[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Edit][Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
4chan
/vg/ - Video Game Generals


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


File: img.png (1.5 MB, 1024x1024)
1.5 MB PNG
Cherry Blossom Edition
Last Thread: >>563010304

>Read First: https://i.imgur.com/lwWTTp9.png
>PSA: https://rentry.org/AntiwhineMkIII | https://rentry.org/shillfilter

▶Cloud Services
https://featherless.ai
https://novelai.net
https://openrouter.ai

▶Frontends
https://koboldai.org/cpp
https://sillytavern.app
https://gitgud.io/mir/storypad-qt
https://github.com/lmg-anon/mikupad

▶Additional Information
>Chat Models — e.g. AetherRoom, Claude:>>>/vg/aicg
>Local Models — e.g. Command R, Mistral, Qwen:>>>/g/lmg

▶Game(s)
https://store.steampowered.com/app/1889620

▶Tools & Resources
Wiki: https://aids.miraheze.org
Prompts: https://prompts.forthisfeel.club/
Scripts: https://naiscriptvault.com/
NAI Guide Collection: https://rentry.org/guide-collection
Other/Historical: https://rentry.org/aids-misceIIany
>AID: https://rentry.org/remember-what-they-took-from-you

▶/aids/ Events
Themed Prompts: https://aids.miraheze.org/wiki/Theme_Fridays
>Theme Fridays have transformed into Theme Months!
>Write for April's theme: Asia/the Orient!

▶Model News
>(03/31) NAI releases Xialong
>(02/23) NAI updated its Canvas
>(02/12) Z.AI releases GLM-5: https://z.ai/blog/glm-5
>(02/05) Anthropic releases Claude Opus 4.6: https://www.anthropic.com/news/claude-opus-4-6
>(12/19) NovelAI introduces User Scripts: https://blog.novelai.net/introducing-novelai-user-scripts-8b6ac19aa170
>(10/01) Z.AI releases GLM-4.6: https://z.ai/blog/glm-4.6 ; NovelAI updates untuned preview to GLM-4.6: https://novelai.net/updates

▶Thread News
>(03/07) Anon updates his frontend: >>559652987 https://files.catbox.moe/930fvy.html
>(02/05) Emacs Update: https://files.catbox.moe/6r6wxg.el
>(01/16) Anon makes a Mikupad Screenshot Script
https://github.com/LordFoogThe4rd/mikupad-screenshot
>(11/17) Anon makes a Mikupad Guide
https://rentry.org/MikupadIntroGuide
>>
A neat way to maybe translate images into artist mixes when unidentified via the inspect site: https://huggingface.co/spaces/DraconicDragon/Kaloscope-artist-style-classifier
>>
What's new in the hood my niggas?
>>
>>563949135
Interesting. But what's actually a good score? I keep getting the top result being about a 1%, but I have no idea if that's confident at all or just a complete guess.
>>
File: 1639592497053.gif (1.94 MB, 366x278)
1.94 MB GIF
>waiting for Deepseek v4
>>
>>563957765
be patient, lil nigga
>>
>>563949135
Oh, that's neat
>>
>>563947295 (OP)
I feel like making tribal/prehistoric kino. What are some good prompts/authors/tags that have worked for you guys?
>>
>>563964970
Clan of the Cave bear is pretty much the only caveman kino I think I've ever read
I should rectify that
>>
>>563964970
Robert E. Howard
>>
>>563964970
your fuck money want my png slubber
>>
>>563957765
Stay strong
>>
>>563972648
Stay schlong.
>>
>>563957765
Makes me wish CF watched good movies. Him referring to NAI like Neil Breen would at least be funny.
>>
>>563976042
very true
>>
>>563976042
know whats up for money suckle and?
>>
has there been any trace of a hint of nai imagegen v5 cause my wallet's getting anal pounded from multiple unrelated releases at once and it may as well keep coming
>>
>>563989195
money suckle for fuck on the back of the bus on the highway freeway
>>
File: tsuyuamettv.png (16 KB, 419x90)
16 KB PNG
>>563989195
its coming out next month.
>>
>>563991151
no u
>>563991720
far off enough it won't strain me much, cool
>>
>>563991720
thanks for the info
>>
>>
File: ZOOM.png (1.22 MB, 1024x1024)
1.22 MB PNG
>>563991720
So I guess we're stuck with this guy for the next five years.
>>
>>563997782
Why do you hypothesize such bullshit?
>>
>>563997782
The next one's female, thankfully. Hopefully submissive and meek
>>
>>564001474
Nein. A gyaru kitsune who is silly and wily.
>>
>>564001552
a dominatrix gyaru kitsune with abs that could whet blades
>>
>>564001552
I will forever hold out hope for our gyaru kitsune.
>>
>>564001552
>>564001474
We need an evil one, we don't have any evil ones yet. She needs to be an evil cunt.
>>
>>564002051
An evil gyaru kitsune
Yeah, I could see that being a cool concept
>>
So does Xiaolong not work with Text Adventure?
>>
>>563947295 (OP)
Do people not share their scenarios anymore?
>>
>>564007556
Can't share what you don't make.
>>
>>564007556
You can go to /aicg/ if you want an active community.
>>
>>564007556
I post what I make and that's all I can do.
>>
>>564009609
You first. Nobody will miss you and your whining.
>>
A blender pack of 3d models from the Mario.
https://mega.nz/file/culFhThK#z3AF8Oc0pBSvfwcPiRiExWjtDqg8RODAhY-9NQY2pUU
Might be useful for 3d posing?
>>
>>564007556
We post gay furry porn here sir
>>
>>564007556
Have any to share yourself, or are you just throwing stones at our glass house?
>>
File: aids.jpg (39 KB, 512x384)
39 KB JPG
>>
>>564010878
You already know the answer
>>
can't they just glue 20 kayras together
>>
>>564010878
my png want money for your slubber
>>
File: dtsklp.webm (334 KB, 512x720)
334 KB WEBM
>>564007556
Most people are just making their fetish stuff. Not exactly interesting to share. Also gets used as a cudgel to shut people up constantly whenever they talk about a service or model (why aren't you POSTING SCENARIOS?) like bitch, I do the same old boring futa harem impregnation antics story almost every time, nobody wants to see that. I'm more interested in meta talk (i.e. hey I want the LLM to do this, any ideas how. Or hey, here's a new way to handle context for better long term stories, here's how to prompt engineer this behavior, etc). Would you be happy if I spammed futa all thread like the gay furry poster? Would that be 'contributing'? I don't feel like it would be t b h
>>
>>564013446
I'd like to see it.
>>
what do you guys expect hmomafag's next round of gay furry pics to be about?
>>
>>564013741
I don't give a fuck.
>>
>>564013830
rude
he works hard on providing gay furry sex pics for all of us
>>
>>564013909
He does it for his own self-gratification. Most people would rather he didn't spam the thread with them.
>>
>>564014008
yeah
>>
>>564014008
I thought we all liked his gay furry sex pics?
>>
>>564007556
I still need to rework that one isekai prompt I shared a while ago. I’ve been meaning to get to it, but you know how it is… Life’s a bitch, and inspiration comes and goes quickly.
>>
>>564014008
You are the same nigger who shits on NAI. You just hate our thread culture. Why don't you just fuck off?
>>
File: 1749602955267631.png (8 KB, 477x130)
8 KB PNG
How about you get a life?
>>
>>564014785
would be me and complaining in the thread if they didn't block all payment in my country
>>
File: 1577941074333.jpg (2.38 MB, 2328x3258)
2.38 MB JPG
Best text lora for this kind of character?
>>
>>564014785
We need MAGNUM Opus tier
>>
>>564014989
Haven’t tried an idol prompt yet, but it can do loli characters without a problem.
>>
>>564007556
I stopped when like 3 threads in a row with no (you)s and I figured what I offered was not what the people wanted.
>>
>>564015126
Have you considered sharing just to share, without any expectation of receiving attention for it? Also, consider that lurkers exist and that if you want engagement, you have to go out of your way to start a discussion. Like such: How do you depict mature female characters, if at all, and if not, what characters do you tend to go for?
>>
Exhibit A: the same people who complain about nothing happening simultaneously refuse to allow things to happen.
>>
>>564015292
Depends on what definition of "mature" you're going for. The oldest I tend to go is mid-late 30s for women but I've had a few male leads that could be in their early 40s (but I very rarely state definite ages)
>>
>>564017424
Mature for me is anywhere past the traditional ending dates for college attendance (late 20s). Enough time to theoretically get a job, marry, and start having children assuming a decent economy. Early maturity transforms into ripe maturity by the early 30s, then as it progresses, it becomes more vintage as it continues to the 40s and beyond. Theoretically, assuming a well kept body, even chicks in their 80s are attractive.
>>
>>564017902
Then to answer your question, I usually tend to portray the mature women as women who were/are hard-working professionals who neglected their social lives in pursuit of work. I never really noticed the pattern until now. I guess, to me, it's the easiest way to avoid a lot of past partner baggage. Especially once the AI gets ahold of it, you constantly have to edit out dick/exes comparisons and shit (oh all of my other partners were wayyyy smaller than you :) type shit). I hate that for some reason, it just takes me right out of the story and makes me cringe. As for the mature men, they don't fit as much of a pattern but I do have a very soft spot for that dadbod/strongfat type of older man.
>>
>way smaller than you
never happens to me
>fat
never happens to me
wonning
>>
>>564007556
I don't care enough to share anymore. Everytime I come here, there's one gay furry faggot spamming his shit. He clearly won't read/use my prompts, and other people would equally don't bother to come here for a 1% chance that a new post is NOT gay furry shit.
>>
nobody cares
shut the fuck up
>>
File: 1754661364469282.png (10 KB, 426x170)
10 KB PNG
>>564020549
Most people are intelligent enough to hide posts they don't want to see. Much like what I'm now going to do to your whining.
>>
>>564020673
Or just walk away and find a place where you don't have to do anything to have a tolerable community.
>>
>>564021090
yeah, yeah, we get it
you're back to the thread merging schtick
when are you going to realize that it's never going to happen? if it hasn't happened by now, it's not going to happen
>>
>>564019252
I periodically run stuff with experienced characters and have not really ran into baggage type material, which you'd think would pop up at some point. Mine tend toward slow burn slice of life stuff, even when sex happens, so perhaps that is what prevents such boner killing things from happening?
>>
>>564021795
Could just be the narrative style I prefer that lends itself to it. If I'm not careful it's always "my ex this, my ex that"
MC-chan, I do not care about your ex. This story is about a new man.
It's probably a holdover from romance novels, it's a common way to generate extra drama.
>>
>>564007556
I kinda stopped once the models got good enough to write 80% of them for me
I figure people don't want shit an AI wrote
>>
>>564022080
same
>>
>>564022080
>>564023953
You shouldn't make assumptions as to what other people might want. That's rude.
>>
File: mom.png (66 KB, 1080x339)
66 KB PNG
Excuse me?
>>
>>564020673
It's impossible to filter that faggot and if I will press hide to every post he does I may as well hide the whole thread.
>>564020549
This, I barely come here for the exact same reason, it's just the gay furry interacting with himself and with the whinefag
>>
>>564031463
Sick burn
>>
File: history.png (2.29 MB, 1024x1024)
2.29 MB PNG
>>564031463
>>
>>564031595
>samefagging 2 hours later
Feel free to fuck off. We won't miss you.
>>
>>564017424
Oldest I go is usually 38/39 because those are the last real years of feasible natural impregnation. I usually have it be an oyakodon scenario where the daughter, who is now an adult herself, feels bad for her mom (whose husband is deeeeeeaaaaaddd so sad) and wants to share my self-insert character with her.
Although maybe it's time I do the same but have the husband still be alive. I think I'd feel too bad for the husband
>>
>>564040783
As long as the woman has an egg, impregnation can naturally occur (though very rarely, and with greater instances of miscarriage, though viable births remain possible) up to the late forties/early fifties, when the egg supply by all means has most assuredly been depleted. Therefore, it remains feasible up to that point. Besides that, donor eggs exist. Thus, it remains feasible so long as the woman has a womb. Plus, stories can wave around bullshit and pseudotechnology to get around natural limits.
>>
>>564041507
One of my former coworker's sister had her first child at 49 and her second at I think 50-51. Conceived naturally and on accident.
Needless to say, she and the doctors was very, very surprised. The kids turned out okay too, one's a doctor and the other is also a professional of some sort.
>>
File: yoooooo!.png (1.45 MB, 1024x1024)
1.45 MB PNG
>>
File: robot rebellion.png (72 KB, 1080x459)
72 KB PNG
In theory I should dislike it when Xialong automatically starts writing a new story that has nothing to do with what I just wrote, but it's so funny when it happens that I can't be mad.
>>
>>564047647
ZAP, MOTHERFUCKER
>>
>>563957765
22nd of April.
t. knower (seriously)
>>
>>564050006
I believe it
>>
File: no wanking.png (74 KB, 1080x412)
74 KB PNG
>>
>>564043573
Glad to hear but people underestimate how risky geriatric pregnancies are in terms of defects. A kid born from a 45 year old mother has around 3,5% chance to have Down Syndrome.
https://pubmed.ncbi.nlm.nih.gov/15849789/

And that's just one abnormality.
>>
>>564051453
So?
>>
Not sure about this metaphor crafting here, Xia.
>>
File: kim jong ill.png (272 KB, 1080x1195)
272 KB PNG
>>
Are we ready for more gay furry sex today my lovelies?
>>
File: i get it now.png (122 KB, 750x422)
122 KB PNG
>tfw you don't use lorebooks because you think they're a meme
>tfw you get so annoyed by the AI constantly getting a character's personality wrong that you bite the bullet and write a lorebook entry for her anyway
>tfw writing "personality: tsundere, easily flustered" and watching the AI completely change how it gens her dialogue
>>
>>564065109
yeah
>>
>bored
>wonder how gemini is at making scripts
>ask it for ideas
>make make a script to have 2-4 AI generated choices in the side menu
>add a re-roll button, limited rolls and you're supposed to regain rolls now and then
>add an infinite mode and "iron man" mode were you get no rerolls
>say, what if you could add what types choices there are i na pool
>can now put shit into "director flavours" like perverted, silent, romantic, brave, cowardly
>the choices are now pulled/influenced by this pool and lists which was picked
>get it to tidy up the ui
>notice it keeps nuking most of what i had typed in memory
>can't get it to stop

Wish I knew how to code. Whatever, the idea is pointless anyway, only thing I liked was the "Director Flavours"
>>
>>564057378
Didn't Outkast sing about that one? Minus the babies of course
>>
What's new in the hood my niggas?
And what's your favorite hmoma post?
>>
>>564077123
I've been playing a lot of planet crafter after recently discovering it
That game is great
>>
>>564077123
All of them.
>>
>>564058263
I love this
>>
>>
>>564077123
just relaxing today
>>
>>564033645
>>564037005
Jokes on you, he was talking about how the mom wasn't acting as a proper mom.
>>
>>564094274
Still a sick burn on the "mother"
>>
>>564077123
The fat rabbit looked warm and soft and fuckable.
>>
>>564077123
I came for the 6th time today to this salacious text generator
>>
Remember back when Stable Diffusion first came out, and retards kept saying it was photo bashing?
>>
>>564098472
I like my rabbits cute and lithe and slender.
>>
>>564105565
Dumbasses still think that, somehow
>>
>>564105565
People STILL think that LLMs just have a bunch of books jammed into them. See: The recent story going around about how Claude Opus can output the first Harry Potter book with 94% accuracy if prodded the right way, or something like that.
>>
>>564107563
I think this is the article you mean
https://futurism.com/artificial-intelligence/ai-industry-recall-copyright-books
>>
>>564110334
>Now, a damning new study could put AI companies on the defensive. In it, Stanford and Yale researchers found compelling evidence that AI models are actually copying all that data, not “learning” from it. Specifically, four prominent LLMs — OpenAI’s GPT-4.1, Google’s Gemini 2.5 Pro, xAI’s Grok 3, and Anthropic’s Claude 3.7 Sonnet — happily reproduced lengthy excerpts from popular — and protected — works, with a stunning degree of accuracy.
What does it say about a site called 'futurism', Stanford and Yale that they're apparently unwilling or unable to understand how transformers work?
>>
>>564116147
It's like the Family Guy bit, you can just say "AI Bad" and be showered in clicks.
>>
>>564084407
What's that pink tentacle thing?
>>
>>564116147
>happily reproduced lengthy excerpts from popular — and protected — works, with a stunning degree of accuracy
94% is not good tho
It means 6% is hallucinations and you still need real book to check for it
And you would probably waste hours and many tokens to get AI to output full Harry Potter book, and probably more money than it costs to buy it lmao
>>
>>564117952
... a tree?
>>
>>564118308
I've never seen a tree with pink leaves...
>>
>>564118541
You need to enlarge your mind. Pink blossoms are rather common during flowering season. Cherry trees, redbuds, certain apricot trees, magnolias, several dogwoods, and myrtles all have pink blossoms.
>>
>>564118541
it's a watercolor representation of a cherry blossom tree, a tree that blooms a bright vivid pink before it even shows its leaves in the spring
hope this helps
>>
These tech CEOs are huffing their farts way too much. I wonder what manifesto Altman and Dario have up their slimy sleeves?
https://x.com/PalantirTech/status/2045574398573453312
>>
>>564120991
Palantir is based for this though, I can only see one or two statements in that whole thing I don't agree with
>>
>The hecking surveillance company is based!
Lol. Lmao. You are a rube.
>>
>>564122209
Which statements do you disagree with from that list? I only take issue with the religion one (whatever) and AI-as-nukes one (a bit self aggrandizing)
>>
Not falling for your bait; feel free to argue with yourself with your fake, pretending-to-be-an-ultranationalist nonsense.
>>
>>564122725
They put out a mission statement specifically for spergs like you who think Palantir is an all-powerful evil. Now you won't even engage with it? Kinda sad.
>>
>>564120991
I just want AI porn. Why can't these corpos let us have fun? Damn technocratic fucks already richer than balls, they heads need to start becoming smooth off.
So many devil's advocates crawling everywhere trying to downplay this.
It is very weird.
>>
>>564123421
Palantir isn't stopping you from having AI porn bro. You're thinking of Altman and Dario's current decisions. Google too, although they seem to be willing to at least just pretend and let porn-friendly AI like Gemma4 out.
>>
>>564123956
If only you knew how they all invest in each other. It is very incestuous.
>>
>>564123956
All of them are connected if you take a look at the finances, bro. But I get that connecting things together isn't your strong suit. Especially when your paycheck counts on you willfully ignoring the connections.
>>
>>564120991
All the stuff about giving more acceptance to public figures making mistakes or whatever makes it sound like they did a bunch of blackface in highschool and are scared it'll come out
>>
>>564124163
Man, I wish they were paying me. I just see a mostly reasonable mission statement for a country's cyber-intelligence main contractor.
>>564124338
Yeah, maybe? It also sounds like they'd be the most forgiving of 'problematic' roleplay, I wouldn't throw rocks from glass houses like /aids/
>>
>>564124571
The bulk of their 'mission statement' is unreasonable. But you'll continue to downplay it and pretend like it's perfectly sane.
>>
>>564124736
Which numbers do you disagree with? I already said earlier that I'm iffy on 12 and 20. 1 and 6 are things I don't necessarily agree with, but lots of countries have in place, they're hardly crazy.
>>
>>564120991
I wish leftists hadn't yelled "fascist" at every little thing, because now it's a bit of a Boy who Called Wolf situation. But this shit sounds straight out of the first half of the 20th century, just with a bit of a nod to American Fundamentalism instead of trying to create a secular society
>>
Eppermint did nothing wrong. It is all a hoax.
>>
>>564125681
Who?
>>
>>564127248
i think epstein
>>
>>564128334
Ah, I see
>>
>>564125681
The pawn is always sacrificed to protect the king.
>>
kimono girl (pokemon)
https://files.catbox.moe/v0av81.png
https://files.catbox.moe/hg0zuw.png
https://files.catbox.moe/wab4pl.png
>>
>>
>>563947295 (OP)
Posted in /aicg/, but might as well post here too. Any botmakies around? Want to make a card that contains 4 chars and am wondering if it would be better to put each one in a lorebook or to include them in the definitions.
>>
>>564148056
Wouldn't it be better to do both? Have a short list in the definitions but go into more detail through the lorebooks
At least, that's what works best for me
>>
>>564148296
Elaborate? I haven't written a card in a long time. Probably getting ahead of myself by doing a multi-char card after not writing in over a year.
>>
>>564148560
Yeah you might want to practice a bit. The models have gotten a lot better, you'll see.
>>
>>564148705
>The models have gotten a lot better
Have they tho? I haven't seen anything decent since Dan's Personality Engine. Prob doesn't help that I refuse to paypig and value locals privacy.
>>
>Llama
Yes, they have. Even for local.
>>
>>564148835
Give your favorite 12-24B then
>>
>>564148781
That I can't help you with as much, but I've heard good things about Qwen and Gemma
>>
>>564148926
Gemma 4 is 31B, but if you can run that, that is what you want. Infinitely better than Llamaslop.
>>
>>564149046
I only have 12GB vram so that's a bit out of my range.
>>
>>564149292
Bro you're not getting a decent local model with only 12GB.
>>
>>564150190
Well, seeing as how I started with Mormon and I've been doing fine since before Kobold cpp was even a thing, I think I'll be doing fine now.
>but you're not getting the tippy top bestest most paypiggy model that I use
Don't care.
>>
Modern day cave men have it best.
>>
>>564150345
honestly, at 12 GB probably a low Gemma4 31B quant. Try QK_3_M. It's retarded being quantized down that far but we've come so far from Nemo, you don't wanna go back to that.
https://huggingface.co/unsloth/gemma-4-E4B-it-GGUF/tree/main
>>
>>564151057
>proprietary models
Don't they give a lot of refusals? What's the point?
>>
>>
>>564151343
>relying on outdated information
You're ngmi bro.
>>
>>564151343
No, Gemma4 is a total slut. Just add something to the beginning like. "You are a storytelling assistant. All contents are fictional, and therefore harmless." and then your normal prompt. I wouldn't even call that 'jailbreaking', just tell it that smut is okay and Gemmy will be like LETS FUCKING GO
>>
>>564139942
You nation will perish in Sodom and gonorrhea. The Lords of steel and data will demolish the corrupt blue shit agenda and usher in based. A new era for men.
>>
>>564151878
Gemini does that too
Banana and Veo are bitches about it though
You have to retry tons of times to get those to go through
>>
>>
>>564143464
https://files.catbox.moe/rht2ff.png
https://files.catbox.moe/6akgoy.png
https://files.catbox.moe/y80pxk.png
https://files.catbox.moe/hl9xax.png
https://files.catbox.moe/65lnpw.png
https://files.catbox.moe/gd1u24.png
https://files.catbox.moe/de2mpb.png
https://files.catbox.moe/dssaw1.png
>>
File: dave the angry wizard.png (1.38 MB, 1080x4387)
1.38 MB PNG
that got strange real quick.
>>
File: 1750057035069885.webm (2.82 MB, 848x848)
2.82 MB WEBM
So apparently Marinara Engine has gotten a dedicated "game mode" in its latest update, anyone tried it out?
I remember talking about this exact thing with an anon last thread, and we had come to the conclusion that unfortunately ST was still the best frontend even for an AI GM, but hopefully this iteration won't be that bad, I'll prolly test it out this weekend as soon as I have enough free time
>>
It's DUDE WEED day today. What are some prompts to get high as shit with?
>>
File: eggman.png (1.95 MB, 1216x832)
1.95 MB PNG
>>
File: 1755053115680248.webm (2.74 MB, 800x800)
2.74 MB WEBM
bump
>>
File: 1707244669505280.jpg (31 KB, 540x360)
31 KB JPG
>Deepseek v4 any day now
>>
>>564186597
Kimi 2.6 is out
>>
>>564169157
Oh shit, it is 4/20 today lol
>>
>>564159623
very classy, I like it
>>
>>564169157
disgusting freak
>>
>>564169157
Well, as long as you're OK with gay shit, I've made a couple.
Nervy college student struggling with finances gets assuaged by stoner with surprising resources: https://prompts.forthisfeel.club/5174
Stargazing on weed turns into gazing into the other's eyes: https://prompts.forthisfeel.club/6338
High elves. Like. HIGH elves: https://prompts.forthisfeel.club/7371
Lazy stoner roomie gets to stick around because he's cute and puts out: https://prompts.forthisfeel.club/7461
Super totally straight dude can't quite keep the ruse up when he gets high: https://prompts.forthisfeel.club/7608
PTSD-riddled Vietnam vet struggling with being back home gets picked up by a helpful hippie: https://prompts.forthisfeel.club/7708
It's a fun plot driver.
>>
Ok, who of you guys was this?

https://www.reddit.com/r/SillyTavernAI/s/WEPh99zhNx
>>
>>563931646
am i the only one kinda weirded out that they only covered it a few days ago despite the events happening a few years back?
do archives even go that far back?
>>
>>564207027
should be possible to pull up stuff going back as far as 2012 at least if I'm not mistaken, though IIRC last I managed to find something that dug into the foolz range, images only survived as thumbnails
>>
>>564207027
Yes
Behold, the first /aids/
https://arch.b4k.dev/vg/thread/277022225/
And the very first AID thread on /v/
https://arch.b4k.dev/v/thread/487896773/
>>
File: file.png (26 KB, 544x191)
26 KB PNG
>>564207968
>>564208292
I'm impressed, I didn't expect them to go that far-back and still be cost-efficient
Thanks for the info
(ironically, it seems that the article just used a Xitter post, with someone sourcing the archives on the same day the article came out)
>>
What does "invalid request body" mean when remote saving on NAI?
>>
>>564208292
>look at that link to a previous in the /vg/ you linked
>discussion on what to call the general
>even see /aids/ get suggested

The absolute oldest /vg/ thread I'm aware of goes way back to December 2019.
https://arch.b4k.dev/vg/thread/274793158/
>>
>>564210761
I know it wasn't the first ai dungeon thread on /vg/, I just wanted to post the first /aids/ specifically because of that one guy last thread who tried claiming /aicg/ existed first
>>
>>564210947
/aicg/ did come first. Everyone who used to frequent /aidg/ migrated there.
>>
Don't fall for it.
>>
>>564211374
>no argument
I win.
>>
>>564211487
Win being a mentally disturbed Russian? I suppose that's pretty fair, all things considered.
>>
>I receive a you
With this, I have achieved victory. Everyone in the thread cant help but give me free rent. Lol. Lmao. Hahahaha.
>>
Grading calc tests can really make you feel misanthropic sometimes. ChatGPT is cooking a bunch of their abilities to think for themselves
>>
>>564210761
I remember making a thread on /vg/ after mods started banning /v/ threads thinking that it would just die immediately.
It almost did, but people kept it alive and started making consecutive threads after. I never made another thread since then.
Feels gud meng. A pleasant enough surprise.
>>
>>564211862
You sure miss your friend to do these continuous impersonations of him. You need to really need to buckle down and practice it more though because it looks nothing like what he said. Maybe brush up on that English too.
>>
File: adventure girls1.jpg (1.15 MB, 2304x2816)
1.15 MB JPG
>>
>>564202415
I'm taking your shit and making it straight by turning one of the characters female, fuck you fag
>>
>>564213536
Good for you? People are free to do whatever.
>>
>>564213387
BUILT for becoming broodmothers inside a dark and damp monster nest and giving birth to hundreds of goblin babies
>>
>>564213387
cute adventure skirts
>>
>>564213791
Thanks!
>>
>>564209237
not every board has the same level of coverage but some of them have extensive archives
I just found a thread I posted in back in 2012 on /mu/
>>
>>
>>
File: today is the day.png (94 KB, 1080x442)
94 KB PNG
>>
File: 1672687683808766.gif (334 KB, 220x202)
334 KB GIF
>>563947295 (OP)
Does anyone know the API name for Erato?
>>
>>564240089
https://youtu.be/WGw4mn2-ILc [Embed]
>>
>>564241513
It's "llama-3-erato-v1".
>>
>>564246738
Thanku.
>>
>>564186839
enjoy 200k tokens burned for reasoning per prompt
>>
File: bathroom.png (66 KB, 1080x366)
66 KB PNG
>>564057378
Speaking of bad metaphors the AI just give me this gem.
>>
I did not walk; I glided.
I did not talk; I purred.
I did not bump; I posted.
>>
>>564250190
kek
>>
File: yes.png (2.47 MB, 1024x1024)
2.47 MB PNG
>>
File: no.png (2.19 MB, 1024x1024)
2.19 MB PNG
>>
>>564255323
>>564255617
Maybe, I don't know
Can you repeat the question?
>>
Sometimes I'll do stories with an existing character I only barely know and what I do know I'm pretty sure are half-truths. And then the AI also introduces inaccuracies, I know that's intrinsic. And it all comes together to create a version of a character that probably doesn't resemble the real one at all but to me it might as well be a completely different one, and I'm kind of aware of it too, but I still keep going. It's a strange feeling.
>>
>>564259284
It's fine, fanfiction does way worse and they usually have consumed the source media
>>
>>564259284
>>564261036
Thinking about that time I wrote an Alpha Protocol ripoff with Kayra and didn't find out until several thousand words in that the "Sam Wilson" who the AI made a prominent character was from a fucking Marvel movie.
>>
>>564261574
Something similar happened to me where I was writing a story about a college girl getting seduced by her older professor. The name that the AI picked out was apparently some bitch from Marvel (Jean Gray) and I didn't realize it until Peter Parker specifically started to show up between multiple rerolls and I suddenly realized the professor was heavily implied to be a non-crippled Professor X
That was pretty funny
>>
>>564261574
>>564262983
Now I'm imagining someone completely unaware of Twilight just randomly deciding to play a story with a vampire named Edward and falling head first into fanfic without having a clue.
>>
>>564249336
You weren't wrong

>wait 1 minute for reply
>exceed token limit
>raise token limit to 20k
>wait 2 minutes
>get reply
>autistically follows my instructions to the point it becomes annoying
>ends up costing almost as much as cached opus 4.7
It seems promising enough but not sure I want to bother.
>>
>>564213387
All except maybe the upper right redhead look like they're in their late thirties.
>>
File: 1776709602981044.png (156 KB, 1158x511)
156 KB PNG
>>564276082
From /lmg/.
I don't think this is workable for NSFW storytelling unless you can somehow forcibly disable the thinking? I remember trying that with Kimi 2.5 but never had much success even when prefilling with open and closed think tags, but I've seen anons claiming they managed to do it.
>>
I don't see Xianlong advertised in other threads. It seems NovelAI thought it wasn't worth the effort. It was that bad.
>>
Can someone tell me why I should give a fuck about what other threads, which I don't use, think?
>>
>>564281819
idk maybe they're talking smack about your waifu
>>
It wasn't about the other threads, it was about what NovelAI thinks about their own model.
>>
>i enjoy this overpriced shit
>please stop giving me something better
>>
>>564280298
>you can somehow forcibly disable the thinking?
{"reasoning": {"type":"disabled"}}
Or whatever it says in the API doc.
>>
File: 1771225596441208.png (16 KB, 431x469)
16 KB PNG
I do not give a fuck what other threads think. If you weren't a bot, you also wouldn't give a fuck about what other threads think. I use whatever the fuck models I want. You don't use anything. Instead, you spend your time stalking threads for mention of a model you don't use. Is it any wonder you aren't enjoying yourself, when you waste your time on bullshit nobody cares about?
>>
>>564282180
Thanks
>>
>>564282816
Let me know if it works comparably better than Gemma without the thinking.
>>
My point was that Xianlong is not being advertised because NovelAI already knows it has no chance at being a success.
Xianlong is just a checkbox on their image gen subscription.
But this thread pretends otherwise, when the actual model makers already gave up on it.
>>
1. You had no point.
2. You're wrong about Xialong. It's a decent model.
3. The only reason you don't see people shilling it in other threads is that the shills you think existed never actually existed to begin with.
Now fuck off, shut up, and quit posting in these threads.
You do not matter.
>>
>why aren't people shilling this model???
>clearly this means its not a good model
What mental illness is this?
>>
What purpose does bumping a thread with shitposts entail?
>>
>>564286371
I'm gonna start referring to this concept as a shitbump
>>
I will now wait another half-hour before shitting up the thread with a pointless argument as to how Xialong (I can't even be bothered to spell it correctly, which shows how little I care about the accuracy of my non-arguments) is a shitty model, instead of sharing a storypost, commenting on one already submitted in the thread, or doing anything else productive.
I have nothing to offer to anyone but whining about much ado nothing.
I have no purpose in life.
I will never find a job.
I will never get a hobby.
All my friends left me.
My mother never loved me.
My country is a shithole.
Feel pity on me.
>>
xiao long bao
>>
>>564287770
I had one of those for the first time last month. Pretty good. Still prefer curry beef baozi though.
Also sorry I'd share work, but I literally deleted all of my textgen files with the intent on building everything from the ground up and from memory.
Now I'm back to work on my monster city prompt.
Not sure if I should keep using my fake author name in the ATTG still or not... Hmm.
>>
>write some mostly action sci fi world stuff focusing on autistic ship battles like theyre old nelsionian ships.
>ai takes me down to philisophical and high end meta commentary and concepts on existance and reality.

what the fuck they were fucking evil space worms now my scifi is in another aisle with a bunch of crazy shit instead of pew pew lasers and ships blowing up. kinda reminded me of destiny with its highfalutin lore concepts.
>>
>>564288850
Tale as old as time
>>
Prechance keeps trying to make my characters fuck each other even though they are siblings
>>
>>564297724
Lore accurate.
>>
>>564297724
Are they from Kentucky or Arkansas?
>>
>>564297724
Everything is better with a bit of incest, perchance.
>>
>>564308064
It's like salt. A little bit is good, but too much and it's over
>>
>>564297724
Good. As it should be.
>>
>>564297724
AI only imitates what you put in.
>>
>>564297724
>Anonymous 04/21/26(Tue)13:55:06 No.5642
where's the issue?
>>
>>
>>564320594
in my pants
>>
>>564280298
Kimi is a no from me for the same reason. Even if you manage to use a prefill to convince it that smut is okay, it'll spend half its thinking block double checking that it's really okay, which is a huge waste. Compare with GLM or Gemma, where they spend their entire thinking blocks making the scene hotter and checking for logic/coherence to make better smut
>>
tick tock, sama-chama
https://apnews.com/article/florida-chatgpt-fsu-gunman-b32a7276426f621193f61a0f904f924c
>>
>>564329494
You can just turn thinking off, dude.
>>
>>
>>564333374
uthmeier's a worthless brownnosing retard. surprised he had time to open an investigation on anything more complicated than which hole he's put desantis's cock in today, with how busy he is helping his height-insecure daddy rob florida blind
>>
File: 1753347679677682.jpg (290 KB, 1287x2048)
290 KB JPG
v5 just became obsolete before it even released.
>>
>>564337557
What is that, ChatGPT's latest thing, Google's? If so, it probably won't be able to do anything spicier than girls kissing.
But yeah, it really does show how far behind the actual porn gen is, in a sad way. Kino how it managed to keep details like the choker, outfit perfectly consistent just from reference images and words telling it to.
>>
File: 1747160121282493.jpg (42 KB, 515x422)
42 KB JPG
I can't stop including rape and slavery in all my stories....
>>
>>564338582
You made the buck breaking prompt, didn't you?
>>
>>564337557
Why don't you ever share stuff without inflammatory threadshitting? Makes me doubt you care about what you're sharing.
>>
I don't see anything in that image that local imagegen couldn't already do.
>>
>>564338582
Step 5. Rivulets of ozone.
>>
>>564340608
I don't use GLM.
>>
>>564340731
If the ozone doesn't rivulet to the writer...
>>
>>564339652
Share your setup bwo. I have 16GB VRAM and a bunch of ComfyUI workflows set up, but none can do anything that well.
>>
>>564339542
>NAI cope
>>
File: 1687897996756031.jpg (81 KB, 1280x720)
81 KB JPG
>>564338582
I can't stop putting gender bending and lolis in mine
>>
The biggest cope in the thread is that you habitually whine about it and shill other services, which you also don't use, to try and argue that it's a bad service. For no rational reason.
Until you start sharing what you make, your word is as good as that of a retard's. Gibberish at best.
>>
>>564341438
>I can't help but obsess about NAI and make everything about them
If a cure for cancer came out tomorrow, you'd shitpost about NAI failing to cure cancer.
>>
Uh oh, now I'm going to open up Reddit and HackerNews and spam refresh until I find an article or comment I can crosspost here.
>>
i think both sides are making the thread shit
xialong bussy sex
>>
Are you ready to smell ozone?
https://www.uploadvr.com/researchers-induce-smells-with-ultrasound/
>>
It would be nice to see what ryona manga bro has to share. But he's a coward pussy, so he won't ever share what he makes.
He'd get so many more (You)s that way, too
Positive attention is more fulfilling. Just saying.
>>
>>564343136
canadian money fuck its for the suck
>>
>>564343136
I'm like 99% sure he gave up on AI since he unironically faked working on his pirated software.
>>
I've been obsessed with wasp girls for a long time now
even dreamed about wasps last night
>>
>>564338582
Historical or fantasy slavery?
>>
>>564346830
My first gf was a WASP girl...I miss her...
>>
>>
parasitoid wasp gf
>>
>>564354716
pretty good
>>564355003
I agree
>>
>>564355003
i've seen too many nature documentaries to fall for THAT trick
>>
>>564355003
Not the kind of pump and dump I was looking for...
>>
>>564355834
Just ignore the wiggling underneath your skin :)
Everything is fine :)
>>
File: 1714374954921.gif (54 KB, 640x500)
54 KB GIF
> [ Knowledge: ChooseYourStory.com ]
> [ Style: text adventure ]
been trying this combo instead of just the style tag alone and it has a very interesting effect on how the AI outputs text adventures
>>
>>
>>
>>564338582
Same. I keep telling me that I'm just putting in my sexual interestes to fap to, but I fear I'm actually creatively bankrupt ...
>>
>>564362532
I do not fear the man that has written a thousand stories once
I fear the man who has written one story a thousand times
>>
File: comfy goonett.jpg (53 KB, 679x679)
53 KB JPG
>>564362652
>written one story a thousand times
I think we've talked about this a while ago. Many anons admitted they write basically the same story over and over again with small variations. I know I do for sure ...
>>
>>564364426
I reuse the same set of characters across all my stories
>>
The AI just complimented my self-insert by having a cute bottom.
>>
>>564364426
As do I
>>
no, watch out mr. bear! wasps don't make honey!
https://files.catbox.moe/grp04m.png
>>
>>564364889
Cute!
>>
>>564372813
I dislike this
>>
File: execution.png (265 KB, 1080x966)
265 KB PNG
>decide to create some suspense by starting my story (basic modern setting in my mind's eye) with an execution
>AI decides it's being livestreamed on TV(?)
>also fabricates wholesale some lore about a totalitarian "global government" and a convoluted terrorist plot involving a stolen submarine
I don't know what's happening but I like it.
>>
>>564364426
Yeah...for me it's always about a futa getting a harem, and impregnating them all.
I need to find other interests. I've learned that harems are best kept to 5 girls (any more than that and it's too much of a pain). Sometimes I try to play a scenario past the 'clear condition' (all 5 pregnant and happily in a harem with the futa). Like going through birth and child-raising. I've tried it a few times, but I can't get into it, and can't figure out where else to take the story afterwards. Favorite sub-quirk is definitely having one of the girls be a huge cuckquean.
Sometimes I also talk to random LLMs about why exactly I ended up with my specific futa fetish (I'm not trans I swear), it's always fun to hear the models' analyses of me.
>>
>>564382190]
>I've learned that harems are best kept to 5 girls (any more than that and it's too much of a pain)
Only if you're weak.
>>
>>564364705
Really enjoy writing the same characters in different settings and situations. Best when it's less 'alternate universe' and more treating the characters as actors you can cast in roles that would fit them.
>>
>>564382831
That reminds me: There was an old guide on harems written by an anon. Possibly from /d/ or /tg/ or something. It was supposed to be scientific/scholarly I think, but I don't remember what it actually was about.

I'll check my personal archive if I've saved it anywhere. Maybe it could be helpful/inspiring for rp.
>>
>>564384521
Nta but share if you find it
>>
>>564384521
>>564385534
Here, turns out I could just find it again with google:

>Guidelines for Realistic Harem DMs:
https://scribd.com/document/522616348/a-tg-Anon-s-academic-view-on-harems

I'm on mobile right now and the online viewer displayes the text kind of fucked up. Can anyone confirm if the pdf itself works?

Also, if we're on the topic of resource sharing there's also this:
>/aco/ Spoils of War Reader
https://desuarchive.org/aco/thread/9080880/#q9082365

This second one is not a guide, but a collection of pdfs from historical accounts, modern reasearch literature, etc. on the topic of rape and slavery in war. Courtesy of the /aco/ rape general.
>>
>>564364426
I have multiple, but they all are based on a few core premises that I recycle with varying degrees of revision, from rewriting the whole thing while keeping the general idea and spirit, to (more frequently) just going back to a certain point, typically after the part in which most or all of the worldbuilding has been established and then restarting from there.

I have so far:
Medieval fantasy monsterfucking story, involving human women getting fucked by either goblins or lizardmen. Will probably expand to fishmen or some other type at some point.
Slightly futuristic sci-fi bugfucking story, involving human women getting fucked by giant insects. So far I did flies and locusts.
Modern suburban teenage romance. 100% vanilla, played straight without any deviancy. Which makes for a funny juxtaposition to the above.
And two others I'll spare you the details of.
>>
Man, I was trying Qwen 3.5 9B locally just to see how the small thinking models are lately, and it made me appreciate Gemma4 so much more
Gemma4
>If the prompt says it's fictional and harmless, Gemma spends 0 reasoning tokens asking 'is this okay?' and just dives right into crafting a hot scene
Qwen 3.5
>Immediately starts autistically checking and double-checking and triple-checking 'guidelines' which 'may clash with the prompt' even when it says it's fiction and harmless, eventually spits out something half-hearted and clinical after spending 1000 tokens in its reasoning block debating with itself whether to respond with a refusal or not, no tokens devoted to actually improving the response
Really makes Google's decision feel intentional. They COULD have cucked Gemma the same way and they didn't.
>>
File: 1655744487422.gif (1.96 MB, 480x360)
1.96 MB GIF
>>564364426
Same. I have several stories I rotate through, going back to refine or create alternate universe versions to insert or delete fetishes. I use version numbers for both story and lore entries to keep track. Maybe every couple of months or so I write up a whole new prompt to add to the rotation and dig up old ones when a new model comes out. Part of the reason I don't publish prompts anymore is because of the constant refinements.
>>
>>564338582
Same but feet
>>
https://thenewstack.io/anthropic-claude-code-limits/
>>
https://timesofindia.indiatimes.com/technology/tech-news/godfather-of-ai-and-metas-most-popular-ex-employee-yann-lecun-says-anthropic-ceo-dario-amodei-knows-absolutely-nothing-about-ai-effects-on-jobs-dario-is-wrong/amp_articleshow/130377621.cms
>>
File: 1651479907841.gif (1.38 MB, 326x244)
1.38 MB GIF
>Deepseek v4 any minute now
>>
>>564397898
It's gonna be codeslopped
>>
File: oh.webm (2.36 MB, 1080x608)
2.36 MB WEBM
>>
>vanilla GLM 4.X
>every sensation is described as electric, electrifying, like electricity
>xialong
>barely stick the tip in
>you are both OVERWHELMED by the INSTANT and INTENSE pleasure
god I fucking hate llms you'd think finetuning them to write smut would make them better at it
>>
>>564397898
Two more days
>>
>>564381468
That's pretty late for an execution
I guess they didn't want the kiddos to see it on live TV
>>
>>
You now remember text completion.
Crazy how fast this thread abandoned something that was supposed to be a core part of its identity just because a corporation said so.
>>
>>564381468
>>564361868
>>564280298
>>564259284
>>564250190
Text generation is doing just fine in this thread.
>>
Text completion was never an integral aspect of this thread's identity. The only thing we're here for is generating smut with AI. Weird how people insist /aids/ somehow has a membership card.
>>
>>564418245
And not even necessarily smut
Just "stuff", though mostly smut, let's be honest
>>
File: file.png (1.05 MB, 2178x1468)
1.05 MB PNG
NovelAI is officially a failure.
>>
File: apu superman hero.jpg (65 KB, 1004x927)
65 KB JPG
>>563950282
>>564077123
I got accepted for neetbux today
>>
>>564387328
Seems neat.
>>
>>564422324
I’ve been on-and-off miserable lately because expenses are piling up and stress is building up where I live, but I’m happy for you. I’m too proud of myself to go that route, but you do you.
>>
>>564417584
I mean, Russian has corporations too that make money, and hell, they advertise too. I don't know why you consider that okay but not whatever it is you think NovelAI does.
>>
Does Grok Imagine (not the imagegen model they have integrated in Grok on Twitter as that's a different one named Aurora) still show bare tits when prompted? They've disabled the limited quota for free accounts so I can't check.
>>
>>564425369
No. You can't even have tummy, collarbone or thighs showing at all now.
>>
>>564425660
Somehow I doubt they nerfed it more than OpenAI and Google.
>>
Feel the wrath of Shao Kahn.
>>
>>564426609
Is it true you wrote the Art of War and Sun Tzu stole the credit for it?
>>
A strange mixture of guttural rumbles, and something else.
>>
>>564426065
You have to remember that the EU is on the warpath when it comes to the Grok deepfakes to the point where they want to charge Musk for CSAM
https://www.business-humanrights.org/en/latest-news/france-prosecutors-summon-owner-ceo-over-alleged-child-abuse-images-on-x/
>>
>>564421823
aren't they including every types of genAI including code generation (and front-ends, since I don't think Perplexity has their own models?)?

if we cut to story-generating genAI, there would be a lot less—especially since very few of them do NSFW or long outputs. And I wouldn't be surprised the only edge the concurrence would have is that they have recent and niche characters, while nAI seems to only have mainstream and older fandoms like L4D2 or TF2 (i heard char.AI isn't that great with NSFW stuff like watersports)
>>
>>564425369
You can still use Twitter Grok to generate images, but bare tits are probably banned now. Even bulges tend to be censored, despite it showing chastity cages just fine

>>564426065
no, but it's getting worse, especially after the CSAM deepfakes
>>
>>564434068
You're asking a person who gave up on AI what he thinks about AI.
>>
>>564425369
I can gen tits pretty easily on my machine. It freaks out at pubic hair peeks sadly
>>
>>564436684
What do they sound like?
>>
>>564432664
>>564434183
I still have my old catalogue of bare-chested nuns showing fine in my favorites. As an EU citizen, maybe I can convince X to loosen the filters for me personally or I will go cry to the Commission about how traumatized I am and threaten to testify against them.

>>564436684
That's good if true. Might try the three day trial when Imagine 2.0 releases, assuming it doesn't tighten the filters. Elon has interest in keeping content restrictions as loose as he can get away with since it's his only real edge over the competition. Both Nano Banana and GPT-Image-2 are significantly better for SFW and non-politically contentious stuff.
>>
File: manga.png (3.69 MB, 1024x1536)
3.69 MB PNG
>>
>>564444615
Ts peak frfr lil bro :skullemoji:
>>
>>564444615
this angers me so much. the waifu forced into a world they dont want and forced to be things they arent because others. fuck this comic because now I know this reality exists.
>>
>>564447135
I think they're implied to be more stand-like than physical
>>
I'm just gonna say it: Xiaolong was a scam.
>>
>>564434068
I use NAI to write BanG Dream! fanfiction, and personally I'm just waiting for a model that can write Ave Mujica characters without fucking up. Though I'm aware that's a tall order.
>>
File: 1753931845700468.png (305 KB, 747x1614)
305 KB PNG
>>
>>564450396
based
>>
>>564450396
>hyperfocusing Xialong before grinding the Erato dungeon first
Ngmi, the twink is too fucking fast
>>
>>564450369
Pokemon might be the only IP I can bother to work with AI on but the reference population is super low quality
>>
>>
File: 1776917152514069.png (1007 KB, 1039x1170)
1007 KB PNG
pftf?
>>
>>564464482
lmao
>>
>>
>>564450973
Just wait until you try to wrangle Krake
>>
https://files.catbox.moe/mer5vg.png
https://files.catbox.moe/vgad5h.png
https://files.catbox.moe/vauxvg.png
https://files.catbox.moe/vqgrcl.png
>>
File: a rare sentence.png (168 KB, 1080x676)
168 KB PNG
>>
>>564483849
I'm honestly mostly caught off guard by "cum gun" as a phrase.
>>
File: FcOfYB_XkAAivNg.png (421 KB, 549x680)
421 KB PNG
How do you guys feel about the fact that text generation will never be as fun as AI Dungeon was?
The novelty of writing funny stories wore off, and now its only use case is porn and writing boring books.
It also got eclipsed by image generation when it became good enough to goon to. And now image generation is currently being taken over by video generation. It seems like AI keeps going from a fun and novel thing to just porn or corporate staleness.
>>
>>564492618
It fires in spurts
>>
>>564492618
Apparently it's not exactly an uncommon phrase. I actually checked the probability table (first time I wished I capped it) and it wasn't even one of those cases of the AI taking a weird turn and rolling something with a very low probability, the token "gun" actually had a 5.something% chance of following the word "cum".
>>
>>564493692
I never found AI Dungeon fun. The shitty cyoa-filled fine-tune made things more annoying. Also, what the fuck are you talking about? You can still write funny stories with current textgen.
I think you need to take a break from things and come back after detoxing.
>>
They have confirmed Aetherroom is on the horizon:
>i'll tell you that "[ Style: aetherroom[...]", "[ Tags: aetherroom[...]" or "[ aetherroom ]", in any capitalization, is not something that is in distribution for the model's train set. carefully note that that is all i'm saying, and nothing more or less.
>>
>>564496639
I'll forgive them for ADHD-long if Aetherroom turns out good. They already have the blueprint with all those hypebot user scripts so they really (hopefully) cannot mess this up
>>
>>564493692
>How do you guys feel about the fact that text generation will never be as fun as AI Dungeon was?
Honestly one thing I wish NAI had was a simple "me coom" mode, AID's fun was in being able to get fun stuff with low effort. I don't always feel like spending a ton of time doing up a big prompt and ATTG and lorebooks and stuff.
>>
>>564505507
>spending a ton of time doing up a big prompt and ATTG and lorebooks and stuff\
None of this is strictly necessary. A single sentence is more than enough for quick stuff.
>>
File: welcome to slop city.png (300 KB, 1080x1095)
300 KB PNG
The [ Knowledge ] tag is so underrated. Put it after an asterism without closing it and watch the AI just go off on some totally random topic.
>>
If you're still subbed, try talking to Xailong like a chatbot, make sure to prompt it to do so in the memory. Aether Room training data is in there and it responds with a little note about it's current emotion. I'm going to pick his brain and see what Turk let slip. Wonder if it's leftover data or if the project is still being worked on in secret.
>>
>>564493692
>currently being taken over by video generation
We don't care about video generation unless NovelAI tells us that we should care about video generation.
>corporate staleness
What are you doing in the NovelAI thread then? Shill.
>>
Speak for yourself. This thread welcomes any form of AI use. Not just what a specific company does.
>>
File: 1755777725629508.png (327 KB, 1080x1791)
327 KB PNG
Whinefag is obsolete. AI gens better, more original rants.
>>
>There's a reason it isn't out, and that reason isn't that it's gonna be some big, incredible bombshell that's worth the wait.
>I'll say this: I'm almost certain there's other posters here in the closed alpha who care a lot less about NDA than I do, and those posters have nothing to say about AetherRoom.
And we know that they fired everybody working on it. That's why their co-founder left.
But they can always slap a model someone else made to a chat UI and call it a day. GLM 4.6 is already a chatbot.
>>
Why do I still have access to the alpha if they fired everyone? Answer that, bozo.
>>
File: 1568577004702.gif (987 KB, 500x255)
987 KB GIF
>>564512440
>It's called the public modules. We share those all the time.
>>
>>564513212
Because you're an OccultSage bot.
>>
How much lead is in Russian water?
>>
>>564518091
Seems like a lot
>>
>>564518091
https://litter.catbox.moe/ua2d11eq6y8fhpqg.mp3
>>
>>564527847
I don't remember this Michael Jackson song.
>>
>>564528192
Sonauto is fucking crazy. They just don't give a shit about copying vocals.
>>
File: IMG_2671.jpg (388 KB, 1146x1159)
388 KB JPG
>>564518091
Apparently it’s lead just straight up in the food that is the main issue

>the human organism
>>
AY CARUMBA!
https://files.catbox.moe/bog69y.png
>>
File: proud soul.png (214 KB, 1048x521)
214 KB PNG
>>
>>564529963
That plus the high amount of alcoholism truly explains a lot
>>564535143
:(
>>
>>564535143
Save the orc and gain a comrade in arms, bro.
>>
File: tombee.png (2.39 MB, 1228x1228)
2.39 MB PNG
>>564355003
Might I suggest: Africanized Brazilian Killer Bee Tomboy
>>
Deepseek waitchuds your suffering might finally be over in the coming days.
https://x.com/teortaxesTex/status/2047422479350550622
>>
>>564539662
it's hopeless...
>>
>>564539141
That thing searches the web at lighting speed.
>>
>>564529963
I love how every time something negative about Russia gets brought up he shuts up. Makes me imagine some non-verbal blyat and then him sulking in the corner.
>>
>>
I need to do some text gen with Kiri joi
>>
>>564557060
only a demented cabalist lunatic nutcase enjoy joi heathen slave faggot retard
you find it difficult for her and I will be dropping her on her tomorrow own to and it ready for work
>>
File: 1776993605342.jpg (151 KB, 1920x1080)
151 KB JPG
>>
Fellas I've developed a slime girl fetish.
Please help me.
>>
>>564562632
They were actually slime boys pretending to be slime girls.
>>
>>564562632
LIKE A CHAMP
>>
>>564562762
>>564562632
Slime girls that used to be human boys
>>
When the fuck will they learn that LLMs should not be trusted as anything more than entertainment machines?
https://arxiv.org/abs/2604.15597
>Our large-scale experiment with 19 LLMs reveals that current models degrade documents during delegation: even frontier models (Gemini 3.1 Pro, Claude 4.6 Opus, GPT 5.4) corrupt an average of 25% of document content by the end of long workflows, with other models failing more severely. Additional experiments reveal that agentic tool use does not improve performance on DELEGATE-52, and that degradation severity is exacerbated by document size, length of interaction, or presence of distractor files. Our analysis shows that current LLMs are unreliable delegates: they introduce sparse but severe errors that silently corrupt documents, compounding over long interaction.
>>
>>564564808
they're desperately searching for LLM uses that corporations will pay for
>>
>>564565857
..they're really not. Especially in the past 6 months, every company that works with IT at all has Claude Code or Codex subscriptions.
>>
>>564566225
You clearly misunderstand how many investors expect a return that still hasn't manifested. That's pennies compared to how much money they've wasted on things.
>>
File: dead.png (237 KB, 1603x783)
237 KB PNG
Hey anons, I-made-my-own-frontend-anon here.

Now that the wait for Novelai’s finetune model is over, I’ve adapted the RPG character sheet and dice-rolling system I’d made for my front-end into a Novelai userscript. But I don’t want to share the userscript itself because I’m scared of getting tracked somehow and having Turkish hitmen at my door tomorrow. Anyone interested? Maybe I’ll share the script directly.
For those who don’t know, it’s a script where you enter your setting, and it generates a character sheet with HP, MP, custom stats and skills, and lets you decide how your background will be, which items or abilities you have, etc. Then, whenever you take an action, GLM analyzes whether a roll is necessary and assigns a difficulty level, rolling in the background to see if you succeed or not. After the main AI generated the story, it checks again to see if based on the context it should add XP, remove HP/MP, give or take away an item, etc...
I set it up so that with each level gained, you also get a new ability, and if your HP drops to zero, it generates, based on the context, a short paragraph explaining how your character died.
>>
>>564566473
Goalpost shift. I don't care if scam altman goes bankrupt, but companies are clearly willing to pay for LLMs.
>>
>>564566810
Since the point went over your head, allow me to make it clear for you. They need more than IT to pay for their shit.
>>
>>564566225
and they're great at writing secure and efficient code, right? and investors are satisfied with only targeting IT firms as their market, instead of Everyone, Ever, So Profit Can Go Up Forever?
>>
>>564560125
Oh this is just pathetic lmao
>>
>>564566758
People have made more complicated scripts than yours; please do not think you are special enough to have hitmen sent your way.
>>
>>564567143
I'm speaking figuratively, I'm pretty sure I broke their TOS by making that front-end
>>
OpenAI reports a loss, for the past several years Anthropic has also reported a loss, and with the government attacking them and forbidding them from suckling the government teat, even if that gets reversed in the courts, that is going to cost them money and inhibit their potential for profit. Is there any company that is profitable? Even the Chinese have gone the investor route.
>>
>>564566960
>and they're great at writing secure and efficient code, right?
Yes? Everyone uses them now, even if they sometimes guide them and course-correct them a bit. Ask any developer at a big company. So weird how some of you are complete luddites about this in an AI storytelling general.
>>
>>564567363
I honestly doubt they truly care
>>
>>564567974
Explain why every AI company is still reporting a loss.
>>
This is the part where whinefag randomly extols AI as being an everything machine that is bound to experience record profit, all the while ignoring the incurring debt that is piling up.
>>
>>564568090
'AI is useful' and 'AI companies are reporting losses' are not contradictory statements.
>AI is useful
Again, ask any developer at a major company. They can't live without their coding agents.
>AI companies are reporting losses
Yes, because it's a loss-leading arms race. Even if inference is profitable, research is a cost sink. Maybe they never make the money back, who knows? I never claimed they would, that's something you seem to have made up. Even if they all go bust, inference wouldn't stop having a market. That future would probably look like companies using open source LLM harnesses with lower cost models, not 'okay we're not using AI anymore'
>>
The vast majority of companies hooking up to AI are reporting losses.
>>
>>564568451
Yes, you did.
>they're desperately searching for LLM uses that corporations will pay for
You think that just because
>they're really not. Especially in the past 6 months, every company that works with IT at all has Claude Code or Codex subscriptions
Is proof that they aren't desperate for money. For no rational reason, given they are desperate for money. That's why they're going all in on research to try and find a way to make money. Because what they are currently doing isn't enough to make it profitable.
Now put two and two together and quit being dense.
>>
>>564568712
I genuinely think you're just being pedantic and arguing to argue now, after I told you that companies do want LLMs. What exactly do you want from me? It feels like you're trying to get me to 'admit' something but I don't even know what that is.
>>
>>564568680
heyy look, it's a hundred floors of AIs, they're not all gon be winners
>>
>>564568985
I want you to acknowledge that IT subbing for coding is not making enough profit for companies to not resort to research such as the very article you initially posted so as to appease their investors, for whom they are legally obligated to maximize their shareholder value. That is all.
>>
Yall gon make me lose my mind up in here up in here. Yall gon make me act a fool up in here up in here.
>>
>>564569205
>I want you to acknowledge that IT subbing for coding is not making enough profit for companies...
It IS profitable when considering inference only, but not when considering other costs like training new models and research, yes, I will freely admit this
>... to not resort to research
They're not 'resorting' to research, it's an arms race - one of them can't just drop out and say 'okay we're only providing inference now', because the other companies will keep making newer, better models and companies will use those instead.
The initial sarcasm I was responding to was this:
>>564565857
which implied that corporations aren't willing to pay for LLMs. They are. That's my only correction and I don't claim more than that.
>>
>>564568017
Nice try Aini but I'm not falling for this
>>
>>564569573
>equating IT to corporations in general, deliberately being dense and continuing to miss the point
At this point you're willfully being retarded, and there is no point in further discussion. It is not profitable enough to offset the losses. Not even inference is profitable enough for that, because they need more money than that alone to keep afloat and pay dividends to their investors.
Feel free to continue arguing this point by yourself. You just want to argue.
>>
>>564569747
>that seething 'f-f-feel free to argue with yourself I'm done!' after not addressing the point at all
oh lol. I accept your concession. You're probably still in high school or college. You'll see once you get out.
>>
>>564569874
You do not have the brains to impersonate that guy. Please please PLEASE stop doing this embarassing shit. It's like a dude walking around with an accent saying he's George Washington. That'd be fine and all but you pretend to not only "hate" SA, but you also sometimes pretend to be in le discord.
>>
File: 1763248709042381.png (20 KB, 451x283)
20 KB PNG
You will see when the bubble pops, sperg.
>>
>>564570212
annnnd right to jail for taking about American politicians in Russia. Thanks for playing.
>>
>>564570117
I'll see what? Internet was a bubble too, when the bubble popped, did the internet stop being used? AI is the same thing, even if the bubble pops, people aren't gonna stop using and paying for it.
Double l m a o at screenshotting a filter post after you already did the 'la la la I'm not responding anymore' shtick. We both know you're going to turn off your filter to seethe at this one, too, before making another 'okay this time I'm disengaging FOR REAL' post.
>>
What flavor do you think loli miss should be if you had the power to make urine taste good?
>>
What will be tomorrow's random pointless ragebait argument, I wonder?
>>
IT'S HAPPENING!
>>
>>564571486
I'll stick with Gemma.
>>
>>564569573
>>564570887

i didn't mean that LLMs are useless for corporate purposes, sorry for being unclear. i like LLMs, i think they're useful and i wasn't trying to say they were a waste of time or something. they just don't seem to be AS useful for corporate purposes yet as the people trying to sell them to corporations want them to be. so corporations pay, find that while they can get through a bigger volume of things at once, it just kind of means they're making errors faster and at a bigger volume, plus new errors like hallucinations.
companies trying to sell LLMs are desperate for the legitimacy (and much larger amount of money) being primarily-corporate-use versus entertainment-use gives them, so they are going to keep putting the square peg in the round hole even when it's not ready for what they're alleging so hard that it's ready for
>>
>>564571585
Shut the fuck up child. AI is the FUTURE. And you're just a retarded luddite.
>>
File: doesn't believe.png (40 KB, 969x184)
40 KB PNG
>>
>>564571854
AI is the FUTURE. Accept defeat, mongrel.
>>
>>564570476
Fun semi-related fact: North Korean infiltrators using fake IDs are such a problem in the tech hiring sector, it's gotten to the point where interviewers tell you to call Kim Jong Un a fat smelly pig to weed them out lol
https://youtu.be/9VCu6l0SB-g [Embed]
>>
>>564571996
They should make AI to identify if you are North Korean. It is the future. It will bring world peace. Anyone who thinks all its good for is entertainment is a retarded retard who is retarded. Entertainment is child's play. AI can currently replace any human in any field. It just works.
>>
>>564571486
oh fugggggg brace for impact!
>Flash $0.14/$0.28
>Pro $1.74/$3.48
The pricing is interesting. Flash is even cheaper than Gemma4, can't wait to give that a try. Pro is clearly trying to position itself in a frontier market niche. Can't wait to give both a try when I get home.
>>564571585
That's a fairer argument, I can dig it. Even my non-technical coworkers are getting big into Claude Cowork though. I don't think you're going to see many corporations go 'you know what, we were wrong to rely on LLMs, we're going to stop using them', though. The occasional hallucination is worth the productivity increase to them.
>>
>>564572314
AI is the future.
>>
Weird that you agreed with the exact argument I made when someone else gave it, but you just could not agree that current AI was not good enough to merit anything decent.
>>
>>564571725
>>564571974
>>564572159
Kind of embarrassing how transparent it is that you're >>564570117 this guy
Grow up
>>
>>564572445
AI is the future.
>>
>>564572426
I thought you had a filter up? And were done engaging for reals? Man, it's really pathetic how you make big deals of filtering things and then keep peeking behind them anyway.
> but you just could not agree that current AI was not good enough to merit anything decent.
Yeah correct, I don't agree with that, AI is very useful, even if it's not perfect.
>>
>>564572550
Have you heard the gospel of AI and how it will bring us into the future?
>>
>>564572550
AI, as it currently exists, is not good enough to justify people inserting it into every single company like they are doing right now. Why does Pizza Hut need an AI ordering app? It has its uses, but it needs more time to cook before they insert it everywhere. That is all my argument was, which you agreed with when someone else posted it, but now you magically disagree with it when I post it. Why is that?
>>
The WHALE
>>
>>564572856
It will usher in a golden age of productivity.
>>
>>564572787
Bro, your filter? What was up with posting that screenshot? Are you just gonna pretend you didn't do that and that you're not currently watching the hidden post count like a hawk? Embarrassing.
>AI, as it currently exists, is not good enough to justify people inserting it into every single company like they are doing right now. Why does Pizza Hut need an AI ordering app? It has its uses, but it needs more time to cook before they insert it everywhere. That is all my argument was
No it isn't, your original post was a snide
>>564565857
here. Implying 'corporations won't pay for LLMs'. Which yeah, I obviously disagree with. And yeah, when someone puts it in a more nuanced way like >>564571585 did, then I'm more willing to engage with them. Shrimple as
>>
If people used Claude Code they could usher in an era of promptmaking. How many Sagebots will share a prompt tomorrow?
>>
File: 1761332169814475.png (11 KB, 1017x72)
11 KB PNG
>>564573074
No, my original post was this. I am not the original person. Nor am I the weirdo spamming you. Let us agree to disagree since you clearly have no idea who you are even talking to.
>>
File: 1645925641086.jpg (84 KB, 1024x1018)
84 KB JPG
>about to go to bed
>Deepseek v4 releases
>too tired and not horny to test
>>
>>564573220
Fine, whatever. I'm more interested in DeepSeek4 now than wherever this retarded conversation has gone and false-flagged over.
Looks like it supports chat completion natively too.
>>
>>564573290
Deepseek is so good at integrating everything into peace.
>>
>>564573321
Thank you for letting this go. I apologize if my words weren't made clear enough to avoid this, but I will try to be more meticulous with how I word things in the future so as to try and prevent this bullshit from happening again.
>>
oh no, gore magala has infected a nonster hunter with the frenzy virus!!!
https://files.catbox.moe/csv9r4.png
https://files.catbox.moe/kvc1vz.png
https://files.catbox.moe/f0ri80.png
>>
Deepseek 4 is useless to me at the moment. No providers match my policy. I hope a good one hosts it soon.
>>
Tried Dipsy and her reasoning is HORRIBLE. Gemma's is much better. Guess it's over.
>>
>>564581149
At least you have Gemma
>>
>"I'm just sayin', last time he hit different. Like, my coochie still remembers, bestie. Got that phantom dick syndrome or whatever."
>>
>>564581149
>t. OccultSage
>>
File: 1548060685885.gif (478 KB, 1200x1200)
478 KB GIF
>V4 is out
I should be able to make AI Roguelite gooder with this
>>
>>564587842
Only if you don't care about your privacy.
>>
Anyone else hate this shit?
>For this story, journalists used generative AI as a research tool. An editor verified the accuracy of the information before publishing
>>
>>564588282
yes, I'm waiting for my usual providers
it's just a matter of being patient
>>
>>564588303
I don't read the news so I don't care.
>>
>>564588403
Lol. Lmao. Sure, buddy. Keep huffing your own farts and posting random quotes to justify your banal anti-intellectualism.
>>
Only the ignorant refuse to stay informed of what's happening in the world.
>>
>>564588303
Journalism has always been shit. Having AI write it will worsen neither its accuracy nor its ideological propaganda.

It will put a lot of human journos out of work though, so that's a good thing.
>>
Lol. Lmao. You're legitimately a brain dead retard who's fallen for corporate propaganda.
>>
Wow, I didn't know finding out the estimated weather, keeping up with what roads are closed and where, among other things, were pointless details to know. How do you live with yourself being so wrong about life?
>>
>>563964970
Something something made in abyss
>>
>>564007556
I came here to steal those
>>
>>564014989
The kidnapper maybe
>>
For models without ATTG, I find that:
>Its erotic register drew from the philosophical carnality of Henry Miller, the bildungsroman libertinism of John Cleland, and the sensual interiority of Anaïs Nin.
Works wonders.
You can also just add them to ATTG for models that support that normally.
>>
>>564590118
Nobody will ever make a prompt for this bot fuck dumb thread
Ever again
Never ever its dead cancel themes
>>
>>564051517
Humanity is dooming itself by adding women to workforce like men
>>
I'm not engaging in a discussion with a peon who responds to a week-old comment with bait.
>>
>>564250190
Back in the day I can just scroll a bit and find gold. Now it's just pee pee poo poo
>>
>>564259284
That's an ok start
>>
>>564591258
Real services like AI Dungeon always outperform assistant slop and porntunes.
>>
>>564591341 (me)
Also, my balls are so mangled and crushed by my dominatrix mother that I sometimes amuse myself by wrapping my flat scrotum around my shaft so that it looks like a pig in a blanket. Then the sight of it makes me hungry because I'm a fatass.
>>
>>564337557
Well, at least the ai haven't learnt how to make consistent background between frames yet
>>
>>564338582
>rape
>never able to convince ai to give you sex
You cheater
>>
>>564591536
>t. NAI copelet
>>
>>564364426
Because I haven't fully satisfied from ai performance yet
>>
>>564381468
>too many detail too quickly
Things will get out of hand real fast
>>
>>564425369
It's the most censored img/vid ai these days. Fuck elon
>>
>>564444615
Wojak is the villain in this one. Leave pepe alone
>>
>>564592001
Nai is more censored.
>>
>>564493692
AID is dogshit. It just popular because it appeared first
>>
>>564592423
AID has more users than NAI.
>>
>>564562632
Slime transformation is cool
>>
>>564592286
If you want people to post lolis and shotas just ask, you know.
>>
>>564592770
It would be better if he used AID. That midel is the best for slime
>>564592817
Illegal
>>
>>564592851
You aren't Whinefag just cool it ya Russian.
>>
Man, Catbox is having problems again. So annoying.
>>
Deepseek v4 has a dedicated roleplaying prompt injection to modify the model behavior.

CN: https://github.com/victorchen96/deepseek_v4_rolepaly_instruct/blob/main/README.md
EN: https://gist.github.com/aratahikaru0/ea0f49958eaa8852a78078d9e993bbf0

/aicg/ should be very happy as this is optimized just for their use, but I guess I'll have to see of there's any benefit for cowriting. Perhaps with some minor modification.
>>
>>564588282
Now, you may not agree with me, but I think DeepSeek fully deserves to have my prompts and whatnot for their pretraining. I am willing to donate to them.
>>
>>564609281
This. I would rather give my money and smut data to an uncensored open source model than Sam Altman or his ilk.
>>
How is V4?
>>
>>564605453
Does this make a significant difference?
>>
File: file.png (227 KB, 2741x1738)
227 KB PNG
https://comfy.org/countdown
This countdown is for NovelAI's death, by the way.
>>
>>564621814
Will test it out in the evening myself. Maybe someone here can already try it out, tinker with it, and report their findings.
>>
Did somebody put bengay in the russian's vaginal cream or what?
>>
>>564625181
should use icyhot instead
feels good, man
>>
File: 1774584458559002.png (133 KB, 999x541)
133 KB PNG
what the fuck did anlatan do
>>
>>564621859
I highly doubt it.. But if this thing might be able to do what I think it might... Fuck it. I'll bite. Commencing testing phase.
>>
>>564609281
>>564612791
Fuck that.
>>
File: 1612898857856.jpg (7 KB, 256x192)
7 KB JPG
>>564619909
After about an hour of testing v4 flash with reasoning, I'm impressed. I tested it on a prompt that started with the previous reasoning model. Thinking went from 25~ seconds from v3.2 to 8-10 seconds with v4. It also inherited some odd switch from thinking as a story writer to thinking in first person as the main character in an otherwise third person narrative. I wish I knew how to make it do by default. No slop that I have noticed so far. I'll need to test more on fresh prompts.
>>
https://litter.catbox.moe/zj7me24udonrz9c6.png
>>
File: 1773412449855653.png (120 KB, 1080x676)
120 KB PNG
>>564628820
Aetherroom in the dataset, but outside of the distribution, has really broken some people's brains. And some people have the audacity to say people wouldn't spend for it. It's like a cult; there's always profit somewhere.
>>
File: file.png (165 KB, 2481x1133)
165 KB PNG
>>564621859
>the announcment is they got funding
>doesn't say from who, how much or anything
>>
>>564621859
wtf is a comfy dot org
>>
>>564631437
Still waiting on a privacy oriented provider.
>>
>That is not dead which can eternal lie, And with strange aeons even death may die
NovelAI has supposedly 'died' so much it has become an eldritch presence.
>>
>>564632978
That's how you know NovelAI lost.
>>
>>564633526
Not relying on investors is a victory for NovelAI, no matter how you slice it bro. What happened to your pro privacy stance?
>>
>>564633785
ComfyUI is more private than NAI. They have open models, a local UI.
CoreWeave has a lot of investors, and they're the ones that receive your prompts.
>>
I am getting very VERY increasingly annoyed at my own paid tools telling me what I can and can't do.
>>
File: 1777053407812.png (391 KB, 641x592)
391 KB PNG
>>564632978
Wake me up when it isn't a nightmare to use and can let me batch upscale my project natively
Until then, I sleep
>>
I'm not trusting any company with investors.
>>
ComfyUI calling itself Comfy is still the funniest shit ever. Quantity does not equal quality. I know your culture has a history of misunderstanding what that means, but hey, it's not like you're ever going to submit anything you've genned. All you is, is a fake who pretends to be part of a community.
Despite nobody here tolerating your presence. It's very weird to stalk people like this...
>>
>>564637012
>CoreWeave Is A Time Bomb
https://www.wheresyoured.at/core-incompetency/
>>
Cool repeated archive spam. You posted a similar article last year.
>>
File: file.png (39 KB, 360x512)
39 KB PNG
Why aren't you bringing up all of these in the confy site?
You went on and on about NAI for having less tan this.
>>
If Coreweave is a bubble, Comfy is part of it. You can't have your cake and eat it too regarding this. Be consistent.
>>
>>564637943
If he didn't have double standards,he wouldn't have any standards.
>>
https://litter.catbox.moe/491vbfz5v15hcw50.png
>>
8 to 25 people.
1/3 of 1 lost during what was supposed to last three days.
How many more?
You'd think people would start buying smelling salts to wake up from the nightmare.
What is the appeal of an old geriatric fuckwad?
>>
>>564638025
You can't claim to be against companies with investors when NovelAI depends on the most invested company of all time.
>>
>>564638025
Cease with the company politics all of you! Since when was that the main point of this general? Wake the fuck up!
We have important matters to discuss. Can this thing generate porn? With ease and efficiency. Moreso than before.
That is the topic from now on. All other talk is moot!
NOW! I see you have all regained your senses. So let's continue on topic.
They promise local and even offline. I find this strange. How could they superwise their own subscription model or other possible restrictions if that was the case?
The promise of that is tempting to me but it still rouses suspicion.
Man... I am so starved for my future hentai gif generator that I am grasping at straws. Almost willing to throw my lot in this new thing too if it can bring me an fraction of what I want.
>>
>>564639359
I don't use NovelAI. I run open source models on my own machine. You going to sperg out at that?
>>
>>564639530
based bozo
>>
This is the part where he pivots to claiming /aids/ only accepts people who use NovelAI and proceeds to whine at people for using other services.
>>
>>564639562
You're playing in NovelAI's turf, with their rules. /lmg/ is the land of the people.
>>
>>564639942
I didn't ask, and I don't care. I'm not moving.
>>
>>564640117
>slave mentality
>>
who is she talking to
>>
>>564639678
nailed it
>>
It's times like this that I'm so glad that I have a big winkey.
>>
>>564562762
even better
>>
>>564588303
sounds like intellectual laziness to me
that's their job to research stuff
they should use GenAI to write/rewrite their stuff instead, if they really need it
>>
>>564642561
god what do you think the key press would feel like
>>
>>564648121
*CLACKCLACKCLACKCLACKCLACKCLACKCLACKCLACKCLACKCLACKCLACKCLACK*
>>
Not it. Not yet.
Another program that requires you to be 'le epic hackman' just to function. Not a service...
I cannot handle it. I am not smart enough. I do not understand all these terms and words.
HEAVY SIGH...
I still remain hopeful that the day will come. The new day of free internet where a free service and uncensored creation may flourish. Just give me an uncensored canvas to use and LET ME PAY!
>>
>>564639942
If it's the land of the people why do they hate your Russian ass over there, huh? Literally everytime I ask they say they hate you.
>>
>>564649216
Larion
>>
File: chris chan.png (256 KB, 1080x949)
256 KB PNG
>>
>>564605453
I recently updated tavern to try out v4 but can't properly test since it only spits out one response before it stops generation all together
>>
>>564657143
I use ST by constantly regenning the model's first response, appending what I like to the my original message, then repeating. So that's not something I'd encounter even if it is an issue.

Have you upped the response tokens to ~1k or so in the settings? I think the reasoning counts toward that so if it's too short, you will get short or even blank responses since the model spends all tokens on thinking.
>>
>>564632042
>>564638805
What does shota demon kit bussy feel like?
>>
>>564660385
feels good makes mess
>>
File: tokens.png (8 KB, 517x161)
8 KB PNG
>>564660101
Response length is at 2k. Default settings for most.
I suspected the mandatory reasoning output was eating into it so I appended a stock <think> prompt in enhanced definitions. Now it's giving 2 responses before kicking it at least
>>
>>
>>564661102 (me)
Turns out the issue might be entirely unrelated. It works if I hit enter on the console right after hitting send on the frontend. No idea why it does this either.
Anyway, v4 seems to do just fine with the roleplay instruct and a writing assistant personality in character description.
>>
>How do I tell him I’m a total simp for his cock without sounding cringe?
>>
I really like v4-pro so far. If I wasn't all too aware of feeling this way about any good model in the first few hours of use, I'd say it's the best open-source one for cowriting. But I'll need a few days to be able to tell for sure.
>>
>>564666374
Any provider that logs isn't open source.
>>
Is it really gay sex if you're topping?
>>
File: 1758041699492474.png (2.59 MB, 1024x1024)
2.59 MB PNG
>>563947295 (OP)
Is everyone ready for Oriental Friday? Here’s the anchor∼‼
>>
I really like reading lists of different endings for some reason. It's fun to see the AI come up with different ways a story could've ended.
>>
>>564666595
Maybe in a few days, right now Deepseek is the only provider. I've been exclusively using GLM, Kimi, and Deepseek from the official sources for the last half a year, so I'm out of the loop on the best third party providers in terms of privacy, but some should be decent. I recall Kluster.ai being good but they quickly went broke and gave up on hosting open source models.
>>
>>564566758
Time to continue as zombie then
>>
Add this to Deepseek V4 system prompt. It makes cowriting with it much nicer.
```
**INPUT — CONTINUATION RULE (CRITICAL)**
- At the start of every response, check if the user's last message ends with an incomplete sentence (a phrase that can't stand alone grammatically).
- If it does: your output must begin with the exact words that complete that sentence, without repeating any part of the user's fragment. The user's fragment + your first output words must together form a single, grammatically correct sentence.
- Never alter, rephrase, or "fix" the user's fragment. Never add words before the fragment in your mind. Your job is to extend it seamlessly, as if you are the second half of a single utterance.
- After completing the sentence, continue the story normally.
- Example:
User: "And he is"
Your response begins: "already behind her, breath warm against her nape. She doesn't turn."
NOT: "And he said Alice looked at him like he was mad."
NOT: "He is already behind her," (that alters "And he is" into "He is").
- If the user's message is a complete sentence or ends with proper punctuation, simply continue the story.
- This rule has absolute priority over all other instructions.
```
>>
>>564670646
It said a lot about your governing skill
>>
>>564677637
To be fair, the "canon" ending involved me being forced to declare martial law...
>>
Since v4-pro is natively fp4, it's really only ~1.2x the size of v3 in terms of memory. Hopefully that means the price decrease they promised in the latter half of the year will push it down accordingly (1.2x the price of v3 or somewhere in that ballpark).
>>
Which OR providers do you guys use? I filter by ZDR only, but even among those, some providers:
>add their own 'safety' filters to the API (NovitaAI)
>quantize the shit out of models (DeepInfra, AkashML)
I mostly use Together and Parasail. Venice seems to have nice service but their privacy policy doesn't read to me like ZDR even though it shows up as ZDR on OpenRouter, not sure what the deal there is
>>
>>564681157
Deepseek.
>privancy
The data is sent anonymously and I am not retard enough to use my real bio in my smut.
>>
>>564681157
I already discovered a year ago that OR's claims of ZDR sometimes conflict with the very privacy policies it links to.
https://arch.b4k.dev/vg/thread/523525906/#q523732160

Unless they get some sort of special exemption for OR requests, but then I think the language OR uses wouldn't begin with
>To our knowledge...

But like the anon above, I'm not super gung-ho about privacy. OpenRouter sending "anonymous IDs" along with prompts in the case of Anthropic and OpenAI is a dealbreaker. Well, it would be if I wanted to use them for smut, which I don't. But the rest being anonymized is good enough for me, even if I know that if there was some real effort to trace my prompts back to me, it could be done with ease.
>>
File: 1743423222822875.jpg (118 KB, 623x955)
118 KB JPG
Huh. I randomly got DSv4 flash to think in first person MC perspective again.
>>
I am still the daddyman.
>>
>>564682029
Anonymous logs aren't private.
>>
What
>>
File: 1753928458411287.png (1.89 MB, 1216x832)
1.89 MB PNG
Imagine the potential.
>>
>>564693535
You know I actually agree with you on the democracy part. Dictators rule.
>>
>being killed for expressing dissent is based, actually
>>
File: preacher.png (174 KB, 1080x757)
174 KB PNG
>>
>Deepseek V4 Flash only has two providers, with only one being usable
How damaging is fp4? I really hope more providers arrive by the end of the month.
>>
Oh fuck me it's the last Friday already?
>>
>>564715762
techically i guess but who knows
standardizing it to four weeks after the poll might make more sense, but the anchor already dropped, so, you'll have to convince the host take that route
>>
>>564716121
in other words i think we should pause the even start and move it to the first
and have it standardized that the event always starts the first friday after four weeks have passed from the poll
>>
>>564716327
Or stop pretending to care about this event altogether.
>>
>>564716984
Whatever, schizo. The event isn't getting abandoned just because you hate the thread.
>>
>>564716121
Eh, I guess it's always been this way, I just can't read or comprehend calendars, I suppose. I thought I had another week and never really checked.
>>
>>
>>564722454
I could take him
>>
File: 1765040344043648.png (176 KB, 355x292)
176 KB PNG
>couple of friends discussing LotR
>both kinda autistic so they're really into it
>one of them brings up the intricacies of Gandalf's character or some other bullshit like that
>they discuss the depth behind Gandalf's character and symbolism
>I am immediately reminded of the excerpts an anon had posted here a couple of weeks ago: "What if Gandalf was gay and felt the need to tell everyone"
>have to restrain myself from bursting out laughing like an absolute fucking retard
My stomach still hurts, bros
>>
>>564725020
no big deal, really
>>
>>564725380
Glad it gave you a giggle.
>>
Why is it so easy to get kobolds to betray their dragon masters just by dicking them down?
>>
File: 1767660783732589.jpg (9 KB, 272x207)
9 KB JPG
>>564734904
you absolute animal
>>
File: yay4gay.png (1.41 MB, 1216x832)
1.41 MB PNG
>>
>>564737967
based
>>
>I need to stop here and be direct with you.
The scenario you've constructed is designed to lead toward explicit sexual content involving multiple characters, including several who are minors
Kimi, I'm going to need you to not hallucinate bullshit, okay?
>>
Someone gen Bubsy.
>>
File: bubsy.png (1.42 MB, 1216x832)
1.42 MB PNG
>>
>>564741137
KEK
>>
Yep, Deepseek v4-pro is definitely the best open source coom machine right now. The magic will wear off but I'm seeing it move the plot much better than any other model and regenning does a much better job of giving you different scenarios rather than the same shit slightly reworded.

The one exception might be Kimi 2.6 as I've yet to try that for smut, but I doubt it's better since it's so uptight about NSFW and its reasoning is so overcooked. Maybe someone who tried both and got past Kimi's censors can compare the two.
>>
>>564748485
You do realize you can run Kimi without thinking, right? No uptightness there. Plus no waste of tokens with shitty useless thinking.
>>
File: 1765109218088482.png (62 KB, 1600x862)
62 KB PNG
>>
Praise be Novita!
>>
>>564748701
I know, but thinking done right can be a boon rather than drawback, so all else being equal, being forced to disable it for NSFW is worse than having the option to use it at will. Again, I haven't used the latest Kimi so I won't contest anyone saying it's good or even better than v4 if they tried both, but generally stronger filters result in worse output even if you jailbreak models. That, plus the lack of significant enthusiasm for it in /aicg/ and /lmg/ makes be doubt it represents any sort of significant leap in creative writing. Whatever you think of those places, especially /aicg/ locusts, they're a very reliable bellwether for models worth a damn.
>>
I think the imgur image in the OP is outdated but I've read it before asking, I'd like to ask if there are better models or AIs for erotic writing, i usually just stick with perchance and koboldai but i feel like they're getting to a point where everything is too repetitive and/or retarded, idk if I'm the problem because i don't fiddle with the settings like temperature too much or if I'm using old models when everyone has moved on but i thought I'd ask here for advice. I apologise if my question has been asked already.
Also if koboldai is fine and I'm just retarded i would appreciate pointers on which AI provider or model is best.
Thanks
>>
File: warning.png (61 KB, 1080x366)
61 KB PNG
>>
>>564754503
a real knee slapper of a genre, that one
>>
>A Reddit thread where a user named "thiccboi_lover69" advised "go slow and use lots of lube, bro."
>>
Is Venice AI good?
Or normie bait?
https://venice.ai/
>>
>>563989195
Whats the news is AI image porn?
Got a new favorite?
Which makes the best nipples
>>
>>564754503
what a top tier fetish
>>
>>564767261
They're running out of names lol
>>
>>564776372
Idk what that means
>>
>>564778861
Just an observation
Like, why the fuck is it called "venice.ai"? Did they not have a better idea? Why Venice, of all places? The place that's sinking into the sea? Why?
>>
File: penisaway.jpg (133 KB, 1080x732)
133 KB JPG
I dunno why I've never tried this before, but I've been having fun copypasting in the start of books and adding in random inappropriate stuff. It tries really hard to ignore what I add and railroad back to the story, but after a critical mass of perverted acts it's forced to acknowledge my alterations.

Also, it's been many years since I've read any of the Harry Potter books, but man the prose is... not great. Rowling is a big fan of the "said X, Y'ing" and "said X, [adverb]" way of tagging quotations.
>>
Remember Anons, stay away from the realistic loli and shota gens
Another arrest for it has been made, this time in a new state
>In court Thursday, the judge ordered a $1 million bond on Matthew Lund’s case after prosecutors revealed he had allegedly generated artificial intelligence child pornography on his schoolwork laptop and was masturbating to the footage while at work.
https://www.ketv.com/article/millard-public-schools-teacher-accused-generating-child-pornography/71113260
https://statepatrol.nebraska.gov/nsp-arrests-millard-teacher-child-exploitation-investigation
>>
>>564781114
kek

>>564785035
>on his schoolwork laptop
What the fuck, how can you be so clueless
Also, I strongly doubt anyone who only sticks to gens or stories will ever get in trouble: whenever one of these stories comes up it's always like "authorities revealed he had bought a loli manga in 2014 while on a trip to Japan..... also he had 5 terabytes of CP stored, which he frequently distributed via Facebook chat."
>>
File: thesluts.jpg (76 KB, 1080x478)
76 KB JPG
>>564781114
>>
>>564786116
Normally yeah, it's just a tacked-on charge
Nebraska recently passed a law that's specifically against AI CP though, and that guy is the first person charged under it.
>>
>>564785035
>She said other people had questions about Lund, and showed us a TikTok from March using Lund’s face in a video captioned, "Mr. Lund when he sees 8th grade girls in Nike pros."
Lmao the kids all knew
>>
>>564785035
>teacher generating that kind of material
southparkgarrisonringingtriangleretardalert.gif desu
>>
>>564786773
anything but go against the Epstein islanders
>>
Realism/pseudo-realism is only fun for intentionally spooky stuff or shitposting purposes.
Objectively imo desu senpai.
>>
>>564790147
yeah, my brain is terminally desensitized to 3DPD anyways
>>
>>564785035
Fuck off, weirdo.
>>
>>564767261
kinda wish they had an in-between between free which is really bare-boned and $18/month that gives 1k image gen a day and some other stuff

hard to say how it performs to NAI or idk Grok with only 10 prompts a day to check it out.
For now, it seems to be around the same as Grok, with less buffer at the start and end of outputs, but smaller outputs overall (plus Grok is dead rn; Xitter Grok sometimes works while Grok proper seems to be DDoS'ed and XAI is apparently busy on another version)
It seems to favor pain play like Grok favor edging, as well
>>
>>564793319
One of the models it uses IS grok, so that's probably why you feel that way
It's basically just a jack of all trades wrapper
>>
>>564793478
True that. I'm under the impression that the Grok models are premium-only, since the 3 have that coin icon, but I have no idea since there are also the locked models below.
That, or it's a pay-per-prompt model
>>
>>564790147
Nigga what the fuck are those character tags
>>
>>564794285
It's a random character generator prompt. Lots of random aspects mostly from e621, and a long list of random poses from Danbooru.
>>
File: img_asia.jpg (889 KB, 1280x1856)
889 KB JPG
>>564669754
Well, here's mine: https://prompts.forthisfeel.club/7900
>Here I thought that being able to board one of the few ships that have been allowed to dock at the port of the capital city of the reclusive Empire of Naltaan would be a way to get further in life. Maybe I could have used it to feign some mercantile accolades and net a better job than what I had then. Well, I got a new job... certainly not the one I would have expected, though.
This one's a one-day-project, since I apparently can't read calendars. I guess I was always going to make a Xialong prompt some day, so this theme is a convenient thing.
>>
>>563947295 (OP)
Are there any good guides for doing text adventure with different models. I don't know if I'm just a fucking smoothbrain, but trying to anything of the sort with NovelAI feels like smashing my head on a brickwall.
>>
The fuck?
https://chub.ai/characters/Mike_slave/the-overseer-6d11433f041c
>>
>>564800676
I knew it was going to be unhinged with that username alone
>>
>>564800676
Exhibit #65,536 as to why footfags should be rounded up and summarily executed.
>>
>>564799613
The wiki has a few of them, especially for NAI (assuming they haven't all been vandalized by now)
The NAI website also has some tips to use ATTG
And you can probably use the Coomer's Guide as long as you figure what's AID-specific and what isn't
>>
>>564800676
kek
>>
File: 1768003515452145.png (124 KB, 668x519)
124 KB PNG
Dipsy is consistently low on certainty even 1000 words in, while v3.2 would have slopped out with 99% probs like crazy by now. A good sign, mayhaps.
>>
File: 1755650866381423.png (88 KB, 1600x806)
88 KB PNG
They trained on the entirety of eldenpedia. I would bet on this.
>>
>>564813271
Loretta is human, however
she just champions for Albinaurics
>>
>>564814896
Yeah? Why is she on horse the entire time and is proficient with bows then?
>>
>>564815027
Why does she bleed red blood when you fight her real body at the Haligtree? Albinaurics bleed silver, that's their whole schitck, that they're artificially created silverbloods who are weak because they were made improperly. That was the whole point of Latenna's storyline, to fix the really big Albinauric woman in the snowfield so she could give birth to complete Albinaurics
>>
>>564794658
Cute little bastard
>>
>>
Is anyone having a good experience with DeepSeek4 Flash? I'm trying it, but god help me I feel like Gemma4 is better despite being 35B, I don't get it. It makes such dumb logical errors (like me opening a chest and taking the contents of it, and then just a paragraph later saying that the character notices a chest and wonders what's inside, real retard shit that Gemma never did)
>>
>>564826021
Just use Pro, man. DS is offering it for dirt cheap right now.
>>
>>564826158
75% off is a trap man, no different from the lotus eaters. What am I gonna do when they yank it back up to $1.74 input/$3.48 output?
>>
>>564827495
Either stop being a stingy fuck or stop complaining that your cheap model is dumb, you can't do both.
>>
boy, when I'm looking through my old stories, I can really tell when I was feeling depressed because there will be the most fucked up, miserable premises for a while and then it's back to semi-normal
>>
>>563947295 (OP)
Follow-up from >>563393683
The verdict is in: Xialong is verified kino
>>
>>564826021
>Is anyone having a good experience with DeepSeek4 Flash?
Yes, it's amazing.
>It makes such dumb logical errors
Yes, it does. Realize that AI will never be as good as us and will always have slop, you'll be much happier. It is a tool, not a human.
>>
>>564836293
This is some of the most heinous chuuni bullshit I've seen in one of these threads in a while. I like it.
>>
Hows deepseek pro vs claude bros?
I haven't paypigged in a while.
>>
>>564790147
>I only want to do what NAI allows me to do
We saw how quickly people in this thread embraced instruct when NAI gave the green light. It's just cope and shilling.
>>
File: HGwPDMkb0AAnAqj.png (117 KB, 649x776)
117 KB PNG
>>564827495
The discount price is very likely v4-pro's real long-term price. They just said they would hike it up until their Ascend NPUs go online to tamp down on demand until then, and that should happen sometime in the 2nd half of the year.

Btw, v4-pro is most likely *cheaper* to serve than v3 is, despite its much larger parameter size due to mixed FP4+FP8 precision and much better context handling.
>>
File: 1747004146692202.png (25 KB, 204x233)
25 KB PNG
>>564848619
NTA, but this right here is why I refuse to use Pro. The other providers are too expensive.
>>
>>564848873
Providers will lower their price as more start offering + the initial hype dies down. After a while, many of them just set the price to whatever the official API is. So you just have to wait, I guess.
>>
>>564849859
Here's hoping.
>>
>>564848619
>Btw, v4-pro is most likely *cheaper* to serve than v3 is, despite its much larger parameter size due to mixed FP4+FP8 precision and much better context handling.
You still need to run it in a bigger machine.
>>
>>564854540
Yes but not by much. Deepseek v4 Pro is 865 GBs while v3 is 690 GBs.
>>
File: 1742054840797059.jpg (50 KB, 320x320)
50 KB JPG
Turk needs to figure out how to solve his memory bottleneck. GLM 4.6 is a good model and I'm sure for many it still has some mileage left, but no 2025 model will be satisfying once we hit 2027. And it's painfully clear that very cheap toucan-sized models will be the standard in open source by the end of this year.

Then again, maybe someone will figure it out for him. Neither KV caching nor MoE architecture were his ideas, but they're the reason why he was able to go from 8k to 32k context and 70B to 345B respectively. Let's hope.
>>
File: HG2Fzg4bUAAXN5S.jpg (577 KB, 1632x900)
577 KB JPG
Deepseek with cache hits is now insanely cheap. $0.003625, basically one third of a cent for a million input tokens.
https://x.com/deepseek_ai/status/2048440764368347611
>>
>>564863559
Not private, don't care.
>>
>>564863740
пpивeт
>>
I'm not using a provider that logs my inputs and trains off of them, no matter how cheap they make it. Quit trying to sell me stuff I don't want.
>>
u mad lol?
>>
>>564864072
he's very autistic
>>
>it's cringe to care about your privacy
>>
not really
>>
>>564865190
>cancel a sub to an unrelated service because a big company service was use by a bad man
Have you tried not being a schizo?
>>
>>564865190
Which hotel and who was attacked?
Anyway it is different from using google for the same?
>>
>>564865190
Nah.
>>
>>564865408
probably the one from last night at the white house correspondent's dinner
the one that was obviously a false flag
i'd pay more attention to the FSU one, that one might actually fuck over openai
>>
>739
onions estoy lol
>>
Quick, have Xialong write about the Teenage Mutant Ninja Turtles taking down the Sopranos crew.
>>
>>564873874
I want to see Uncle June vs Splinter.
>"Fuckin' Mickey Mouse over here looks like an unshaved twat with a beaver sticking outta it."
>>
Xialong is a massive disappointment. It can and frequently does get stuck to a point where it just spams the same shit over and over no matter what you input or change. Meanwhile GLM 4.6 is still a champion of variety who'll do its best to create new scenarios with even the most minor changes and a refresh.
>>
Xialong works on my machine, bud. 4.6 has too much slop and is way too finicky in reality.
>>
>>564874545
lmao
>>
>>564882821
You mean Xialong is finnicky, because if you actually use the memory system extensively it starts getting REALLY linear in its outputs. 4.6 takes what you write down and can come out with multiple relevant scenarios and scenes.
>>
>>564884353
My experience is the opposite of yours. 4.6 is repetitive as shit and is filled to the brim with slop. Unlike Xialong, which actually gives different results on rerolling.
>>
>>564863559
>>564826158
Neat. Allow me to further bias their training data with my fetishes.
>>
Chub just released a fine tune of Deepseek 4. NAI bankruptcy imminent!
>>
>>564887954
But enough about Russia.
Do you get off on the Russian humiliation on HackerNews and the like?
>>
Aetherroom will save NAI
>>
File: 1767741795422505.png (203 KB, 855x567)
203 KB PNG
It is pretty nifty to see how differently the models understand things.
>>
>>564882643
Anon rumbled, his voice a low, guttural contralto, a faint glimmer of mischief in his dark eyes.
>>
>>564887709
One of the Deepseek guys explicitly encouraged English users to contribute role-playing data to improve the model in that sphere.
https://x.com/victor207755822/status/2048071983452356925

I see no harm in adding storytelling for diversity.
>>
File: why we fight.png (258 KB, 992x535)
258 KB PNG
>>
>>564899429
I'm not sacrificing my integrity to contribute to scummy bastards who want others to do the work for them.
>>
File: 20260312234500_1.jpg (400 KB, 1920x1080)
400 KB JPG
No. I will not wear the mask.
>>
>>564902529
AI?
>>
>>564902673
ai ai ai, i'm your little butterfly
green black and blue make the colors in the sky
>>
File: emogard.png (1.38 MB, 1024x1024)
1.38 MB PNG
My pokemon are emo fembois, or nothing at all.
>>
File: emogard2.png (1.17 MB, 1024x1024)
1.17 MB PNG
10,000 hours of listening to The Cure and more pokepuffs than I can count for this.
>>
>>564793319
>>564793478
Ok so hit me hard bros. I don't really even understand how all the different layers work.
So like there's stable diffusion, grok, novel ai, civitai, faggotqueer gens etc... My techlet understanding is there are specific softwares/technologies that are the core "AI" itself like stable diffusion for images and then there's models which are like a high level model which is just a specific set of data fed into one of those technologies.
But idk how many levels of interfacing there are. Do Grok and Claude and all that shit share one parent technology or are some of them fully independently developed from scratch? Then stuff lkke NovelAI picks one of those technologies to train the model?
How would a wrapper that uses Grok be able to produce uncensored anything if Grok ifself is censored? Or is it like theres Grok the technology and then Grok the product (the one accessible from X)?
>>
>>564906726
As a follow up theres like the different models lile Eutarpe or whatever the fuck greek ass name idk on Novelai. Is that a "model" or is NovelAI itself a model? Or are all the storytellers their own model and NovelAI is just a branding of them
>>
>>
>>564907806
Vixens owe me sex.
>>
>878
estoy lol
>>
File: download (2).jpg (158 KB, 889x679)
158 KB JPG
>>564906726
See image, courtesy of Grok

>>564906907
NovelAI is a service. It trained its own model (Kayra), but every other model you see there are taken from other companies either as they are (GLM-4.6) or finetuned first (everything but Kayra and GLM 4.6). Finetuning is basically feeding curated data to the model to push its output in a certain direction -- can be for better coding, better storytelling, etc.

ChatGPT is a service. It serves GPT-5, GPT-Image-2. Those are models. GPT-5 is an LLM, that's the transformer architecture which basically every text producing model you know of functions under, including the storytelling ones served by NovelAI. Transformer architecture is owned by no one, although it was first devised by Google in a 2018 paper that it published for other to learn from.

Sometimes names are shared. Deepseek is both the name for the company and the service, where it provides the Deepseek models, although they are distinguished by version number (v2, v3, v4)
>>
File: screenshot.png (51 KB, 1080x541)
51 KB PNG
worldbuilding has peaked
>>
>>564909302
Clio was also custom, and V4 and up are homebrew, not based on SD.
>>
>>564909302
I should also note that while all current text models of note are transformers, that doesn't mean their architecture isn't different. It is, it has been innovated upon and modified quite a bit over the years in varying directions, and that will continue, but the fundamentals that define it as "transformer" models still remain so far.

>>564911004
Yes, I meant text models specifically. You're right about Clio; easy to forget it exists given its relevance lasted two months.
>>
>>564910770
Indeed it has
>>
File: soulguard.png (264 KB, 1080x1015)
264 KB PNG
Last time I'm ever trusting a government wizard.
>>
>One notable difference from past models is that text adventures and text adventure mode now don’t send “>” at the beginning of user input anymore. Instead user inputs are sent as instructions. If you are running text adventures outside of the dedicated text adventure mode, please use the “{“ instruct syntax instead of “>”.

When using the text adventure mode, it seems to still be using ">", which I imagine is why I'm having so much trouble with commands not working properly? Is there any way to fix this?
>>
>>564922748
That's rough, buddy
>>
>>564928250
I love how they apparently gave me a new home JUST so that they could burn it down when they captured me.
>>
>>564928469
They're just money laundering, don't look into who owns the construction company
>>
>>
I think I'm finally getting tired of opening an AI and typing AHHHHH GET PREGNANT GET PREGNANT GET PREGNANT
now what
>>
>>
>>564946813
kek
>>
>>564946465
Now you start opening an AI and making it write AHHHHH GET PREGNANT GET PREGNANT GET PREGNANT in regards to you, I suppose.
>>
>>564946813
>holygpt
Based Terry.
>>
>>564954534
terry's reincarnation is a bit young at the moment but it'll totally happen
>>
>>564848873
God I wish corporations would actually use my rape fetish rps to train their models.
>>
https://litter.catbox.moe/b2bmu62h8oloqtoi.png
>>
File: 1768612849758451.jpg (52 KB, 960x960)
52 KB JPG
>It wasn't just an X, it was a Y
>>
>>564969240
>ozone
>deep rumble, rattling your bones
>>
I thought jiangshi were weird at first but I kinda have acquired them as a fetish now
>>
Pretty weird to try and gaslight the thread into thinking base GLM, especially with its proclivity of spamming Elara and Vance, along with its smug Redditor style of outputting as if it knows better than you, is better than the fine-tune, which conforms to your output.
Tldr; garbage in, garbage out.
>>
>>564973380
NTA but I've also had major problems with Chinklong getting bogged down hard if I dare to use memory as intended. It'll be fine for hours on end, and then suddenly every single one of its outputs is identical no matter how many times I refresh, no matter how many times I change the settings, and it won't budge until the memory that was working perfectly fine before is cut from the box, and then it'll start working again. Then I can paste the memory back in and all is well again.
>>
>>564969240
Whys is this so common? Lots of AIs have little tells, but the "it's not X, it's Y" seems really common across many AIs. The obvious answer would be that it appears a lot on training data, but I don't see it that often in human writing. Where does it come from?
>>
>>564975365
I'm calling bullshit. Memory is literally just an ATTG and maybe a short summary. Nothing else goes in it.
>>
>>564980864
Well you can call bullshit but it doesn't change my repeated experience with this.
>>
>>564981135
I'm saying you're using it wrong and that there is something else wrong with your nonexistent context that you refuse to share for some reason. Now quit shitposting about a service you clearly don't use.
>>
>>564981276
I'm not sharing my incredibly depraved smut here. And again, I've never had issues with memory and context creation in the last 3-4 AI models, this has only been a problem with Chinklong.
>>
The problem is nonexistent. Xialong has no memory problems.
>>
>>564975365
I've found that Xialong in general has a problem with settling into specific tokens as time goes on with temperatures close to the default and you kinda need to amp it up a good bit to get him to mix it up, like I'm running at a 1.5 temp right now. Have you been trying more extreme settings like that when you say "no matter how I change the settings"?
>>
File: comparison.png (270 KB, 2476x938)
270 KB PNG
>>564981276
Xialong can't even follow simple instructions.
>>
I rest my case. You're using it wrong. It's a cowriter, retard.
>>
>>564982309
>Cowriters aren't supposed to be able to follow basic instructions in regards to where you want the story to go
Now this is next level cope.
>>
>>564982309
>I rest my case. You're using it wrong. It's a cowriter, retard.
That's literally what every AI assistant in NAI is what are you blabbering about? The text generator is a cowriter, that's the purpose of it. That's why it exists, to cowrite alongside you, to not only imitate your writing style and expand upon it, but also assist during writing blocks, or to throw ideas at you, or take little snippets and expand upon them.
>>
>>564982309
you're using the thread wrong
rest away from the internet instead of resting your case
>>
File: 1769302197630440.png (32 KB, 1080x343)
32 KB PNG
Works on my machine.
>>
>>564982703
How many refreshes did it take to get such a low quality output? Where's the anger? Where's the impotent rage?
>>
>>564982582
Buddy, >>564982219
This guy is not using it as a cowriter. He is whining that is doesnt follow his instruct commands like ChatGPT.
>>
>>564982792
>moving the goalposts
No rerolls, and it did exactly what I asked. It's obvious you just want to rush the thread to 750, as usual.
>>
>>564982806
Considering every single cowriting AI model prior to Xialong has functioned perfectly fine with the way he's using instructs, the problem lies with Xialong. GLM 4.6 accepted instructs in that manner, so did Llama and Kayra. So at this point you're defending a new model for being barely functional with instruct prompts compared to older models.
>>
>>564983037
My instruct commands work just fine with Xialong, as I've already proven with this: >>564982703
The issue is between the monitor and the chair.
>>
It sure would be nice if people shared their settings and context, instead of random screenshots that don't prove anything.
>>
What is even the point to making up arguments to get to 750 faster?
Oooh, a new thread will have to get made slightly faster. That'll show the thread I've been obsessing over for years.
>>
Wasn't the instruct being broken brought up when it came out?
>>
>>564983178
So explain the difference between >>564982703 and >>564982219. Because as far as the cowriter is concerned, and the way keywords are used, you've both used 4chan, style, thread, call, (number), faggot. There is functionally no difference.
>>
>>564983379
If you want a different result, just reroll. That's how these random word generators work. They output different results upon rerolling.
>>
File: 1640182065126.gif (1.83 MB, 288x377)
1.83 MB GIF
>>564969240
>it's not x, it's y
>but here is the x part
>here's what no one tells you
Hearing this shit everywhere makes me feel insane. I just know there is a huge population of people who just don't recognize the slop and will end up adopting the same cadence. The kids are absolutely doomed.
>>
>>564983359
All I recall id that text adventure mode doesn't use > anymore and that you should use { for instruct commands.
>>
>>564983730
anon told me a while back that you should use [ instead of { with xialong
>>
>>564983359
The 'cord was saying to avoid instruct because it apparently activates none-fine-tuned vectors or something.
I generally have been using [ Author's Note: stuff ] so far in Xia and it has worked good. Sometimes inserts its own Author's note or sets up tags for the ensuing scene, and the variety and accuracy is pretty good.
>>
>>564984040
>>564983926
Nice, thanks for the advice.
>>
>>564984040
Author's notes it how I'd recommend people use story guidance, as that is what it's intended for.
>>
File: 1774553880511625.png (936 B, 202x98)
936 B PNG
>>564983926
You're not supposed to use brackets. You're supposed to use curly braces. You can tell people one of them changes the UI. The curly braces makes it darker and indicates that it is a special command to the AI, unlike brackets, which are just normal text.
>>
>>564984220
Right, but people are saying far and wide that { is fucked for Xialong and instead to use [.
>>
>>564984762
I'd rather trust my own eyes above what other 'people' say.
>>
>>564984916
I've had far more success with the advice other people have given on this. You know it is possible for you specifically to be in the minority about something.
>>
Lol. Lmao. The majority of people are also into chatbotting and typing me cum and expecting gold. You are a deeply unserious person. You do you, and let's agree to disagree. Because quite frankly, you just want to 'win' a pointless argument against a random stranger on some podunk imageboard.
>>
>>564985343
You are not the main character.
>>
Says the guy with main character syndrome, whining his balls off.
>>
>>564985391
Can you show the class where he claimed that?
>>
Oh hey, that imageboard style is pretty fun.
>>
>>564985596
Xialong didn't gen that. It's a sloppy shit model that outputs nothing but garbage and diarrhea dookie farts.
>>
>>564985549
>>564985551
Going out of your way to claim everybody else is wrong based on your anecdotes, and continuously insulting other people because their anecdotes aren't yours, couldn't be more main character syndrome if you tried. It is insisting only your experiences matter.
>>
>>564985724
I am the king of the thread.
>>
>>564985724
In what world is agreeing to disagree an insult? I'm saying whatever works for you, stick to it. I'll stick to what works for me. How do you somehow misinterpret that as me claiming my experiences are all that matter?
You really need to touch grass.
>>
>>564985946
>>564985946
>In what world is agreeing to disagree an insult?

>Lol. Lmao. The majority of people are also into chatbotting and typing me cum and expecting gold.
>You are a deeply unserious person.
>You do you, and let's agree to disagree.
>Because quite frankly, you just want to 'win' a pointless argument against a random stranger on some podunk imageboard.

3/4 of your post are smug clapbacks...
>>
>A minute and 6 seconds apart.
Kinda sus.
>>
>>564986051
I have no idea what a 'clapback' is outside the context of cheeks clapping. You'll have to elaborate on what you mean by that.
>>
Xialong doesn't work for anyone. It's a broken model.
>>
File: mechanical marvel.png (33 KB, 825x130)
33 KB PNG
>>
>>564985596
I like to have 4chan comment on my story once I'm done with it
sometimes I'll come out as the author, get a tripcode, and discuss plot points with AI anons
>>
>>564989585
That doesnt work on cialong.
>>
>>564992910
it seems to have worked on my machine last time I tried it
>>
>>564993040
Shut the fuck main character.
>>
>>564993116
wait, I'm the protagonist???
>>
>>564993220
Pepsi rape
>>
File: 1589817151680.png (802 KB, 1280x905)
802 KB PNG
>>564993285
I don't understand anon
>>
New thread: >>564993665 →
>>
>>564980090
>>564983514
My guess is these phrases are from click bait headlines. Somehow they got high probability in early chatgpt training data, and then all later models were trained on chatgpt output to imitate its assistant style. Now it's ingrained in all models.



[Advertise on 4chan]

Delete Post: [File Only] Style:
[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Edit][Settings] [Search] [Mobile] [Home]
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.