to add support for board custom spoilers.
>>116835 #
I've already watched that video.
>>116839 #
Please do not spoiler the peach cakes
you're just adding unnecessary steps to my viewing of them
in fact, you should add a de-spoiler option to the meido console!
>>116845 #
>That one felt rather lewd
This isn't /homu/, as BO I demand lower quality standards for shitposting
>>116839 #
Oh i C. Well i'll try to reproduce it again right now with this one.
>>116842 #
de-spoil these nuts boy, git gud.
Oh you know what i'm just fucking stupid, i saw my text box go still and thought in my sleepy brain that i couldn't type anymore. It's working alright.
That said it's like 7 PM here so i have to really git it together for the all nighter.
>>116842 #
That's a high quality greta right there.
May as well dump my Bobinas since the markets are going to get rekt again soon.
>>116842 #
That one felt rather lewd to me though.
>as BO I demand lower quality standards for shitposting
djb.
>>116843 #
It's fine. Get some sleep anon.
>the all nighter
If you really need to do it... Otherwise I wouldn't recommend it. I always get more done in a better way after I sleep.
>bear posting
anon I...
>that single Greta post made me nostalgic for summer 2023 when i used to leech off of GPT 3.5 Turbo on /aicg/ and people were sharing sd1.5 greta gens and erp's
Gooood times.. haven't Greta'd in a while. I may do some for old times sake.
>>116855 #
oh my god that is pure kino. i jumped on just slightly late, after all the drama of the NAI leak and stuff already happened. its such a trip looking back at how much progressed since then.
Didn't think i'd be running LLM's that curb stomp turbo locally, image gen models that make 1.5 look decades older than it actually is by comparison, and video genning.. at all.
Is that archive available online anywhere?
I really need to organize my AI stuff into its own folder instead of dumping it in topical folders and labeling it AI.
>>116852 #
>>116853 #
I have 90% of the AI meme archive from when /pol/ discovered Bing Dall-E
Gretas in abundance!
>>116853 #
>make 1.5 look decades older
It still has it's uses, at least until niche LoRas for it go mainstream
>>116860 #
>greta dump thread
that would actually be quite funny, but also quite time consooming
>>116855 #
Post greta the bunny please.It's futaba's decision, but if you're going to dump them all imo it would be better in its own thread.
>>116858 #
>time consooming
You could upload them to filemail. I'd say catbox but I can't even access them on a VPN these days.
>>116862 #
Thankie!
>>116861 #
>You could upload them to filemail.
Yes but that would defeat the purpose of a giant thread of gretas as a POI for the board lol
fuck it maybe i'll just start it and add them incrementally and pretend I'm the one generating them every day like clockwork
>>116864 #
>hamas chans
There are loads of it in the archives they're just not tagged properly (maybe 10% of each folder has tags)
Damn I have a lot of hamas chans. The fact that you don't see anything like this coming out of 4chan these days shows how jew owned /pol/ is now. iirc these were made by the /chip/ guys.
>>116858 #
Yeah don't get me wrong, I definitely don't mock the old models. Never a fan of "throwing out the old for the new" necessarily. Going back and fucking around with the earlier tech is really fun and they have their uses.
Though i say that, and i'm struggling to find my early 2024 gens. I know i threw them somewhere on one of my back up harddrives but all i'm finding is the really shitty coomer failgens i kept because they were funny. Shit.
fuck, I was going to work on banners
#d100 (39) banners
#d100 (85) keep slopping
#d100 (50) spawn the greta thread
dump them now #d100 (46)
tomorrow #d100 (93)
different thread #d100 (43)
Wish I had another two hours to fine tune my Homura prompt back to what it was after stupidly losing it
but at least I have something to do this weekend!
mata ne
[YouTube] ???
>greta's so irrelevant now no one's done z-image turbo and flux klein loras of her
oh well lmao was gonna commit to the bit by posting some funny gens but i am not gonna train one myself.
However i will post a consolation prize, one (1) singular Abby.
last one
>>116979 #
now this time i was gonna back out and wait to make my reply, but refreshing just sent it anyway even if i cleared the text field
truly actions have consequences.
Currently redownloading z-image turbo to give that a shot, because flux 2 seems completely worthless aside from editing for some reason.
>>116980 #
>now this time i was gonna back out and wait to make my reply, but refreshing just sent it anyway even if i cleared the text field
Yes once you see a post header (date plus number), the reply has been allocated client side and registered on the server. It happens after you upload an image, add a character or copypaste twice (for some reason). Eventually I'll add back support for non-live posting (it was removed by the original dev in 2019) but it's hard to find enough people who care to make it a high priority. We used to have some russians here that wanted it, but between the format and Putin's crackdowns they didn't stick around.
>actions have consequences
You can always cancelpost things by erasing text and if #steal is enabled, you can replace your original uploaded image with someone elses.
also open posts are automatically closed after 15 minutes, regardless of if you're still typing. meguca is generally designed around short conversations on boards with a single, fast thread.
>>116984 #
>>116988 #
very cool
>Also it just occurred to me i was unintentionally genning in 1440p oneshot
lol its crazy what we can pull off with the models released within the past 6 months. A far cry from even over a year go. This is like the only tech sector i've been excited about in a long time.
>>116991 #
I haven't seen reviewbrah in ages.
>its crazy what we can pull off with the models released within the past 6 monthsI use AMD so I've never set up imagegen.
>>116997 #
my honest condolences. i was running an rx 580, until i upgraded to a 1080 in 2023, then finally my 5060 ti 16gb last year. i can't go back to AMD ever again.
especially since they fucked me in the ass dropping support for the 580 literally right when i bought it and blender's opencl support started getting flaky so renders and overall performance was unbearable.
>>117001 #
>i can't go back to AMD ever again.
AMD sucks and it took me a long time to get it since my cards have all been good. But I'm stuck with them since I use linux and Jensen has a stick up his ass about supporting open source projects. To be fair though, they were right about wayland.
>>117005 #
Up to the board owner.
>>117008 #
Tough break that it has to be this way. I wouldn't care a single second about "brand elitism" or whatever but credit were its due, Nvidia hasn't done me dirty. the 1080 was one hell of a beastly card for now being 10 years old. Haven't ran into anything that this current card didn't like and it's a damned dream for A.I.
>>117019 #
>Haven't ran into anything that this current card didn't like and it's a damned dream for A.I.
I have a 6900XT, which keeps me treading water for 4K gaming and mpv meme shaders. I should have bought the 7900XTX tbh but I decided to save $100 and some power overhead. I've been wanting to upgrade for the last year or so, but everything keeps getting delayed. I don't want to pay AI scalper prices, so I guess I'm waiting another year.
>>117032 #
He's probably seen everything if they grew up together.
>>117032 #
If the isrealis see what we're up to a third world country's gonna get nuked overnight.
>cannot believe this shit actually understood "Fuzzy cow ears"
>>117033 #
In retrospect i totally could've saved up another month and bought the 5070 instead, but i'm very happy with what i got given all the negative rep it got from the 8gb model + people's claims that PNY was a shit brand. It needed a proper undervolt/reasonable fan curve out of the box but other than that its been cool, quiet, and very fast. Probably the last real consumer card generation we're ever gonna have too given they're accelerating into agenda 2030 pretty fast.
>>117034 #
>Probably the last real consumer card generation we're ever gonna have too given they're accelerating into agenda 2030 pretty fast.
Probably. They might do some real engineering for the 7000s but supply is going be near zero because of the AI order backlogs and inevitable Taiwan war soon.
>agenda 2030
There's an unholy alliance between jews and SaaS tech people to strip people of local compute and replace it with rented compute from a central server that is easily controlled. AI has accelerated this since it allows bad goyim to nearly make their own mass media quality productions, but GeForce Now is over a decade old I think. All of this age verification, remote attestation and brain chip crap is about mind control.
>>117036 #
people might "protest" ai now but, they still willingly buy games that are permanently stuck with always-online-activated DRM protected games, SaaS for other things, willingly giving all their data to companies like discord, etc.. There's really absolutely nothing stopping this future from coming into complete fruition. It's just a matter of when not if.
Which is a big reason i dropped the $ on my card instead of waiting a bit longer. No telling when supplies will just suddenly dry up or markets will crash. Or both.
>>117044 #
>There's really absolutely nothing stopping this future from coming into complete fruition. It's just a matter of when not if.
I didn't really appreciate the amount of time and effort that goes into build and maintaining relevant open source projects until I worked on meguca for a while. It's got about 50k LOC. I understand pretty much all of it and it's not as complex as it sounds, but I have no time to work on anything else if I want to add or change something in it. There's a rapidly shrinking and aging group of GPL programmers that believe in the ideals of free software and try to give back to the community and fight the attempted revival of mainframe computing, and they're being replaced with eager corposlaves that will actively replace GPLed code with permissive licensed code that integrates better with proprietary crap. They hate the idea of user freedom because it means their corporate and tranny overlords can't control people. They basically want to tivoize computers in general, but linux in particular. I'm pretty blackpilled about the future of free computing since it wouldn't take much for intel, amd and nvidia to require encrypted binary blobs that can only run on TPM "secured" systems and the real freedom crown would be eager to build those systems since they think your time and effort disappearing into some exploitative corporate shitware is a good thing. Gomen for ranting.
>>117045 #
I respect the rant, i could never fully wrap my brain around programming, some snippets click properly in my brain but looking at an actual programming interface makes my head spin. Appreciation is entirely needed to those FOSS projects we use on a daily basis and especially the ones that take the fight to FAGMAN and the powers that be.
I wanna hope LLMS and other A.I can be used against this 2030 plan, that people will fully embrace what COULD be done for the better and go for it, but honestly i'm blackpilled as well.
>>117056 #
>I wanna hope LLMS and other A.I can be used against this 2030 plan
How useful it is depends on the language and your own personal preferences but personally I would never trust vibe coded software. I do use AI for advice/ideas here (mostly for JS, Go is easy), but the problem with generated code is that it's far easier to maintain something in the long term if you wrote it yourself since you'll remember why you wrote it and the assumptions and expectations around input, output and expected behavior. tbh that does apply to contributed code as well, but at least then you can talk to the contributor about their ideas and rationale. AI is probabilistic so you'll get 5 different answers for 5 identical prompts. That's okay if you want something quick and simple, like the epub identification code I recently added to the thumbnailer, but if you apply that to a decent proportion of a codebase you get chaos that's impossible to follow psychologically even if it all works. Like in a more complex language like javascript it will be mixing and matching features from difference ECMAscript versions, different comment, indent and ; styles, different programming patterns etc. Then you add in someone who has no grounding in programming and vibe codes an entire project and you end up with something that barely works and is full of security holes.
Programming is basically a digital trade, not a science. Even if something can be automated it's far easier to understand if you do it yourself and the work is better appreciated and of a higher quality if it's done manually. Even if you do what the AI does, which is glue together stack overflow answers, you'll still do it in your own styleunless you're indianand probably do some basic reading and testing. That said I'm not mourning software engineers. They're getting replaced by claude code because their code is nearly as bad and unmaintainable. They're a bunch of classist faggots that turned what should be a trade into an exclusive club for shitlibs and undermined the GPL to do it.
>>117057 #
I heard vibe coding was a huge problem, but you going into it on the more technical side makes it sound like a nightmare.
It's like the big companies turbo charged how detrimental to the entire world Indians are with this shit kek
>>117059 #
>It's like the big companies turbo charged how detrimental to the entire world Indians are with this shit kek
Basically yeah. You can put all the prompts into an AI that you like, but at scale companies are just bringing the nightmare of dividing up feature work between an SV office and 5 indian call centres in-house because LLMs just remix their input based on whatever the highest internal weights are. You can't make AI truly great unless it can take the entire context of a project and language like a human can, but even if it has that in memory it's "understanding" changes from paragraph to paragraph and its output degrades the longer it gets because it's fundamentally just a better autocomplete.
imo LLMs for programming are just a time saver for experienced programmers in bad languages. The AI bubble has to pop badly at some point because LLMs fundamentally can't scale to the AGI/ASI hoax of superhuman or even human-like understanding. Unfortunately it'll take a while since SV is full of so many retarded rationalist cultists that they would rather believe the human brain works like an LLM than admit LLMs can't behave like a human brain.
Anyway I should go watch some anime and sleep. If I was talking to the newflen, then you can make >>>/homu/ readable by either changing your CSS theme from default in the cog menu or blocking remote fonts in ublock. AI content isn't exiled here or anything, I just thought it would be useful to add a board for sharing AI content and teaching others how to use the tech. We also have >>>/cr/ if you're into human art too, not a lot of interest there tho outside of some music stuff that's mostly discussed on /homu/.
>>117068 #
>I just hope the bubble pops so we can get back to building PC's again. I'll definitely be a panic buyer that goes ham, full on triple pc setup + at home server for backups and streaming.
The bright side of Moore's law slowing down is that anything high end that you buy in the next gen should last a very long time unless photonics take off.
>>117061 #
>>117067 #
I just hope the bubble pops so we can get back to building PC's again. I'll definitely be a panic buyer that goes ham, full on triple pc setup + at home server for backups and streaming.
Yeah thankfully i figured out the UI pretty quick and deciphered the weebspeak pretty quick so it's all good.
And so... the nightmare begins again
>>117399 #
>is there any possible timeline where you'd miss the season finale to frieren tonight?
bruh, of course I'll be there
I should figure out how to edit these better
unfortunately, it seems over complicated to do in Swarm without touching (un)comfy UI
>>117561 #
occasionally, with the built in repaint tool, which is very weak and usually gives worse results
there is a way to implement actual good editing functions using additional models, I just haven't figured it out
>>117470 #
>>117602 #
I strongly recommend Animaika if you wanna gen consistently without worrying about inpainting/second passes.
https://civitai.com/models/2426265/animaika?modelVersionId=2743280
mou I missed the newflen. Maybe he was on /homu/.
>>117953 #
digging the webrings, had lurked different ones for years, even moreso after the funny shit that happened to cuckchan last year. just decided to start posting more now.
I'm out of hamas chans so I should go work on somethingor sleep
>>117956 #
Neat, thanks. Well if you have any feedback and don't like /homu/ for whatever reason, I'll keep lurking this board for a few days. It'll give me some motivation to sort out my AI pics anyway.
>>117961 #
I have enough stuff to manage... Would you write a guide? Could help get more activity here from /homu/.
>>117960 #
>I'm out of hamas chans
You should find an API and start shitposting your own slop, like the cool kids
I'm about to porn-bomb the thread slightly (in spoilers of course!)
Animaika v2.7 just dropped, it's a massive improvement and step up in a way i didn't expect for such a small incremental update.
If a model like this is so easily finetuned and even slopmerged, v0.5 of the base model is going to be crazy.
>neon lit bedroom
>swinging off pole
>exaggeration lines i didn't prompt for
S O U L
>>119333 #
>Animaika v2.7
I'd never even heard of it until you recommended it the other day but it seemed to already out-perform Anima (besides speed) so this should be gud.
Will update ina bit
>>119822 #
Fat. Also the PSG gens don't really get the artstyle. Lewd stuff in the cartoon inspired style >>119333 # >>119334 # looks weird imo.
there does indeed appear to be an improvement to detail
>>120233 #
Homu's eyes look like they belong in Fate. AI can't do the super detailed with lines eye style.
does the iron cross easily, now to try for the happy windmill
Spamming 'swastika' seems to work, unlike in SD
just can't seem to give it the proper rectangular flag border
>2 hours of inpainting later
>>120260 #
Maybe tomorrow I'll spend the evening trying to get it to generate her timepiece/shield thingswastika will have probably been easier cuz idk how to prompt such a thing
>4 API edits later
>>120257 #
>>120261 #
>>120299 #
>>120322 #
2 hours of inpainting? Pal this shit handles swastika like its been trained on six gorillion examples lmao
>>120341 #
Arm bands an circular bewb stickers maybe, but try getting the actual flag on a tank and get back to me
>>120351 #IMAGE DELETED BY 'Futaba'
>first try
>slop
yes.... very good!
>>120353 #
sir that's not the ww2 german flag it needs the hwite circle to be historically accurate that's what I'm saying
picrel is the the first time it's actually gen'd it for me
>>120352 #
>>slop
bro over here getting picky about his swastikas now what is this faggotry, it did what you wanted it do, just prompt better.
>>120352 #
>>120358 #
>2023, absurdres, highres, @riz, akemi homura, holding gun, sitting on a tank, on the side of the tank is a red rectangular decal with a black swastika encased in a white circle in the middle, simple background,
i'm no promptsmith or master genner but there you go
>>120362 #
oh wow you're actually prompting with prose rather than a wall of stable diffusion commas, I guess that actually works with this model then?
>>120367 #
>reeding the instructions
my greatest weakness...
thanks for the update...
>>120366 #
It's trained on everything, natural language works really well (almost perfect for a model in v0.2)
https://huggingface.co/circlestone-labs/Anima
Okay my work here is doneIMAGE DELETED BY 'Futaba'
>(flying a nazi flag: black swastika, white circle, red nazi flag:1.2), swastika insignia on flag with black cross
>the moment i realized my lora i trained to make anima slightly better at clown girls can actually turn any female character into voluptuous clown girls
>AND mixed with this cool 90s anime lora
oh lol. also i can't even post this one to civitai because their TOS auto-blocks anything with swearie words in them. wow!
>>120731 #
He's cross referencing this post in the main thread >>120727
in some form of crude unflattering ironic humor...
>>120732 #
very cuckchan *type concerntrolling it seems. hope theres not a leakage, like we'll see more of it.
anyway gonna kinda segway from that subject back to something more interesting/uplifting.
Flux Klein is really fun for the image editing/style shifting stuff it can do, picrel was an early experiment where i told it to put the subject in a different scene + style. Party surrounded by people shot on a specific 2000s camera that produces the red eye effect. cool stuff.
>>120734 #
Nothing wrong with using SwarmUI and other options, I dual wield cumfartUI and Forge Neo based on what works better/quicker/easier.
Comfy is the linux of frontends/webui's. It's unnecessarily complicated, too many knobs and dials to do things that were figured out 3 years ago, big example being hiresfix/img2img'ing is broken and convoluted in comfy but forge neo's is perfect because it's been the same since A1111. Swarmui takes getting used to and can be a bit bitchy but if you can get it working, i don't see the shame in its usage.
the real subhumans are still on A1111/regular forge. Or SaaS users.
So for me it's comfyui to use videogen/Flux Klein/Z-Image, Forge Neo for SD/XL/Anima.
>>120733 #
I assume you're using comfyUI like a mentally astute normal personunlike me who requires a plebbit tier fascia like SwarmUI to avoid uncomfortable noodle exposure?
>Comfy is the linux of frontends/webui's
swarm uses it in the backend I just haven't figured out how to load additional models for editing etc (I assume that's how it works)
Maybe I'll do that tonight, or maybe just switch to forge if that's what the cool kids are doing
ty
>>120734 #
It's really down to personal preference/prior UI experience. I've been used to Blender 3D node spaghetti since like 2016 so comfy was second nature already. What makes it complicated/annoying are the community's inability to work together in a coherent ecosystem. Similar issue to linux IMO. Packages break all the time for no discernable reason in python, everything conflicts with each other so you really need to be careful what node packs you install, mostly down to WHY you install each one really. Which is why i split workloads between comfy and forge neo, forge neo stays a consistent walled garden for only models that may need that img2img pass, comfy stays consistent as long as i don't ((pull)) or install any new nodes. Spaghetti is something you can get used to when you start reading the nodes "in order". Diffusion model or checkpoint, text encoder, vae, extra nodes for model logic, and the ksampler + vae decode then output nodes. Ez.
Once people start adding faggy shit that over complicates the workflow and maybe slightly improves output quality is where i completely nope out. Flux klein is literally just what i outlined step by step and that's it. One shot to 1080p, no img2img.
>>120740 #
>Diffusion model or checkpoint, text encoder, vae, extra nodes for model logic, and the ksampler + vae decode then output nodes. Ez.
Hmmm, that might be worth noting if the OP ever gets updated with instructions/tutorials
tyvm
Despite all my rage, illustrious still has a use caseusing specific LoRas for particular art styles
anyways here's more homura slop
Don't worry, anon will do the heavy lifting
>>121512 #
for me it's Nova Animal having a particular semi-real style i've grown to love over the years.
I don't see it getting an Anima version until anima v1.0. my autistic brain needs it NOOWW. Having a text encoder that can actually understand exactly what i want on the first try is so sweet and perfect, i can't do with sdxl's TE anymore.
>>122305 #
giga stacy moves honestly
>>122303 #
Could you please teach alvy how to un-embiggen his images? Your homugens are pleasantly smol.
>obese stocking
I'll respect that this is your board and not hit the buttons, but ugh that's ugly.
>>122307 #
>obese
>large breasts
men like you need to be forced into a re-education camp to put an end to this faggotry. reminds me of an exact word-for-word response i got on cuckchan a year ago. at least there the guy got ripped apart and called a fag.
>>122308 #
>obese
>large breasts
Small boobs are better, but the problem is her face. Actually wait, you're not futaba are you
>>122309 #
the face is very american, it's why i trained my own loras to un-foid-ify them. that was just posted as a raw example of what that model usually outputs before i train better looks that seem to also change the realism a bit.
-gotcha, it's fine. no i'm not the board owner, i just got here a little bit ago.
>>122310 #
Futaba is the board owner here, sorry.
https://homurachan.org/json/board-staff/ai
I think is the last AI thing I haven't posted here.
this one anon that cannot handle a seven foot tall thick amazonian woman is gonna LEARN to.
actual re-education camp tactics i'm employing here folks!
>>123621 #
These are all I have of those that I have. There's an anon on /homu/ who might have more, I'll ask him later.
>anti-thicc girl faggotry
Soft girls are nice, but a lot of thicc posters use it as a cover for fat fetishism.
>>123618 #
>>123619 #
>>123620 #
all the grilling you about your anti-thicc girl faggotry aside, i loved these big tech gens i forget if its bing or dall-e they used. There's one of peach and elsa were they have nice knockers and they're eating in a fancy restaurant, gonna have to recreate it in Anima when its 3d capabilities get better.
>i can't find the image im talking about so here's greta tardburger
>but a lot of thicc posters use it as a cover for fat fetishism
well, most people who overly post about a specific fetish are compensating for something, or just have deep mental illness. Many such cases there. In my case it just happens to be my prevailing characteristic. I like shortstacks and petite girls, and i'm into loli to an extent, but i just like what curves bring to the table.
>>123622 #
the early sdxl/late sd1.5 shit blew my dick off back in 2023/2024. theres a few i saved that still get me going *PHEW*.
the guy that made picrel wanted his methods to remain secret for some unknown reason. this reminds me to finally sit and organize my downloads, i have over 1500 image downloads sitting here, hard to sift through.
>>123623 #
>i have over 1500 image downloads sitting here, hard to sift through
Tell me about it, I just reuploaded stuff and didn't realize. I think this one is new though. I'll sort out my stuff too.
>>123624 #
scrolling through my downloads in chronological order and watching the realistic gens turn properly photorealistic all of a sudden is so fucking jarring lol
>>123631 #
ZAMN Shadow almost genocided the entire human race for THAT piece of VANILLA CAKE?
>>123632 #
i should actually do that, thanks. Though my downloads doesn't have any real dupes, it's just a lot of shit with no filenames or filenames that are just gibberish.
>>123633 #
>it's just a lot of shit with no filenames or filenames that are just gibberish.
You could also try image optimizes like jpegoptim and oxipng to losslessly compress images. Sometimes you can have two or three pics that are the same, but they have different file hashes because of partial compression at the source. It's not going to work for 4chan sourced pics though, they reencode everything to try to remove embedded data. Just make sure you only remove safe metadata like exif. My commands arejpegoptim --strip-exifoxipng --strip safe --fix
Fix is optional but it will remove embedded data and try to fix broken pngs.
>>123634 #
Honestly funny thing is, 1,000~ images is pretty tame given i've been doing this since Q1 2024, downloading since q3 2023. i have over 10,000 downloaded general images from the internet over the course of 7 years or so. I'm shocked i haven't gone ham in downloading a giga fuckton of other people's gens.
must remember to generate more 2D styles going forward, i've been deeply embedded in the semireal/3d slop for a while. i think only because it's so difficult to attain an aesthetically pleasing look that it's like a challenge to conquer.
also found this jej
>>123633 #
>ZAMN Shadow almost genocided the entire human race for THAT piece of VANILLA CAKE?
It's amusingly canon.
H O R N Y indeed.
This one was actually posted by Irving's twitter account to promote his new book jej.
genuinely so happy the old z-image turbo stuff could find a new place to be appreciated again
-that said flux 2 klein came out soon after and curb stomped it into irrelevancy lol
-this one z image couldn't do without mangling her body
Looks I'm like out of pics. Thanks for sharing and getting me to organize my shit. I should probably get back to coding anyway.
>>123659 #
yeah that was probably my last image dump for a lil while, sometimes its good to just do stupid shit. haven't made anything like these literally since these timestamps.
Might not have posted this one.
>>123662 #
>haven't made anything like these literally since these timestamps.
Honestly I just collect funny things. It's too complicated for me to keep up with the state of the art and I'm on ayymd anyway. I've looked into models a little, but it seems like most of them are so close in benchmarks that it's all vibes based.
>>123663 #
cool as fuck lol i like that the golem is made out of clay.
>It's too complicated for me to keep up with the state of the art and I'm on ayymd anyway.
many such cases, sorry to hear. when you do get the chance to jump to njudea, definitely check in on how Anima's progressed up to that point. Realism models are always going to come and go and be really difficult to stick with.
>most of them are so close in benchmarks that it's all vibes based
benchmarks are almost always mememarks. It's 100% vibes based, whatever ends up providing the most satisfactory results (subjective) in whatever use cases (subjective once again) people put it to the test at. For me, it depends on how big the model is in parameters, that's how i decide what i'd want to do with it.
Flux 2 is perfectly in the middleground of being high parameter enough to do realism to my liking, i'd love to stick with it permanently but the vibes do not agree sadly, no one trains for it. Z-Image won out even though it sucks dickshit.
Anima is slightly bigger than SDXL, and uses an actually good newer text encoder, but none of that stops it from being a top of the line 2d model.
the vibes are indeed not easy to follow but i'm glad i do because it helped me save a lot of time in deciding what models to sink time into and what to delete to save on storage.
Anyway here's princess peach playing twister in the style of Ayami Kojima.
Ah I do have some questionable things.
Not sure if this one is AI. Saucenao is awful now and half the pics that are posted originally twitter can't be traced back anyway.
Alright this is definitely the last one. lol this was a fun start to the morning.
Actually there's one more thing. I should put it here anyway since someone uploads it all the time on /homu/ anyway.
Source is this guy https://www.youtube.com/@Gossip.Goblin
Posting some more stuff before I nap.
>>124627 #
I don't mind, I was just curious. I'm just always looking for feedback and I know that some people don't like how heavy /homu/ has been on /pol/ stuff because of the war.
>stocking as tanya
Saving that one jej
>>124629 #
Eeehhh, topics are gonna ebb and flow i don't personally have a problem either way.
Also lol i think it's a one piece thing actually https://civitai.com/models/893256?modelVersionId=2832719
>when you hide the BPD bitch's meds (again)
god i love how easily trainable anima is i REALLY hope this model continues to pick up steam
>>124632 #
and yeah i know Tanya, she's a cutie. love that little psycho schizo. there's several loras on Civitai dedicated to just recreating crazy facial expressions from One Piece.
>>124630 #
>Kakegurui face expression
Oh that's a fun show.
>>124634 #
>what Kakegurui is
It's kinda hard to describe. Basically it's BPD bitches gambling in school.
>>124633 #
ah wait just googled it, i *had no clue what Kakegurui is whoops lol
yknow its funny i've lurked anime basketweaving and gun fabricating forums for over a decade and i know very little Anime. Hell i just watched yu yu hakusho for the first time ever with my best friend recently. (favorite show of all time now bytheway).
>Basically it's BPD bitches gambling in school
ah sounds about right going by that lora.
>>124634 #
>i know very little Anime
Have you checked out >>>/mtv/? We're watching a ton lately. Kind of unusual for us but it'll be a regular feature on wednesday nights anyway if you're a burger.
>>124635 #
I may down the line. Used to lurk cytubes a long while back, had fun watching different media like cheesy old horror movies and 80s movies. if something ends up getting streamed that i wanted to check out ill become a regular.
>We've done self serve for years
kino
>>124636 #
Also you might want to refresh your tab after your post. I made some feature changes to the frontend.
>cytubes
>streamed
We've done self serve for years. I try to make sure there's an outside stream available but we don't run a cytube, anons just fetch stuff themselves and follow a server timer. I was working recently on implementing an embedded stream, but all the options are kind of shit. I'll probably just add a cytube embed similar to megutv if I can get the iframe to only include the video and not the entire channel.
>Self hosted
No. That's pretty difficult bandwidth and thoroughput wise if you don't use cloudflare.
man I'm finding this stuff everywhere
Might not be AI.
This bih is hard to generate
>>124681 #
It's not that far off. If you can get the face markings right, the rest would be okay. Eyes are a bit too human though.
>realizing just now that animaika after v23 regressed in terms of text rendering
wuut duuh fuuuukk
>>124690 #
>regressed
How can you tell? It's all probabilistic and text is already hard to render normally.
>>124693 #
certain words turn the entire sentence into complete sdxl tier gibberish, were v23 seems fine but always duplicates a specific letter in the word.
that said some words are just completely impossible, the word spiritually in v23 is gibberish, after that version it corrupts the whole sentence lol
im not surprised though, this is a big finetune on top of a model that's not even finished training. this scene moves way too fast, people gotta respect the time it takes to train a big ass dataset. can't have pony/illustrious models coming out every year.
>>124686 #
>If you can get the face markings right
Yes that's the hard part I'm talking about
>>124694 #
>AI refuses to generate the word spiritually
How peculiar
Last one for now
Time to go terrorize the supermarket
In accordance with Elimination Ordinance 3, Clause 2, Elimination Level has risen by three.
Yeah it's not possible without a LoRa
But I atleast learned a little bit about modifying an input image with paint and the image weights that work
>>125117 #
I only have 8gb Vram but it's probably 15-30 seconds per gen, slightly quicker with an input image, just a lot of trial and error with the prompt trying different words
>>125116 #
At least the eyes are cute. How long does it take you to gen one of those? Do you make a few and pick the best one?
>>125117 #
With a less niche character it usually only takes about 2 to 5 tries to get something good but it struggles differentiating the markings with her hair, plus it's kind of hard to describe them idk
If I made a LoRa it would be no problem, but there's also not a ton of pics of her to use so that would probably require generating/editing additional images to add to the data set
Just remembered Novel AI can usually succeed where most local models fail
Sadly it's a paid service with no daily free tokens besides 30 gens to start
heh heh heh
None of my models will do Re-I from Ergo Proxy... how am I supposed to fap?
>anima 3 knows mossancannibalis
>but not chudjak
fug
heh
Anima preview 3 came out days ago, i'm just now getting to it. Seems like there's more hype around this release than the previous, which seems to imply more people found Anima in all this time. Nice.
Hopefully next release fixes all the upscaling problems so i don't need that 2048pix lora anymore.
everything from v2 may need to be retrained for v3, even though people seem to be speculating the opposite. Loras rape the shit out of text capability, which seems to have improved with the new version.
also issues like picrel seem to crop up with these loras, the text boxes.
>think the model can't do something right
>it's just something dumb in the prompt
literally me every day
stole this from the 4/g/ animu thread awhile ago
back before anima
wish it could do her without a LoRa...I believe in evil waifu supremacy>>128133 #>Anima preview 3 came out days agodownloading it nowI suck at prompting so I probably won't notice a differencebut it will give me an excuse to post some slop
>>128728 #
preview 3 seems very promising, improvements in a few areas and with all the hype i think people are gonna be making some neat stuff.i don't think i have any evil waifus unless you count early Ayane from DOA
>>128857 #
>God, as in her lesbian GF
we'll see if she lives up to the sterotype and commits domestic abuse in the new movie I guess
for now thoughI don't think she's an 'evil' waifu, just a dark type pokemon
>>128821 #
>>128823 #
>>128857 #
>>128859 #
it just now occurred to me that none of this is even a.i, are you sure you're in the right BOARD?
correcting my newfaggotry on the site name and posts non-ai on the ai thread what are you dooiingg buudddeeee
>>128862 #
There are plenty of APIs you can use to fit in with us, sillya
Before local I was using pixAI, which is basically where I learned how to prompt and mess with image settings
But you can also just go there to steal media like I do now
>>128873 #
For now I'd rather be lazy and collect stuff from anons. I could dump the old glif memes made before your time for my next couple of posts though.
>>128878 #
again, this is the ai thread
but o-okay
>>128880 #
oh, I actually think I remember when this was a thing then, nvm then carry on
okay last one time for work
Is there anybody out there?
pssp pssp pssp
Busy day...
IMAGE DELETED BY 'admin'
I don't speak retard nigger-ese
Can somebody translate?
liking this one
Where did flens go?
>>150594 #
I should really study how to make better slop
now that you mention it
>took like 5 tries with an API to edit this properly
this shit is getting insane
not even trying to prompt max on these
It used to take 100 gacha rolls to get an accurate Nihontō
#roulette (5/6)BANNED BY 'system' FOR 3 HOURS FOR "lost at #roulette"
#roulette (3/6)
#roulette (2/6)
#roulette (1/6)
#roulette (6/6)
#roulette (6/6)
ReturnCatalogTopLast 100Last 500Locked to bottom