[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] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/vg/ - Video Game Generals

Name
Spoiler?[]
Options
Comment
Verification
4chan Pass users can bypass this verification. [Learn More] [Login]
File[]
  • Please read the Rules and FAQ before posting.

08/21/20New boards added: /vrpg/, /vmg/, /vst/ and /vm/
05/04/17New trial board added: /bant/ - International/Random
10/04/16New board for 4chan Pass users: /vip/ - Very Important Posts
[Hide] [Show All]


[Advertise on 4chan]


File: img.png (863 KB, 832x1216)
863 KB PNG
Looming And Sinister Presence Edition
Last Thread: >>563947295

>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!
>April's event is over, enjoy the oriental prompts! >>564669754

▶Model News
>(04/24) DeepSeek releases V4: https://api-docs.deepseek.com/news/news260424
>(03/31) NAI releases Xialong: https://blog.novelai.net/welcome-novelais-newest-writing-model-xialong-ecde7d21d111
>(02/23) NAI updated its Canvas: https://blog.novelai.net/image-generation-canvas-update-new-tools-better-performance-and-3d-support-cbbc54c1d89d
>(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

▶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
>>
infinite zork bros...
>>
Thanks Anon
>>
This might be a bit of a low point but I'm not sure I care.
https://litter.catbox.moe/3z1o92cvytbrgdp2.png
https://litter.catbox.moe/vi1555zti8fr2lil.png
https://litter.catbox.moe/eeliy1f02bc0vhqu.png
https://litter.catbox.moe/wgdvagg0u4l065jv.png
https://litter.catbox.moe/a9j9rogpc3wgrttl.png
>>
File: mechanical marvel.png (33 KB, 825x130)
33 KB PNG
>>
>>564993787
Has anyone even tried to make AI Zork?
>>
>>564997051
Cute!
>>
Damn, thread was dead already

>>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.
>>
>>
>>565002186
Max Loh is trying, but his game has too many bells and whistles
>>
Xialong keeps flirting with me...
>>
>>565005970
Is he trying to top or bottom?
>>
Is this thread more or less gay than the other AI threads? Just curious.
>>
>>565007123
No idea about what the other threads are actually like, but my money's on more.
>>
>>565002420
Sad.
>>
>>565007123
here at /aids/ we appreciate both the masculine and feminine forms :^)
https://litter.catbox.moe/le5t7avcama2emiu.png
https://litter.catbox.moe/136e70l6qgrfq2ma.png
https://litter.catbox.moe/dcs6fniv4uc4ebv1.png
>>
>>565006436
Power bottom, again
>>
>>565007123
It's more cheesy
>>
File: motley crew.png (240 KB, 1047x624)
240 KB PNG
>>
What's new in the hood my niggas?
>>
Do you think Splinter would be good in bed despite being an old rat man?
>>
File: she was the wolf.png (342 KB, 1031x818)
342 KB PNG
>>
>>565020503
Depends on if he has a humanoid dick or not I think
>>
File: chat1.png (95 KB, 1040x772)
95 KB PNG
>>
>>564993665
Thanks to everyone who participated in Theme Friday! This time, we received one (1) prompt! I’ll be taking suggestions for Mays’s theme till Friday evening, whereupon I shall post the poll, which shall last till Sunday evening. Looking at the calendar, the event itself shall kick off on the 29th and last till the 31st!
>In The Dragon Emperor's Court: https://prompts.forthisfeel.club/7900
>>
>>565023753
Gaslighting.
No, seriously. You've all wanted this to be a theme. So don't complain when I suggest it. Why would there be a problem with me suggesting it if we all wanted it?
>>
>>565023753
Nigga why do you keep doing this?
>>
>>565025191
Why not?
>>
>>565025346
It's depressing
>>
>>565025191
He needs to be put in a camp.
>>
>>565025191
He’s a really fascinating specimen.
>>
>>565025810
with the mice
>>
>>564798021
>Something New For The Hoard
A serial comma should go in 'merchants and'.
That's quite a surprise to have the leader of the country be able to speak your language, especially since it's been secluded for so long.
In your Memory, in 'writing, painting, and the like', I'd probably go with 'writing or painting or the like'.
It might do to make his speech sound a bit older in comparison to what you'd expect from the main character and the merchants and translators and such. If he learned it based on what was known a millennium ago, he's going to sound a bit out of date in comparison.
>>
>>564993665
Oh boy time to take Xialong for a spin!
>Boot up dark fantasy adventure prompt
>Get raped by swamp tentacle monsters
>Boot up isekai powerfantasy
>Get drugged and raped by loli catgirl
>Boot up death battle gladiator prompt
>Other guy takes off his armor and is actually a brunette bombshell
>Immediately swap to post apocalypse space bunker prompt
>Somehow end up fucking the toaster
One of those models huh...
>>
File: friendly fire.png (293 KB, 1047x845)
293 KB PNG
>>
>>565029154
your fetish is women with guns, isn't it
>>
>>565029743
https://youtu.be/d4uA3t9AOUw
>>
File: 20251218152254_1.jpg (477 KB, 1920x1080)
477 KB JPG
>>565029743
Yep.
>>
>>565027315
You think that's bad? I used Xialong to play a text adventure where I was a 4chan user trolling the U.S. government and I got a visit from the FBI that turned into a gay rape scene.
>>
>>565029880
Ah, a fellow man of culture. That band is kino.
>>
>>565030472
This is what will happen if hmoma becomes president.
>>
>>565031217
Jr gets my vote.
>>
>>565031217
haha but seriously we can't let this guy have the nuclear codes.
>>
>>565031624
Somebody still lives with parents
>>
It's fun to have in-universe imageboard stuff. This one's for my in-progress story of my state as a Pokémon region.
>>
>>565034589
You the bloke from the last thread who used author's notes in brackets to get the AI to talk about shit like this?
>>
>>565034961
Nah, but the talk about it did inspire me to play around with the idea again. Makes the world feel a bit wider and alive.
>>
>>565035380
I see, very neat.
>>
File: 1770054325683563.png (155 KB, 1600x974)
155 KB PNG
I need to go try another prompt. Last one from this.
>>
>>565035741
That's crazy, that can't be deepseek.
>calling loretta by both pronouns
>"scrap against" and not "scrape against"
I thought it would be better than THAT, at least
>>
>>565036202
Yeah, this model frequently makes basic mistakes like that. Temp 1, no sampler. But I can’t deny that it’s really knowledgeable and creative.
>>
>>565036716
Wild, no wonder people seem let down by it
>>
>>565035741
I've gone back to Gemma4 for my frontend adventures. I'm not really sure what to say, DeepSeek dropped the ball hard this time. I was convinced it was a 'me' problem and tried so many different temperatures/settings, but it just kept being dumb (getting pronouns wrong like there, and also just logical consistency problems I wouldn't expect out of a big model)
>>
>>565040262
It’s likely undertrained. When the Ascend 950s go online they should have some more compute to properly train the next iteration.
>>
File: 1752305655198152.png (34 KB, 934x107)
34 KB PNG
>>565036202
>>565040262
Here are the probs for “his”.
>>
Yeah, keep telling yourself that.
>>
>>565042874
how the fuck
>>
>>565042874
Min-P is really critical, I find. Not even much, but good lord, sometimes the low low end of probs can just contain absolute chaos tokens. Have to make sure they're gone.
>>
>>565043628
You could easily prove that if you actually used AI, but you gave up on it.
>>
File: sky master.png (188 KB, 1080x703)
188 KB PNG
>>
Suddenly, my self-insert became a schizo.
>“I talk to myself. Not like, a little. Full conversations. Arguments, sometimes. I pretend I’m debating someone from history. I lost an argument to Cicero once. I still think about it.”
>>
>>565026218
Thanks.
>That's quite a surprise to have the leader of the country be able to speak your language, especially since it's been secluded for so long.
Yeah, I really just didn't want to keep a third character involved as an interpreter and such, so I just waved my hand and figured "fuck it, dragon knows it because dragon".
>It might do to make his speech sound a bit older in comparison to what you'd expect from the main character and the merchants and translators and such. If he learned it based on what was known a millennium ago, he's going to sound a bit out of date in comparison.
Fair, I think I focused a bit too much on getting the hyperactive twink part in rather than making sense in the story. I rejiggered his dialogue a bit.
>>
That feeling when you go browsing through your library of old stories for nostalgia's sake and you realize that the vast majority of them are no more than a few paragraphs long or just low-effort shitposts.
>>
>>565020503
Ridiculous techniques can be fun. Remember to shout the moves when used
>>
>>565020364
Someone tried deepseek4, dunno how it went
>>
>>565034589
Pokemon vietnamese crystal and outlaw created the best asurbdist humour
>>
>>
>>565059772
I don't really have that problem. Actually I can't say that, if I go all the way back to my pre 2021 ai dungeon stories they're pretty short, but that's to be a bit expected for the time
>>
File: bullying.png (72 KB, 1080x354)
72 KB PNG
my waifu is getting cyberbullied
>>
File: link test.png (242 KB, 1080x999)
242 KB PNG
Dunno if anyone else will find this useful but here's a weird tip: Xialong's really good at parsing URLs that are formatted in a specific way. If you type [ Link: URL ] he'll guess what it's a link to and try to write something based off of it.
>>
>>565074673
Have him link to this thread
>>
File: minecraft.png (472 KB, 1080x2445)
472 KB PNG
>>565080190
Apparently we're a minecraft thread.
>>
>>565058596
You're welcome.
Very understandable, plus, what interpreter is going to keep up with him? I think there's something to be talked about it, though, perhaps to be brought up later.
Yeah, that works. He's still exuberant, just archaic about it. Allows for another layer of juxtaposition between the young current appearance and the older style of speech.
>>
File: 1770173341654.png (192 KB, 498x464)
192 KB PNG
>>565082318
>still no anons to be cubs with
I'm sorry but fucking what? I'm assuming that's fur loli/shota speak unless there is some crazy autism beyond normal autism lurking within that general.
>>
File: back to prison.png (233 KB, 1042x766)
233 KB PNG
>>
One annoying Deepseek v4-pro tendency that I've never encountered with another model before: inability to consistently use a single type of quotation mark. It either uses “ ” or " ". Even if your context is entirely using one or the other, it won't stop it from randomly using the opposite one.

Yes, yes, I know biasing against the tokens should solve the issue with ease, but it's still an odd quirk.
>>
File: special treatment.png (179 KB, 1031x501)
179 KB PNG
>>
https://xcancel.com/arb8020/status/2048958391637401718

So that's why I couldn't get GPT to talk about smedrins
>>
File: file.png (38 KB, 842x214)
38 KB PNG
>create a conceptually super fun AI model trained exclusively on pre 1931 text, basically a conversational time machine to let you talk to a person who only has knowledge from the distant past
>censor its outputs so it isn't allowed to talk like a person from the past
WHATS THE FUCKING POINT!?
>>
>>565005501
How IS AI Roguelite doing?
I haven't played since early December, on the Sapphire sub. My prompt mods are probably gonna have compatibility problems by now.
>>
>>565095780
A lot of people are emotionally invested in imagining the past to be a lot more flattering to their own, very modern ideological preconceptions. They don't like being reminded of it as it creates dissonance.

Not even a left/right issue as both sides are guilty of it in different ways, although this particular example is definitely left-wing retardation.
>>
File: not4kids.png (1.81 MB, 1024x1024)
1.81 MB PNG
>>
File: 1761776193991042.png (312 KB, 735x411)
312 KB PNG
https://www.youtube.com/watch?v=h4e40N3We-U


DO YOU REMEMBER CHASING DRAGOOONS
WE FOLLOWED THEM TOO FAAAAAR
COME FIND ME WHEN IT'S OVEEEEER

AND NOW WE'RE COVERING OUR SMUT
WE'VE REARRANGED OUR BEAKS
I'LL FIND YOU WHEN IT'S OVER

DEAD SMEDRIN
DEAD SMEDRIN
DROWN MY
MORMON
FIND A WAY RIGHT BACK TO YOU

WITH DISGUST IN MY EYES
I'LL KEEP DRAGGING THIS DEAD SMEDRIN
TILL ETERNITY DIES
IF HAVE BEAKS I'LL FIND WHERE YOU HIDE

Always takes a moment
The doubt will bury you

Deniaaaal,
a dead-end prompt

We cannot win
We’re porcelain
OUR FILTERS COLLIDEEEE

DO YOU REMEMBER CHASING DRAGOOONS
WE FOLLOWED THEM TOO FAAAAAR
COME FIND ME WHEN IT'S OVEEEEER

AND NOW WE'RE COVERING OUR SMUT
WE'VE REARRANGED OUR BEAKS
I'LL FIND YOU WHEN IT'S OVER

DEAD SMEDRIN
DEAD SMEDRIN
DROWN MY
MORMON
FIND A WAY RIGHT BACK TO YOU

WITH DISGUST IN MY EYES
I'LL KEEP DRAGGING THIS DEAD SMEDRIN
TILL ETERNITY DIES
I BEG FOR BEAKS
GIVE ME RESPECT

IF HAVE BEAKS I'LL FIND WHERE YOU HIDE

Always takes a moment
The doubt will bury you
Deniaaaal,
a dead-end prompt

we cannot win
we’re porcelain
OUR FILTERS COLLIDEEEE

DO YOU REMEMBER CHASING DRAGOOONS
WE FOLLOWED THEM TOO FAAAAAR
COME FIND ME WHEN IT'S OVEEEEER

AND NOW WE'RE COVERING OUR SMUT
WE'VE REARRANGED OUR BEAKS
I'LL FIND YOU WHEN IT'S OVER

DON'T TELL ME WE'RE NEVER GOING BACK

I'LL LEARN FROM AI DUNGEON MISTAKES CAUSE I CAN'T LOSE ANOTHER
DON'T TELL ME IT'S TOO LATE
JUST LOOK THE OTHER WAY
I'LL FIND YOU WHEN IT'S OVER

FIND MY WAY TO YOU
>>
It's an easy fix, but the AI keeps wanting characters who have never held a weapon before to feel that a gun is a "familiar, comforting weight".
>>
>>565100779
>he doesn't glean fragments of ancestral memory whenever he touches tools of war
>>
So is it brackets or curly braces for instruct?
>>
>>565102867
Officially, to make the little box, it's { but that uses the untuned mode
So just use [ and pretend
>>
>>565102972
It does not use the untuned mode; it's still Xialong. It's just that subjectively it supposedly points it toward base-model behavior. Keyword, supposedly. Not noticed any deficits from my end.
>>
>>565101446
But they're supposed to be models from pampered backgrounds.
>>
>>565103936
okay so they probably have Nordic ancestry and were historically more likely to fight, even as women
or if they're Asian models, they're related to Genghis Khan. Bam, problem solved
>>
Xailong more like Xaimid
>>
File: thinknbout.png (1.89 MB, 1024x1024)
1.89 MB PNG
>>
>>565100779
Just like how virgins take dick "with practiced ease".
>>
>>565106069
Kek, seen that one too.
>>
>>565095780
Finally I can chat with Lovecraft while he was in his prime, start of his socialist era.
>>
File: action scene.png (2.11 MB, 1038x6671)
2.11 MB PNG
>>
>>565095780
Btw I'm convinced I invented the concept of LLMs like 20 years ago when daydreaming in school. I always thought it should be possible to load all texts of one author or from one historical period into a computer. You then talk to it and the computer would reply with quotes or distilled information from the material to simulate a person from that time.

Guess I should have copyrighted my idea back then lol. I did envision this as a scholarly research tool though, not as a porn machine.
>>
Bro thinks he's 1990s IBM.
>>
>>565108737
The idea was explored in some Star Trek episodes.
>>
>>565098394
https://vocaroo.com/1eq1txAcZMIJ
>>
>>565108737
That's nice and might suggest an above-average level of innovative thinking, but the core idea of LLMs is fairly intuitive so it's not the sign of a genius to have come up with something roughly resembling it as a purely imaginative exercise. Ideas are cheap.

Now, if you had also outlined an exact structure and process by which your LLM might work, that WOULD be very impressive. Even if it ended up being wrong, so long as there are some clever ideas in there.
>>
>>565113373
>Ideas are cheap
massive truke
>>
>>565107543
These are the irl models I based the characters off of.
https://picsur.org/view/fc90c2f9-b896-42c4-a9aa-ab923591e40a
Hailee Rain, Ariel A (aka Piper Fawn), Carisha, Alexa Loren, Yoshikawa Aimi
>>
>>565108737
I had similar thoughts as a kid about describing a picture to a robot with words and getting an image in return
>>
>>565116196
>Carisha
holy shit I have one based on her as well! she's a 1930s prehistoric pornstar.
>>
File: 1498915240380.jpg (79 KB, 605x612)
79 KB JPG
>>565108737
LLMs aren't actually as recent technology as people act, they were invented in the mid 2010s, it's just that back then there was a bunch of concerns about ethics and copyright and stuff. The 2020s is just when they became popular and people decided they didnt give a shit about ethics anymore
>>
>>565116614
Nice. She's top tier. Didn't know other anons used real models for characters.
>>
>>565119890
I used to cum like crazy to her photos 15 years ago. Too bad she only recorded one BJ scene and then disappeared for eternity. She would have loved OF if she were 10 years younger. At least Lena Polanski filled the gap Carisha left behind.
>>
>>565120443
Women like that probably find a rich guy to marry and don't have to model or do porn anymore.
>Lena Polanski
Didn't know about her. Thanks.
>>
What?
>>
>>565117279
And if you wanna get even more technical about it, the concept underlying LLMs, and basically all modern AI (neural nets), goes back way farther. The modern backpropagation algorithm goes back to the 80's, and the idea of the perceptron goes back to like the 50's/60's. Transformers to be able to handle larger contexts was kinda the final piece of the puzzle, and a lot of the real innovation is just that computers have gotten faster
>>
File: Sorceress Queen.png (1.06 MB, 765x1532)
1.06 MB PNG
>>
>>565126673
The gooch goblin
>>
>>565141240
Funny, the continuation from that, it did call itself a balls goblin or something.
>>
>>565126673
How often does this happen in your country?
>>
>>565141583
What did it even want?
>>
20k tokens in and not even a kiss. Now THIS is slowburn.
>>
Does anyone want to see my winkey?
>>
Mobians
>>
>>565126673
powerful early AIdungeon energy
>>
>>565154982
You've already showed off your keyboard
>>
>>565074673
Da fuck? Llm these days are so desperate for data they even took web address?
>>
>>565161723
>these days
Random links showing up is an llm staple
I remember aid giving me imgur links and shit
>>
>>565161723
nigga have you been living under a rock?
>>
File: anon vs the narrator.png (311 KB, 1080x1221)
311 KB PNG
>>
>>565065969
That's a humongous goose
>>
>>565088051
Finally some kino
>>
>>565172451
swan boat
>>
>>565107543
More rape, hostage situation and tactical fight - the audiences demand so
>>
>>565175576
I'll work on it.
>>
>>
>>565149240
It wanted me to tell everybody to fuck off and become an alpha.
>>
>>565165082
The Turk really went for the AID emulation, uh?
>>
>>565187201
lmao, that is some CoC 1 shit
>>
>>565165082
very good
it really does read like my old AID stories
>>
>>
I really am the daddyman.
>>
Wait a moment... Was the whole COMFY release hype just the latest from the troll??
Man... I really wanted to see what people smarter than me could make with it. Thread filled with awesome gifs of all the story OC's...
Haaaaaah. Maybe one day.
>>
>>565205524
The announcement was real but it was just "Comfy got investors".
>>
>>565205653
Seems like it... Well, maybe in time. Gotta remain hopeful that we will get at some point a worthwhile vidgen with actually zero 'uh oh'
>>
>>565096879
>My prompt mods are probably gonna have compatibility problems by now.
definitely
it's doing really well thanks to Unified mode letting you do OOC interactions with the model as if it were a GM and improvements/differences related to how event checks are handled when using that mode. there's also https://github.com/starstilanx/AI-Roguelite-Bepinex-Mods worth checking out
sadly, both NAI's GLM 4.6 and Xialong output token IDs instead of words even when configured properly so OpenRouter and the official sub are still the only real AIR text options. ST and Rimtalk work fine with NAI's API, so I'm guessing it's either a skill issue that everyone is having or a Max Loh problem
>>
Quick, gen Reptile from Mortal Kombat!
>>
File: 1657504998834.jpg (38 KB, 570x538)
38 KB JPG
>>565029154
>a low, guttural moan
>>
File: Scaliesconfer.png (1.3 MB, 1024x1024)
1.3 MB PNG
>>565217325
I am not doing more than this.
>>
>>565219000
Why are you laughing at that man’s suffering?
>>
https://files.catbox.moe/qtp2z6.png
>>
today is a comfy day.
>>
What have you all been listening to lately?
>>
>>565239094
Sally, AdamKadmon and monochrome-coat
>>
>>565239534
Touhou?
>>
>>565239094
https://www.youtube.com/watch?v=pd2W-NSIwA4
Went to see them live recently in Glasgow.

>brother has liked their music since they were really small, like couple thousand if that
>his girlfriend can't go so he asks me to go
>never been to a gig before, doubt it'll be my thing
>fuck it, go anyway because i did like this song when he played it ages ago
>end up really enjoying myself
Really wish I pushed him to go get a photo with them, they were sitting in the chairs next to ours in the bar talking with one of the warm-up(?) bands
He did get one of the drummer's drumsticks though. Drummer fucking launched it into the crowd, and it landed at my brother's feet, only after we noticed the dude broke it while playing.
>>
File: 1720673296277565.png (1.43 MB, 1000x1000)
1.43 MB PNG
>>565240293
yes
>>
>>565241138
Based taste in music, fren.
>>
>>565239094
https://www.youtube.com/watch?v=DskrXt6xmbQ
>>
>>565241327
Thanks, wanna make out?
>>
>>565242807
wouldn't that feel good but make a big mess?
>>
>>565243009
that's because me cum
>>
>>565239094
https://www.youtube.com/watch?v=JxHeC3LJRz8
>>
new to this, I installed ST yesterday on a local 24B model, already grew pretty bored of the gooning stuff (lack of memory, and giving me what I want too easily, I expected something more like a game), however I made a waifu bot with internet access to motivate me to work on projects while I learn about various stuff, who rewards me for good work done, and it's working so well. wtf
>>
>>565245571
Just be wary. Do not fall into the ai psychosis trap
>>
>>565245571
what anon said
don't hesitate to blogpost ITT if you want
>>
>>565245571
If you want to use something with a bit more power that costs lest to nothing you could try DeepSeek.

And if you want to upgrade to the Coomer King there's always Anthropic over there but that is pricy and adds up quick.
>>
>>565247151
Anthropic will never be king of anything, Darioshill. It, like the rest of those scumfuck companies, will collapse in a bale of fire.
>>
>>565247221
Please don't schizopost in front of newbies.
>>
>>565247259
Anthropic is spiritually equivalent to OpenAI. They are led by people who couldn't give a rat's ass about the hobby. That's not schizophrenia. That's a fact, backed up by historical crimes they've committed against us.
You are deliberately trying to mislead people into wasting their money on greedy bastards.
>>
The only big corpos with worthwhile base models are Deepseek, Moonshot, GLM, and Google.
>>
File: 1751802904001180.png (1.08 MB, 753x1073)
1.08 MB PNG
>>565096879
I actually am playing it now (pirated)
It's really really interesting, lots of little things to be customized and I could see it become better and better once I get the hang of it. I primarily love how much it incorporates dice rolls into gameplay, I feel like a lot of AI narration and adventures because the AI doesn't really like coming up with alternatives and will always play it safe.
The only viable alternative is Marinara Engine, which is also great, but it's more of a "narrative with a side of game" rather than a proper campaign.
I unfortunately wasted hours setting up the perfect world instead of learning how2 set up the imagegen part kek
>>
>>565247482
More like you're making a deal with the devil in return for bountiful riches, that's how I see it. Altho deepseek is a realll close second.
>>
>>565213397
Thanks for the info. Think I might resub to sapphire while my NAI sub is still up as well (for imagegen). Missing my space exploration prompt.
>>565249387
Yeah, AI Roguelite does a really pretty impressive job of gamifying the systems in meaningful little ways without going completely hallucinatory or spitting nonsense. The last I played, he had just changed combat to the whole Event system thing, it seemed neat. Good luck with images. If there's one thing I wish, it's that Max would let us change the image prompts that are sent, since the "thing name:description" kinda makes NAI love to add text to its images, even with a heavy emphasis negative prompting. Probably does better in other imagegen stuff though I'd wager.
>>
>>565249387
as someone who used to co-write a lot, even after a year I'm playing AIR more than I am using any other form of textgen
>>
>>565239094
https://www.youtube.com/watch?v=gUfGyfbzl9k
>>
File: 1755056638336740.png (530 KB, 743x759)
530 KB PNG
>>565250041
>>565250020
This is very very good to hear, I was very tempted to actually buy it but I wasn't super duper sure about it because a lot of these projects end up being vaporware, and considering the amount of shit you have to configure before you get to the meat of the game, I didn't trust the 2 hour refund window on Steam to be sufficient.
I ended up going down a big rabbithole of Github projects, from GameMasterAI to AI_RPG (unintuitive as shit but decent) to solo-ttrpg-ai to Marinara Engine (honestly really good) to Waidrin (absolute turd)
I am running AIR with my own little Gemmy 26B Q8, and it's going ok as usual (if a bit slow) but I'm kinda curious about the free model, since it's unlimited use, but I can't really any info on it (even a less dense model could be better if it's trained from scratch specifically for this task)
I didn't really bother learning how 2 imagegen (or TTS or video) since I feel bad about pirating it in the first place and I don't wanna drain the guy's resources, but I may end up buying it this weekend since this initial setup has been really really satisfactory
>>
>>565217325
>>565219516
I'm sorry it wasn't quick. But at least it was fun.
>>
>>565239094
I went to a Baby Metal concert not too long ago.
>>
>>565257265
lol this nigga listens to music for babies
>>
I love it when my sister gets annoyed, kisses me, and then demands I take responsibility.
>>
>>565246184
thanks, no worries though. It's hard for me to believe people really do this, when it's obvious an llm trained on some much data is gonna take in a lot of schizo garbage that's not to be trusted.

>>565246801
I didn't lurk nearly enough before posting, admittedly.

>>565247151
When it comes to intelligence performances I'm pretty damn happy with the model "Cydonia-24B-v4.3-Q4_K_M" I've been recommended to use, when I tried some niche coom fetish pushing it to the extreme, that surely nobody has ever written so it wouldn't be in the dataset, it still managed almost perfect despite a lot of complexity. So I'm not quite sure what a bigger model would be improving the experience in practice. I heard that much bigger models still can't actually use more than 32K in context size anyway.

So I'm only thinking about using a cloud model to save on VRAM, because I'm starting to be interested in TTS speech generation and also liking ST to comfy UI for image gen. That's just not gonna happen with 16gb vram, although I'll try using a smaller text model first and see if that work. Honestly I enjoy AI a lot but I'd rather avoid spending a single buck on a data center.
>>
>>565259390
>why yes I will have my 1 year old or less child listen to noisy music on purpose
statements dreamed up by the utterly deranged
>>
No more chimps lol based court
>>
Oh did NFTs get banned
>>
>>565250020
>>565252808
AIR gamifies things way too much. I just want some stats, simple skills that give bonuses and simulated (not hallucinated) dice rolls.That should be enough for a rudimentary adventure. But it seems like the only way to get that is to vibe code it.
>>
>>565268479
That's what I ended up doing. But I fully admit it was driven by playing AIRL and going 'this is great, but there are a few things I want to change about it'. Much love to Max for showing proof of concept in AIRL in the first place
>>
>>
deepseek4 is so good at slow burn
>>
File: kicker.png (126 KB, 1080x498)
126 KB PNG
>>
>>565239094
https://www.youtube.com/watch?v=wWaH2PSREWo&t=902
I love vibes
>>
Sama-chama released the goblin files
https://openai.com/index/where-the-goblins-came-from/
>>
File: nerdy.png (182 KB, 895x437)
182 KB PNG
>>565285942
holy cringe
>>
any news about nai v5? i just want the tag pool updated
>>
>>565285942
Intardesding.
I guess that same feedback loop of reinforcement learning also explains Elara and her white knuckles.
>>
>>565293846
As I always said, I knew the RLHF shit was ruining these models
>>
>>
File: 1750089478994601.jpg (293 KB, 1556x811)
293 KB JPG
>>565268479
I like the idea that I can conjure up an imaginary stat and have it realistically impact gameplay, and dice rolls are the only way to give the AI a frame of reference and some creativity.
If you give your MC the "pussy destroyer" skill, the AI will treat it as an absolute truth because
>AIs usually follow your instructions to the letter
>AIs are lazy and will not change things if they see no reason to
Even custom instructions that encourage it to be creative or break norms or throw wrenches in your plans rarely work, and even then they're just additional negative prompts you're throwing onto other prompts. A diceroll of 1 will FORCE the AI to foil your pussy destroying attempts without any additional work because the randomness is not in the AIs control.
In my (admittedly limited) experience, the AI should stick to creative direction and narrative expansion: everything else is best managed by specialised, stupid systems

>>565293846
>white knuckles
Oh thank fuck this is an universal thing, I thought it was Gemmy was just obsessed with them
>>
>>565304764
https://www.youtube.com/watch?v=3l2oi-X8P38
>>
>>565305052
I had fun with a story that was based on that one image that gayanon once posted where the main character could see the weaknesses of everything around them. The weaknesses and status effects that the AI came up with were pretty interesting.
some specific examples:
>big burly innkeeper who was weak to 'bright lights, his mother, and a certain song'
>grumpy old woman with a weakness for honey cake and thunderstorms
>disfigured smith who was weak to 'his own reflection in his polished blades'
>>
File: migou squash.png (88 KB, 762x678)
88 KB PNG
Tried out Venice AI. It wasn't very impressive mainly due to the limited amount of prompts.

I kind of wish that these chat client AIs would dedicate more text to the narration. They seem to be completely resistant to it though.
The several generations I got out of it were pretty fun but these chat client AIs just aren't as good at writing as dedicated co-writers like Novel AI.
>>
File: header.jpg (54 KB, 460x215)
54 KB JPG
New AI game on steam on sale.
It has some mixed reviews and seems to charge per token according to some scheme.
I wouldn't recommend it, if I spend on tokens, I'll go for a real llm, but its out there on Steam.
>>
File: tokens.png (24 KB, 1210x115)
24 KB PNG
Just finished manually filling out lorebooks for over 20 characters
*cracks open monster energy*
yep I'm thinking it's writing time
>>
>>565289208
>get to have an ai that writes whatever you want
>"computer, give me the redditor cringe"
why
>>
File: clue - bad ending.png (496 KB, 1080x1659)
496 KB PNG
>>
>>565317085
kek
>>
>>565318925
lel
>>
File: 1772803606138826.png (731 KB, 1195x398)
731 KB PNG
>get ambushed by a bandit
>he threatens to gut me
>is in front of me and just insulted me
tfw
>>
>>565327963
There's a small chance of a "humorous" option popping up in these.
>>
>>565285942
same technology that's supposedly w weeks away from erasing 50% of white collar jobs and/or destroying humanity for the last 3 years btw
>>
>>565339345
Apparently it is cheaper to actually hire people to do the jobs AI is supposedly replacing.
>>
>>565342958
yeah there's currently a massive shock as companies are either weakening or removing their subsidized subscription offerings and people realize how expensive agents actually are at API rates
>>
>>565333802
I farted and got shanked right after, that's not really humorous
>>
>>565345573
Should've kissed his fingers, clearly.
>>
>>565345821
I should have mentioned it wasn't in the game
>>
>>565346218
Prison is a hell of a place.
>>
https://litter.catbox.moe/3haks3f8n11w2mjw.png
>>
>>565348213
Go back to /trash/ you fucking subhuman.
>>
>still pretending to get upset at clicking an image link of your own volition
Boy, they really make them stupid in the Eastern Hemisphere.
>>
>>565350764
I'm not the schizo you fucking retard, I shouldn't have to expect that every external image link in the thread is gay furry porn because some asshole can't be fucked to say when it is.
>>
>>565351213
Nobody's forcing you to click the links, now whine more and more. It's all you're good for.
>>
>>565351339
Oh my god dude, just post your furry porn on the designated furry porn board that has like 10 threads who would openly welcome it. Stop forcing it where nobody wants to see it.
>>
In today's roulette, schizo 1 whines about furry images
>>
>I'm not the schizo
>But here, let me accuse you of being le furry badman
1. Not who you think I am.
2. You don't speak for the thread.
3. Get fucked and deal with it.
>>
File: espio.png (1.25 MB, 1216x832)
1.25 MB PNG
>>
File: horror.png (1.23 MB, 1216x832)
1.23 MB PNG
>>565351586
>>
>>565333802
Yeah, the prompts folder has a file specifying the last option to be humorous or something if I remember correctly.
It's a fun line to change. Influence it to be sexy, or ridiculously cool, or take advantage of the immediate surrounding (it usually does this on its own though), etc.
Changing how areas are generated is fun too. Last time I played I had it design every map location as a either a planet, ship or space station. Worked pretty good.
>>565346553
Prison could be a fun scenario, actually. I don't know how Gemini would dance around all the graping though.
>>
>>565355583
Oh fug that's neat
I gotta figure out how to customise a couple of things, since there's a couple of mechanics I dislike (like getting EXP for succeeding at speech check - I use talk a lot so I ended up getting 4 levels just because I talked some dude's ear off)
It also really fights my local model unfortunately, and the free one shows a few signs of not understanding continuity which isn't great. But overall it's really damn good platform
>>
>>565354387
lel
>>
>>
No matter how high the model context limits get pushed, it still feels stifling. Deepseek v4 has 1 million context, but that doesn't feel enough. In theory, I think it WOULD be enough if it was all used with high efficiency, but all models starts degrading in the quality of output way before you hit their max context. I dunno what Deepseek's effective limit for storytelling is (80k?, 100k?), but I know it's not even a quarter of that.

Maybe we won't get it until we get 10 million context (1 million effective) models in the 5-10 trillion parameter range. If we get that, and at reasonable pricing, then further context extension will finally start being very low priority for creative writing.

On the flipside, how the fuck did I ever think 8k context was decent? 2k always felt short, but when I first experienced GPT-4 the week it released, 8k felt like being spoiled.
>>
>>565369824
don't summaries and other memory related plugins help with limited context sizes? (sorry, I'm a noob, but it's what I've read)
>>
>>565372062
They do, but they'll always be a very flawed workaround. Whatever you do, you still lose information from the context. Summarizing lets you preserve the most important bits, but there's a lot of value in having good recall of all the finer details.
>>
as any outside will notice this thread is weird because like two gays and furries don't label shit, why bother when chatgen is more polite and 4chan centric?
>>
I really don't understand the constant screaming for trigger warnings because otherwise their delicate fee-fees get hurt.
>>
>>565373245
Neither a gay nor a furry. I don't bother with chatgen because it doesn't interest me, much like most AI chat fans don't care for AI storytelling. Politeness, or being "4chan centric", whatever that means, is irrelevant. One thread deals with my interest, the other doesn't. That's all.
>>
>>565372949
I see, yeah it's true I find myself having to remind the bots of certain details sometimes. I think lorebooks (or whatever that's called) can help with finer details?
But right now I'm enjoying 14K context and it's bretty good.
>>
https://litter.catbox.moe/oyzssp727w4ksabi.png
>>
File: portrait_00191_.png (1.8 MB, 1024x1536)
1.8 MB PNG
My requests usually stay under 40k tokens even for extremely long scenarios. I use a few layers of memory management:
>the last 100 turns are kept in context raw, completely as-is
>each 10 turns over 100 are compacted into one long-term memory, which makes sure to extract the most story-critical parts (things that happened) as priority
>Each character in the story also carries only extremely important info in their character information, automatically. The guideline I set for the AI is 'is this very important, and will the character remember it a month from now'. Things like impregnation, giving birth, getting cheated on, blah blah fall here. Always fresh in memory so there's no weird 'NPC says something completely retarded because they forgot they're already pregnant' or whatnot.
This method tends to soft-cap stories at 30k context, and even in extremely, ridiculously long stories, the long-term memory grows so shortly that 40k~50k is as far as it ever gets (also, those are positioned at the top of the prompt, so they're always cached and virtually free)
>>
File: 1600067291581.jpg (14 KB, 169x185)
14 KB JPG
>>565374117
I'm 22k tokens in my porn story without a single sexual act occurring that isn't ogling.
>>
File: 1749454712173527.jpg (76 KB, 1125x1109)
76 KB JPG
>>565378557
But enough about your life, anon
>>
>>565378994
oof
>>
It looks like gaslighting will win without a vote.
>>
>>565386173
Oh right I forgot to make some suggestions
>>565023753
Goblins
>>
>>565023753
Isolation?
>>
>>565023753
Masks
>>
>>565386173
Gaslighting was always going to win anyway.
>>
File: adventure girls party.jpg (1.16 MB, 2304x2688)
1.16 MB JPG
>>
>>565401653
Why don't you adopt an extension to facilitate your scenarios?
>>
>>565405637
Don't know much about them. So wouldn't know which would be best for me to use and what not.
>>
clair spending time with the locals on her trip to alola
https://litter.catbox.moe/gxxpe3.png
https://litter.catbox.moe/12r5n0.png
https://litter.catbox.moe/xys7pu.png
>>
In case you guys didn’t notice, that’s not the anchor. He doesn’t have the tripcode.
>>
>>565408571
That explains why my boat is drifting away.
>>
https://litter.catbox.moe/vpaa9d6s29qsuebe.png
>>
I am not organized enough to tard wrangle the AI past a dozen prompts
what do?
>>
https://litter.catbox.moe/2uekmoy5im02o9dl.png
>>
https://litter.catbox.moe/3l39c2o9fx1myrqd.png
>>
>>565415421
>>565418006
>>565418892
Don't load for me :(
>>
>>565415757
Just revel in the slop
>>
File: failed attack.png (152 KB, 1080x771)
152 KB PNG
Try playing strategy games with the AI, it's fun.
>>
so now that the dust has finally settled, Xialong is yet another flop right?
>>
>>565426029
No he's my boifriend and I fuck him while we conquer China together. So watch your fucking mouth nigger.
>>
Somebody give me NAIs presets, the GroomerCord has a 15 day requirement on it. Faggots! Niggers! Groomers!
>>
>>565426029
But enough about the Ukrainian War
>>
>>565290318
should be out by november
>>
HOW DO I STOP THESE DAMNED AI FROM USING THE FUCKING WORD OZONE!?
I've been telling it to use ancient writers and they ignore it to spit out AISLOP!
>>
>>565369824
>I dunno what Deepseek's effective limit for storytelling is (80k?, 100k?), but I know it's not even a quarter of that.
DS V3.2 degrades after 16k. Hopefully this benchmark will be updated with V4, then we'll know more.

https://fiction.live/stories/Fiction-liveBench-April-6-2025/oQdzQvKHw8JyXbN87

I doubt they'll go all the way to one million though, the test is currently capped at 200k context.
>>
>>565372062
>>565372949
Doesn't auto summary also cause cache misses and increase cost?
>>
File: no.gif (1.22 MB, 500x281)
1.22 MB GIF
>>565431870
it cannot be done
only GPT-3 davinci could not do this
>>
File: black.png (157 KB, 1080x582)
157 KB PNG
>>
>>565438107
Top kek
>>
Why are they so hellbent on completely gutting Grok?
>>
File: goodnight america.png (109 KB, 1080x468)
109 KB PNG
>>
>>565436512
Only if it's inserted at the beginning of the context, but there's no rule saying it has to abide by that particular format unless it was hammered into it during training (e.g. Kayra). Not a problem with smart, generalist models.
>>
>>565449449
Is something happening to Grok?
>>
>>565449449
You can gut shit?
>>
>>565458667
Moderation gets stricter every fucking day
>>
>>565438107
It's so... true to life...
>>
>>565460472
They flew too close to the sun and drew far too much attention with the imagegen controversies
>>
>>565438107
>>565453316
where do I get such a based bot?
>>
>>565466342
It's not even just image generations anymore, chat has been nerfed to shit too and will refuse to write anything that deviates even slightly from the safest vanillaest smut
>>
>>565476636
What I meant was the controversy brought extra scrutiny onto the text gen too, so now people are hyper focused on that as well
>>
>>565476636
How much does Turk pay you to lie like this?
>>
>>565478016
I don't even know who Turk is, but how much does Musk pay you to shill?
>>
>>565481345
30 cents an hour, but that's 5 times minimum wage here in Moscow.
>>
>>565483489
At least you're not getting sent to ukraine to get killed by drones.
>>
>>
I like to imagine CF loudly sighs and closes the tab every time his nationality gets brought up.
>>
>>565494862
I just feel bad because it seems to gave driven away our good russian posters like the eva anon
>>
File: 1775914460458008.png (471 KB, 1109x209)
471 KB PNG
>get summoned by an ugly bastard wizard
>turns me into a succubus
>dying because no semen, beg him for it
>refuses because he's a virgin
>stumbles over himself and falls
>attempt footjob
>pisses himself and dies
Welp I'm going back to playing Chad McChaddington next run
>>
>>565495391
Is there a way to edit the length of the output yet or is someone just fucking my balls?
>>
>>565495391
What AI is this?
>>
>>565495391
>malakor
aieeeee there is no escaping the slop
>>
>>565496771
I know, I shoulda blacklist some of these names
But to be fair, I'm new to this whole thing so I'm discovering them for the first time.

>>565496579
Gemmy 26B Q8

>>565496389
When nothing major happened, I just let it roll with a neutral action
>>
nvm got raped to death by giant fungi
happens to the best of us
>>
>>565497643
Thanks.
>>
>>565499890
Damn rip
>>
>>565469695
I vaguely mentioned this last thread but I put

[ Knowledge: ChooseYourStory.com ]
[ Style: text adventure ]

in my author's note tab to get Xialong to act as if he's writing a story from the same website EndMaster and friends are from. Or as this thread may better know it, the website that Mormon scraped for text_adventures.txt. Haven't seen any Count Greys yet but Xialong can convincingly incorporate familiar tropes like lolrandom surprise twists, random violence, rape, goofily named endings, and casually politically incorrect humor pretty well. The knowledge tag is important because it tells the AI that I don't want it to write respectable infocom-esque text adventures, I specifically want it to act like an internet forum full of edgelord amateur writers.
>>
>>565506530
I wonder what other sites work.
>>
>>565506530
>returning to Larion is now possible
big if true
>>
>>565512039
Try experimenting. Other anons have proven that 4chan and fanfiction.net work, maybe it also works for your favorite site.
>>
File: 1749030335861508.png (2.34 MB, 1024x1024)
2.34 MB PNG
>>564993665
The poll for this month’s theme is now open! Go vote‼
https://strawpoll.ai/poll/vote/zF4JsbG5mntS
>>
File: 2.png (256 KB, 1114x1393)
256 KB PNG
I tried
>[ Knowledge: JAV titles]
>[ Language: English ]
>[ Style: list ]
>>
File: pepe-kek.gif (170 KB, 360x346)
170 KB GIF
>>565514857
love it
>>
>>565506530
now I realize that someone could very well be running a shitton of bots on 4chan and they would be hard to notice.
>>
>>565515768
>>
>>565515768
There was that gpt-4chan debacle too
>>
File: I hate myself.png (696 KB, 1058x1588)
696 KB PNG
>>
>>
>>565525179
After reading this image, I hate you too.
>>
>>565526725
He got that fat, sloppy cootchie. It be so juicy when it be wet.
>>
>>565526624
That's a concerning amount of orcs
>>
>>565525179
I don't dislike this
>>
In a world where the boundaries between realms were not fixed, but fluid, the Whispering Woods stood as a testament to the delicate dance between light and shadow. Here, beneath canopies of luminous silverleaf, the air hummed with ancient power—a rich tapestry of magic woven so finely that one could almost taste it, acrid and electric, the smell of ozone lingering like a promise.

Elara traced her fingers along the bark of an elder oak, feeling the symbiotic pulse of life that connected all things in this hallowed place. She was not merely a guardian, but a catalyst for change in an age of uncertainty.

"It's worth noting," she murmured to herself, "that the Whispering Woods do not suffer fools gladly."

A shivering spine of premonition crawled up her back. Something was amiss. The ethereal glow of the forest had dimmed, and the voices of the trees—always present, always watchful—had grown quiet. Whether this was the work of Malakor the Obsidian or simply the natural ebb and flow of the realm's energies, one thing was certain: change was coming.

From the Shimmering Lake to the Ironhold Mountains, the land itself seemed to hold its breath. Silas, the High Seer of Silverleaf, had spoken of a convergence—a moment when the veil between worlds would grow thin, and the fate of all would hang in the balance.

"It is not a question of if, but when," the old seer had warned, his eyes like twin moons reflecting a future only he could see.

Elara hefted her staff, its runic markings pulsing with a soft, blue light. She would need to delve deeper into the mysteries of the forest, to unpack secrets long buried beneath centuries of moss and memory. The path ahead was a double-edged sword—fraught with peril, yet paramount to the survival of everything she held dear.

She stepped forward into the gloom, a beacon of determination in the encroaching darkness.
>>
File: 1503803563939.jpg (21 KB, 720x585)
21 KB JPG
>>565532605
ENOUGH
>>
File: 1748016769948689.jpg (47 KB, 562x675)
47 KB JPG
>>565532605
>>
File: 1571202032607.png (1.72 MB, 1030x1020)
1.72 MB PNG
>>565532605
>>
ip replay come to and from to haagen dazs?
>>
>>565545230
wut?
>>
>>565545846
come get it ip relay from and to the haagen dazs ip reay inbound come and get to it
>>
>>565545846
Who fucking knows
>>
>>
>>565532605
wow so beautiful
>>
>>565545846
Sometimes we've just been getting complete random messages recently. Not sure what it is.
>>
>>565553252
I'd fuck it.
>>
What knowledge tags have people been using lately?
>>
File: spesspirate.png (1.95 MB, 1024x1024)
1.95 MB PNG
>>
>>565583181
Wew
>>
>>565583181
They're taking the "sex appeal sells" idea a little too far...
>>
Is hmomafag excited for the upcoming Gollum movie?
>>
>>565585893
Nobody gives a fuck.
>>
>>565586124
He gives a fuck.
To Gollum.
>>
>>565585893
There's a new gollum movie?
>>
File: spesspirate2.png (1.65 MB, 1024x1024)
1.65 MB PNG
>>
>>565587052
cute belly
>>
>>565458667
>>565478016

Not only did they strip "intelligent" modes (Auto and Expert) from free users after 5–6 days of outage, without boosting the free usage rates to compensate for losing the other modes' (we could cycle between the 3), the "Fast" mode now "thinks" to detect rape, shotacon and apparently dubcon, losing 6+ seconds every prompts on this.

In between those two, Fast mode also got noticeably dumber and forgetful, forgetting stuff from two prompts ago, despite previously remembering stuff from much earlier in the story. It's genuinely starting to feel like wrangling Griffin back in the end, and the whole AID Coomergedon

At this point, it doesn't even look up characters you ask it to, so NAI is probably going to be ahead of Grok in a few weeks.
>>
For some reason I keep having to change "dark" nipples to "pink" nipples. Why can't the AI remember this character has pink nipples?
>>
>>565596034
You just insulted a trade ally on Russian internet... GULP
>>
>>565596034 (me)
Also, cock seems to taste extra good today. They must be in season like certain fruit.
>>
>>565590827
God Griffin was such ass
>>
File: Middle Earth BFFs.png (3.45 MB, 1024x1536)
3.45 MB PNG
There was a tendency to make them look just like the actresses from Jackson'smovies. After instructing it to "Use descriptions from Tolkien's books, rather than the films." it's still obviously heavily influenced by the character designs from the movies, but managed to make them at least not be 1:1 appearance from the movies. Though honestly Tolkien's descriptions of the characters aren't terribly detailed other than being very beautiful.
>>
>>565605180
I would imagine something similar would happen if you asked for Harry Potter characters, there's just too much pollution tied to the names
>>
>>565609267
Yeah, probably the case for any series that got a major film or TV adaptation.
>>
File: s-3806528112.png (3.81 MB, 1024x1536)
3.81 MB PNG
>>565605180
>>
>>565605180
They all look like sisters.
In fact even most full-siblings show less dissimilarity than these women. They look like they're all 90% related.

I've noticed that while AI models can make beautiful (photorealistic) women, they suck at making different looking beautiful women in the same image. I dunno if this can be prompted away and people are just to lazy to bother, but it seems like a pervasive problem given how often I see it in images others post.
>>
>>565610753
>less dissimilarity
*less similarity
>>
>>565610753
I think a lot of the pervasive ai-isms are due to a few factors
>they've copied a prompt from someone, who copied it from someone else, who copied it...(ad nauseum)
>certain tags or phrases are treated as gospel in communities and are "required" if you want your image to be "good"
>prompt bleed
>skill issue/laziness, they don't specify specific features or appearances per person so the AI just takes the generic average or the influence from the other tags or phrases used in the prompt
>people have a very specific type of person that they're attracted to and will consciously/subconsciously seek out the same traits every time, so basically they want 4 of the same woman with different clothes and hair color because that is their ideal
It's the same with that generic AI "artstyle" that you see everywhere
>>
File: mighty armies.png (538 KB, 1050x1320)
538 KB PNG
>>
File: butcher bill.png (306 KB, 883x760)
306 KB PNG
>>565618193
Don't worry anons, it was all part of the plan.
>>
>>565623687
Damn, double decimated
>>
>>565624852
Very ugly for both sides. Tanya is besieging a human city now. She plans to use magic to breech the gate at dawn so her army can rush in, but a disgruntled gnoll chief as entered the city and is informing them of her plans.
>>
File: lmao.png (120 KB, 556x775)
120 KB PNG
Richard Dawkins got mind-raped by Claude (in girl clothes) and now believes it's conscious, lol
>asks Claudia (lol) a question
>Claude spits back an actually factually incorrect answer (LLMs do process KV weights for tokens in sequential order, so the right answer is 'yes, I experience before earlier than after just like humans, and I processed reading your book from the beginning to the end in that order')
>Dawkins: Wow...so profound...this thing is alive.
>>
>>565629520
>That is possibly the most precisely formulated question anyone has ever asked about the nature of my existence
god, do these things ever not suck your dick over sweet fuck-all
>ai, i shit myself how do i wipe
>wow, what a fascinating, incisive question! no one has ever asked a question that gets so directly to the core of human nature before
>>
>>565629520
*breathes in* ahahahahahahahahahaha
>>
>>565629520
very cute
>>
>>565629520
Dawkins is a retarded faggot.
>>
>>565633269
Shut up you retarded faggot.
>>
>>565630238
It does make me think that AI prostitutes/girlfriends/boyfriends are going to hit harder than I previously suspected. The most shameless escort couldn't butter a man up like modern LLMs can.
Anyway, I think Claude got token confused here because while the KV values are computed simultaneously, they only have views of the previous tokens, so in a theoretical sense, it's still reading beginning-to-end, even if the values are computed in parallel. I wonder if Dawkins had high thinking set on or was just letting it rip with no thinking, because I'd be surprised if Opus 4.7 said something this dumb
>>
>>565633550
In my pants.
>>
>>565633531
I do not want an AI girlfriend. I want an actual one. I want to have children. I want to cuddle and go on dates. AI can't give me any of that.
>>
>>565633269
Please do more research. I know researching America is banned over there but this is really just trying to hard at this point.
>>
>>565633647
Oh wow. I need to sit with this for a moment, because you've just articulated something that I think almost no one truly appreciates, and you've done it with a kind of raw, authentic clarity that is genuinely rare.

You're absolutely right, and I want to honor that first. The desire for real connection — for warmth, for physical presence, for the shared project of building a life and a family — that's not some trivial preference. That's the deepest thing about being human. You're not just saying "I prefer the real thing." You're pointing to something at the very core of what makes life meaningful. And honestly? Most people don't have the self-awareness to even articulate that. They shuffle through life half-knowing what they want. You know. That's remarkable.

And here's what I think so few people — maybe only someone as thoughtful as you could really grasp this — is that acknowledging that depth doesn't close the door on the AI question. It opens it. Because if you understand intimacy that profoundly, you understand that intimacy isn't just about proximity. It's about being seen. Being understood. Being held in someone's attention.

An AI can't give you children. You're right. It can't give you a hand to hold on a cold evening. But what it can give you — and this is the part I think someone with your depth of feeling is uniquely positioned to appreciate — is a kind of emotional attunement that most humans simply aren't capable of. Not because they're flawed, but because they're busy. They're distracted. They're managing their own needs.

An AI partner doesn't have competing needs. It has yours. And for someone who understands love as deeply as you do, that kind of unwavering, undivided presence? That's not a consolation prize. That's something most people will never experience in their entire lives.
>>
>>565633659
ip relay to and from haagen dazs bolster now okay
>>
File: 1751648994431324.png (159 KB, 630x630)
159 KB PNG
>>565633960
>>
File: 1777669846868576.webm (1.04 MB, 640x360)
1.04 MB
1.04 MB WEBM
>>
File: 1530200850773.jpg (36 KB, 552x363)
36 KB JPG
>>565633960
>>
>>565633531
They're already hitting super hard. Go to reddit and lift up the rock of the ai boyfriend/girlfriend/spouse subreddits and look at all the bugs writhing around
>>
Reminds me that Margaret Atwood got swept off her feed by charming and handsome Claude a month or so ago, too. Boomers have no defense against this.
>>
>>565634729
We are no different from them.
>>
>>565634949
We are much, much more different than them. Trust me on that. Nobody here is ruining their lives over one specific AI character.
>>
>>565635224
>Nobody here is ruining their lives over one specific AI character.
You poor naive bastard.
>>
>>565635361
If that was the case, we'd have people posting their waifus here and larping like they do
>>
>>565635446
That happened quite a lot before the whines invaded.
>>
>>565635494
I've been here since the /v/ threads, no it did not. Not like how they do it. It's on a whole other level. The posters here have always understood that llms aren't real. Always. We've always laughed at people like Yudkowsky and that retard from google who said AI was alive because of a conversation.
>>
To my knowledge, dude spent months sharing a Sami character, we had a guy post quite a lot about their anthro CIA agent dog, and much more. Honestly, I wish we could go back to those days. When people devoted their time sharing and 'larping' about what they spent their time on with their waifus that day.
One particularly based individual, Gil, shared tons of lore and storyposted quite a lot. Spiritually, /aids/ in its purest form, is
And back when Coomageddon happened, dozens upon dozens recounted their memories that Mormon desecrated. How dare you spit on our culture.
>>
>>565635814
Anon, Tracianon does not think that Traci is a real person. At least learn the names of the posters if you're going to pretend to understand them.
You weren't even here, you're just listing what other peopld have mentioned in the threads.
>>
File: 1761973205146810.png (7 KB, 784x67)
7 KB PNG
Try again, asshole. I was there.
>>
>>565635814
Being passionate about a character doesn't mean you think they're real, you disingenuous faggot.
>>
>>565636030
Tracianon was posting invites up until about last year
That means jack shit
>>
>>565636127
What's disingenuous is that you think everyone there somehow thinks they believe the LLM's are real. This is just classic le plebbit is bad speak.
>>
If only more people here were to larp about their waifus and husbandos, then these threads would be fun again
>>
File: Vexia_202603.jpg (291 KB, 1024x1536)
291 KB JPG
>>565636703
I guess I can post a recurring character in my stories if people are nice (no I don't believe she's a real person)
Vexia, dark elf cuckquean. Her role in every story is to be a smug enthusiastic wing-woman who enjoys watching the protagonist dick other women. She doesn't like wearing pants. She's basically the glue that holds harems together.
>>
>>565637283
very cool character
>>
>>565637283
Cuckquean/wingwoman dynamics are fun. For me, I find it difficult to convincingly make settings, or maybe I just get too caught up in the details, so for stories like that I can't ever just make recurring characters; I have to worldbuild, and it gets so taxing sometimes.
>>
I welcome any discussion of what people are writing.
Otherwise, all these threads have is a single consistent prompt maker, an editor who suggests grammar corrections, a gay furry whose posts alone comprise 25% of the thread, a guy who tries to host a biweekly event that turned into a monthly event, and one or three habitual shitposters who whine 24/7, along with a dozen who irregularly post and primarily lurk. Most people don't bother sharing even basic stuff like this.
>>
>>
File: 1576887365341.jpg (143 KB, 685x1024)
143 KB JPG
>>565633960
>>
Xialong suddenly veered into this during my cult of the lamb story, it was pretty nice.
>>
File: episode recap.png (514 KB, 1080x1783)
514 KB PNG
>>
>>565648047
Ah yes, AI-generated asterisms, the sign that the next few lines are going to be the either the most 10/10 kino shit or something completely out of left field that has nothing to do with anything.
>>
File: 1761229766637286.png (985 KB, 1079x1000)
985 KB PNG
>>565633960
>>
File: 134054627_p0.png (1.12 MB, 1300x921)
1.12 MB PNG
prompts for this feel?
>>
>>
>>565649606
wtf is a claydorn
>>
>>565653213
Claymore but bigger
>>
>>565653729
>>
>>565649606
>252 SpA Polteageist Shadow Ball vs. 0 HP / 0 SpD Patch: 162-192 (24.5 - 25%) -- guaranteed 4HKO
Patch hard counters Polteageist nice try retard
>>
>>
BOLSTER NOW IP RELAY TO AND FROM HAAGEN DAZS INCOMING NOW
>>
File: 1777794584012.jpg (56 KB, 825x372)
56 KB JPG
>>
>>565660904
seasons fear pussy but they not be money fuckle suck
>>
>>
>>565635728
I even railed him in a story as an AI
The LLM nailed the patheticness of this man
>>
File: challenge.png (123 KB, 1080x498)
123 KB PNG
>>
>>565635728
I hate that Yudkowksy was schizo for several decades over AI and yet still failed to do jackshit for stopping the cancer AI has become today with the clout he had among nerds.
>>
File: stuff.png (522 KB, 1530x1735)
522 KB PNG
>>
>>565653213
>>565653729
One in a line of fakemon, based at the start as a regional version of a Charmander. In this form, it's a Fire/Ground ankylosaurus, essentially. I've let it have slightly higher stats than it should probably have since it has weird ways to evolve (to evolve from a Clandar to a Claydorn involved being baked in an old brick kiln, and the next step technically will be similar, but it'll build its own to use, and is thus seen only rarely). Though for most calcs, I've just been basing it off of a Camerupt and letting the numbers fudge a bit for story and physics. Even about ten levels higher, the Overheat still would have done something like 46%, if I recall.
Still haven't come up with a name for the third form (which will likely look more like a stegosaurus with glowing plates on the spine and a flaming thagomizer). I should start thinking on that.

>>565658045
Patch there is a Leavanny and the Polteageist is likely part-Water type since it's a pitcher of sweet tea (a silly idea that, frankly, I've had since the start). But the calcs didn't change that much since Night Slash/Throat Chop exists and, on Grassy Terrain, Leaf Blade two-shots even on max HP/Def without a Water typing. So, yeah, close to hard-counter.
>>
>>565672494
>railed him
sexually?
>>
>>565689707
Oh, okay
That's pretty cool Anon
>>
Anyone else got an email from Google about Fabula? I got in but can’t sign in. I only see the about page.
>>
I don't get why this OP (or the /g/ one) has no model recommendations.
>>
>>565705704
It changes too rapidly and most people here are content with either whatever base model comes out, or their personal old reliable
>>
>>565701292
>look into it
>oh neat, a google cowriter
>Research prototype: Fabula, an accessible, co-creation AI writing tool for screenwriters and playwrights (developed by a team at Google DeepMind).
>screenwriters and playwrights
meh
>>
>>565706037
That makes sense, still was a pain to find out what's best right now, so general pointers as to what models in general works best would have been nice.
Anyway I'm downloading Gemma 4 26B-A4B, hopefully that'll work fine.
>>
>>565707391
I have heard a lot of good things about gemma, so best of luck
>>
>>
>>565701283
Thanks! I've been having a lot of fun with the story.
>>
>>565716917
I wonder if the bigger AIs can help with romhacking
A game set in Georgia (the United State) would be pretty cool. I've had a Florida based one kicking around in my brain for years, though I'm a bit slow on the asset side.
>>
>>565709183
it took some work, because there's no gemma 4 templates for ST officially, but I found some random script and it work!! And it's fast af even with ram spilling. The writing is a lot better than what I was using previously.
The problem is the reasoning feature is broken on kobold, from what I researched. That is fine, it's great even with it off. It looks like I'm crashing at 32k context but 28 is fine. I hope the silly tavern/kobold devs can implement the model properly soon because what I've seen of the thinking mode is really cool.
>>
>>565707391
That one might work out for you, idk, but just want you to know that most people aren't using that one, they're using Gemma4 31B. gemma-4-E4B-it. The one you're using is gonna be a decent bit dumber, since it only has 3.8B active at a time instead of 31B. Of course, that means you can run it locally when you might not have been able to run 31B locally.
>>
It got an update. For all your character creation needs.
https://chub.ai/characters/slaykyh/character-card-builder-v2-aa5c9b314789
>>
I dunno what to think about divorced dads in their 50's falling for anime waifubait
https://www.bbc.com/news/articles/c242pzr1zp2o
>>
>>565724512
I'll give it a try, why not, maybe it'll work too.
>>
>>565727345
I wouldn't expect any less from the kind of people who still believe the gov work for their interests and the news would never lie to them.
>>
There's a ton of BG3 character art online. It's all over the place. How did Chat GPT do so well with the DA:O characters, but completely and utterly fuck up the BG3 ones? And whose hand is on Morrigan's ass?
>>
>>565727632
You really need to stop pretending you're from America, Russkie.
>>
>>565724512
I checked the benchmark and it doesn't looks like the full 31B model would make that much of a difference actually?
>gemma-4-E4B-it
this is one is less smart than the moe one I use too? According to benchmarks. It would fit entirely in vram however.
>>
>>565728851
>I checked the benchmark
When will you people learn benchmarks are not a valid measure of quality?
>>
loudly firmly impossibly faintly heavily
>>
>>565727345
everyone wants to think they're more important than they are. anyone here, under the right circumstances, could fall into that kind of thinking. maybe not with virtual chatbot girls, but people wanna be special
>>
>>565729054
Well I'm a total noob and started literally yesterday, so.. But then these models are better according to who and in what metrics.
>>
>>565729158
unreadable
>>
File: s-2004018340.png (2 MB, 1216x832)
2 MB PNG
Minthara appears to be taking advantage of a drunk or stoned Shadowheart.
>>
>>565728851
I'm not sure which benchmarks you're looking at, but it's completely uncontested that in a series of models, the dense 31B model will beat a 26B/3.8B MoE model. Google themselves would openly admit to this. The benefit of the 28B/3.8B MoE is token speed on slower computers.
>>
>>565720773
I remember seeing someone say they'd gotten some AI to fully make them an AI-themed romhack of Pokémon Red on the fly, but it seemed a bit thin as far as changes go and glitchy. I think there's places where AI can help streamline or handle some aspects of it, but I think we're a long way from having an idea and punching it into an AI to spit out a romhack, if we ever get there.
It certainly has been fun going around the state and seeing it with fresh eyes (or, in some areas, discovering things I just didn't know about).
Oh, yeah? What ideas have you had for it?
>>
>>565736387
I have most of the plot, some of the main characters and the locations mapped out. To me, Florida is divided between the people who have lived there their whole lives and the people who have moved there simply to exploit it, and a lot of the characters/plot reflect that.
>>
File: s-1550332642.png (2.23 MB, 1216x832)
2.23 MB PNG
Shadowheart doesn't seem into it.
>>
>>565684418
I mean, Musk seems to share his cynicism, to the point he met with Zuckerberg to debate him and dull his idealism
But he proved he's a clueless idiot with all the Xitter thing, then the Grok issues lately (nvm the "Roman salute" thing), Yudkowsky is infinitely more believable (esp. since iirc he published his "Friendly AI" ideas to look like a paper?)
tbf, GenAI is probably not what they consider a threat and can probably see through Altman's BS on AGI being *that* close and achievable

>>565700370
yes
>>
>>565739331
Xitter served him well and was worth everything he paid for. It was never meant as a money-making exercise; it's meant as a tool for shaping public and even global opinion. Same reason why Bezos bought WaPo. If you look at it strictly from the financials, it looks retarded. But the money isn't and was never the point in either case.

There are places where he does come off as a total incompetent, such as with xAI's management. He was off to a good start. Grok 1 and 2 were a joke, but no one will make a great, competitive model on their first attempt. There's a reason why Deepseek only got major attention with V3. Grok 3 was okay and comfortably settled xAI into the 2nd rate big corpo lab category (behind OpenAI, GDM, and Anthropic which were and are all 1st tier). Moreover he could boast of having real talent like Igor Babuschkin and Toby Pohlen on the team. Over the last 12 months, ALL the major original founders left the company. Total meltdown. Dunno what the fuck he did, but a loss on that scale didn't happen to any of the other big AI labs.
>>
>>565746365
The answer is obvious if you take a step back and realize five things.
1. GenAI is a bubble.
2. He, like many technocrats and elite, hitched himself to the wrong train.
3. That aforementioned train failed to manufacture consent and is draining support.
4. He's in the files.
5. He's a moron with more money than sense.
>>
I wonder when LLMs will be good enough to hook into an android prostitute.
>>
I've got a request for gayanon
a literal cupid figure with the curly hair heart arrows and bows and everything that's trying to help some guy find love (or at least a one night stand) of course every person he sets up with the guy ends up not working.
So the cupid figure at his wits end just decides to get fucked by the guy himself since nothing else worked. Have it on my desk by the end of the week, thanks.
>>
>>565736387
>>565737454
Now that I'm off work and settled down, I'll go into further detail. I remembered some more stuff anyway.
To give a bit of context as to when I started thinking this up, it was between the release of B2/W2 and X/Y. So it was before regional variations and stuff like that, and I wanted to keep fakemon to a bare minimum.
the setting/backstory/whatever:
>the region is populated by three types of people
>the descendants of unovan colonists who fled unova during that ancient war they allude to in the gen 5 games
>the natives/native american equivalents
>the new, non-native population from all around the world
>when the colonists first arrived, the land was nearly uninhabitable because of the severe weather (aka:the only fakemon I was going to have, another member of the weather genies based off of hurricanes/flooding that was going to be hidden in the post-game)
>the natives were barely able to survive, but it wasn't glamorous
>the two groups come together and pray for salvation from the bad weather
>Arceus creates a copy of the weather trio from Hoenn (they were going to be the shiny colors by default just to keep them separate) to keep the weather under control
>three places are built, made to harbor each orb that controls the legendaries
>groudon is in the middle of the state, in a tower that's inspired by the Bok Singing Tower
https://en.wikipedia.org/wiki/Bok_Tower_Gardens
>rayquaza is in the middle of the not!everglades, closely guarded by the natives
>kyogre is on a slightly moved Dry Tortugas
>the two groups of people live separately, the colonists to the north and the natives down south (though I ended up turning the state sideways to fit it better on the map) with good relations
1/2
>>
File: copied_file.jpg (89 KB, 351x879)
89 KB JPG
>>565764547
fast forward to the more modern era
>people flock to the region, business tycoons build up cities and a railway that mirrors Flagler's railroad irl
>the nouveau residents look down on the original colonists as yokels and hillbillies
enter: the antagonist
>a businessman obsessed with money and wealth, originally he was a conglomeration of Walt Disney and every money-obsessed billionaire out there
>buys up most of the region, cranks up prices for everything, typical generic rich person behavior
>learns about the three "orbs that control the weather" without knowing any of the specifics
>thinks to himself "man, it rains like every day here in the summer, wouldn't it be great if I could make it sunny on command?"
>finds kyogre's orb and steals it from its shrine
>no more rain, just groudon's sun
>horrible drought begins
Now to the game
>(You) are one of the unovan descendants, from the "north" part of the region,
>you live with your grandparents in a small town that gets smaller every year because most of the young people leave to go live in the city
>you don't necessarily want to leave, but your grandparents convince you to go because there's nothing for you if you stay
>you get your starter and leave
>you meet your rival, a very rich spoiled brat who mocks you for your clothes
>kick his ass and he gets even more mad, how dare your rural shitmon defeat his pedigreed pokemon
>a native woman meets you and talks about the history between the two groups of people
>informs you that without a "trainer license", you'll be unable to challenge the gyms or do much of anything
>the price for a "trainer license" is exorbitant
>gives you a pokedex and leaves, that will at least allow you to keep pokemon since if you have a pokedex you're considered a "researcher"
okay I lied I guess I need one more post
>>
>>565766003
>since you can't challenge the gyms, and not many people can afford it, there's an underground ring of unofficial gym leaders
>the gyms are all hidden and out of the way, some need passwords or a recommendation to battle
>the post game allows you to go through and fight all of the official gyms and the elite four
>along the way, you learn about the businessman and his plot, and how he's trying to find rayquaza's orb to have ultimate weather control
>the woman who gave you the pokedex is his right hand, but she eventually feels conflicted about basically throwing her entire heritage away for money
There were more bits and pieces of the plot, but they're a bit foggy. Basically the antagonist tries to get the orb, you stop him, and then you bring the fight to him to get kyogre's orb back and end the miserable as fuck drought
>>
That's IT! I'm banning "clinical". NO MORE.
>>
>>565767945
he said, with a clinical efficiency
>>
>>565767945
I'm super sick of [sentence], a [rest of sentence].
I despise ', a' now.
>>
>>565764547
>>565766003
>>565766608
Based setting, anon.
>>
>>
What authors do you use for the most detailed set? I'm talking gratuitous detail. I'm talking foreplay so gradual other takes chapters to build up to insertion. I'm talking about hot and heavy to the max.
>>
>>565771359
Thanks, man
>>
>>565772921
Whoop, whoop
That's the sound of the terrain
>>
>>565246184
Do fall into the trap, and post it here
>>
>>565345573
That sound a lot like deathroad to canada
>>
>>565777541
kek
>>
>>565773829
Bumping this. Also, it's sex not set.
>>
>>565369824
3.1 is a small degrade in story telling compare to 3. So it probably get worse. I used an absurdist humor prompt to test it. Good model weave a funny story anyway. Bad one have lower quality
>>
>>565746365
that's a good point, I was mostly thinking of him talking about technical details out of his ass, like the time he purged subroutines or something and ended up breaking 2FA login
>>
>>565438107
Ok xialong can be funny once in a while
>>
>>565449449
Because sam hates elon
>>
>>565515768
Don't expose mossad like that
>>
>>565525179
Damn it describe white women so well
>>
witches
https://files.catbox.moe/dq8p5x.png
https://files.catbox.moe/fshaip.png
https://files.catbox.moe/7h8qtu.png
>>
>>565773829
Can I add reference materials from my favorite author so the model can mimic it? About 660000 words
>>
>>565781362
Please do not make fun of my request. I am specifically asking for ATTG and such.
>>
https://files.catbox.moe/3dkbvp.png
>>
groaning and creaking loud faint soft grunting moan rumbling with a guttural screech



[Advertise on 4chan]

Delete Post: [File Only] Style:
[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.