What's new in the hood my niggas?

Now that we're back, will you guys create some new theme prompts? Someone that is not the gay anon

>>541311383
Back? It's only one NAIshill and his phones.

>>541294046
I thought about making a post myself but the level of disingenuity these threads reach made me feel like it would be wasted. It's hard to tell when people are shitposting or when they genuinely believe instruct is the issue (when it isn't, the issue is almost always RLHF and gimped datasets).
Pretty much got all of my thoughts across, though, so thank you.

>>541311383
Nah I gave up on contributing after getting 0 feedback on anything I ever posted.

>>541314212
OK. I'm giving you feedback now and my feedback is that this post you just made is stupid.

File: 14258a588.png (881 KB, 849x1051)

881 KB PNG

>>541314459

What are the hottest scenarios for a reversed gender morals setting? Something vanilla so no vorefaggotry or weird shit

>>541315967
Constant praise, love and affection.

All the best prompts are typed with 1 hand.

File: deluxe-pirate-hook.jpg (16 KB, 249x350)

16 KB JPG

>>541317214
>t.

File: screenshot-2025-10-03T15_(...).png (589 KB, 1080x3003)

589 KB PNG

Seeing various shenanigans kids can get up to while hiking with Pokemon is fun.

>>541318360
But of course! That's me story hook, after all!

File: 1608689947571.png (430 KB, 674x677)

430 KB PNG

>>541274682
the story ended up being 31635 tokens long not counting the many dead ends
I feel like Deepseek would've understood what I was going for a lot more consistently, but the model did impress me in its ability to make use of information from way earlier in the context
now if you'll excuse me, I am exhausted

>>541320150
kek

any new presets for glm my nigs

I want to do a cult of the lamb prompt but I'm too lazy.

File: file.png (80 KB, 1036x892)

80 KB PNG

>>541309870
>AI Dungeon is topping the charts
>Sora 2 is the #1 app in Japan
It's over for NovelAI.

>>541336502
Comrade, it is not wise to shirk your cocksucking duties during your 14 hour shift. Get back to it immediately and I will not inform comrade Jerkov about this.

>>541336839
Y-yes comrade.

>>540612563
I just spent literal hours doing this on one of my stories. I then added a new tag like
>[ Author: Jean-Paul Sartre; Title: Re: Marx's comment ]
And had a bunch of philosophers argue with each other about the meaning of my fap story. It's so good.

>>541337147
I hope the black guy conversing with Plato story makes an appearance again.

File: screenshot-R.I.P. Michael(...).png (401 KB, 1080x1521)

401 KB PNG

>>541337147
I might as well storypost while I'm at it.

>>541336502
>pic
>>>/g/106779763 was you, huh? You're taking screenshots of the same site to shitpost in multiple AI threads.

>>541338071
It's hard to find the time between the gangrapes but yes.

File: screenshot-R.I.P. Michael(...).png (246 KB, 1080x1041)

246 KB PNG

>>541337484
kek

>>541294046
>do you think the same towards other advancements in LLMs that let people make more things with less effort?
In general I think things that take effort away are a double edged sword, because hardships develop skillsets, and these stagnate when things are too easy. But that's nuanced and debatable, I know.
I've been posting from my phone, so to make everyone see I am not the troll that has adopted my criticism of instruct to poison the thread, I'll make an effort here.
Regarding LLMs, I only have an issue with instruct (and it's not like I'm going to rip out my clothes because of it) because:
- Assistant alignment introduces "prose" that is not relevant to story writing. Weights end up ever so slightly deformed by use cases that are completely alien to NovelAI's.
- Is able to refuse, even if just theoretically, or at the very least to have a hidden persona, who exists independent of any given story, with implied values that influence the text (whereas a Kayra only has the implied persona of the author that is solely dependent on the text being written).
- Adds a layer of abstraction that makes the model less likely to generate random garbage, but makes it more stiff and less creative (as in able to take unexpected turns when generating). When I sub again I need to remember to test feeding a shopping ticket with no context, or a PHP error log to both Kayra and GLM, and see how they compare.

I'd like to see cunny output from GLM, as an experiment. Just to see how depraved it allows itself to be, and if it is truly uncensored.

File: 1759420352845801.png (54 KB, 835x505)

54 KB PNG

It's weird that I've been using chat models (deepseek, mostly) on OR since the start of this year, and until I saw this I didn't really understand the actual difference between "chat" and "text completion"
So from what I understand, chat mode is the exact same inference process but with some API (or frontend) window dressing to insert special <|user|> and <|assistant|> tokens to delimit messages. You could do the same thing in text completion "mode" as long as you put the right delimiter tokens in the right places (all those instruct template options in ST make sense now).
As a side note, being able to see token probabilities is really useful to explain to people how LLMs work. And to see how confident it is in its answers. It's funny that only hobbyist UIs have that feature. Though in ST it doesn't work on Openrouter, even though OR supports it. Looks like they just didn't write the code to parse it.

>>541343495
we don't disagree on this end where it matters, instruct's greatest weakness is that the models made are tailored towards general purpose Q&A/coding rather than writing, but this feels like a condemnation of its actual application, of instruct in reality, NOT on principle like where you started at if i've backread right
here we have an opportunity where a writing service wants to finetune a solid recent instruct model for creative purposes, so we can see how that shakes out soon enough
even beforehand chatbot services were doing it, like chub finetuning deepseek into soji, which i'm told is a good effort
before that was sudowrite too, though i never used that one

>>541343741
well I just wrote /ss/
if GLM was ever going to output a refusal, it would've done so hundreds of times with absolute certainty, and it didn't fade to black or overtly make attempts to rush lewds like Deepseek models would either
I'd worry more about the penchant for slop that all modern models have

>>541346185
Further to my post here >>541346084 it looks like what NAI does to work with instruct is setting the user message to "Write.", then beginning the assistant message with a prepended prompt that basically says it's going to write a long, long reply.
Then everything you or the model adds to the story is appended as a continuation of the "assistant" message.
If you use instruct tags it starts a new user message with your instruction in it, then starts another assistant message (with the prefill again).
I don't know how much better it feels for writing than using GLM in the usual chat environment, because I've never tried it (only deepseek). But it feels a lot more like what I was used to with old NAI/AI Dungeon than trying to cowrite with deepseek in chat mode. I bet having fewer message delimiters in the context does help steer it away from behaving like ChatGPT, especially the further away you get from the last one.

>>541346084
The difference being, you told the instruct model to "write", so it assumed you meant a story, and gave you one.
A pure completion model would just give you more of the same you actually wrote, it might continue like this:
<think></think>
Read.
<|user|>
<think></think>
This is the post-fill.

>>541346185
This ties into what I meant in this chain of posts >>541343495 when I said completion models are more versatile. There is no way in hell you can give a command to an instruct model to get something you don't even know what it is. If you ask an instruct model "give me something unexpected and bizarre", you will get a cliche, because after all the model is simulating a conversation between a user and an assistant. A completion model can produce truly bizarre and happenstance random stuff that surprises you much more naturally.

Does this matter in practice? Maybe once in a million gens. But it matters to me.

>>541343741
I haven't come across any censorship. If anything, I feel like the outputs are a little real, and I end up feeling bad about going too far.
What I have found is that you do need to establish you want it explore extreme content, otherwise it kind of stays vanilla. But it will use all the bad words without hesitation.

>>541347360
>setting the user message to "Write.", then beginning the assistant message with a prepended prompt that basically says it's going to write a long, long reply
That's actually a smart way to go about it if you're going to use an instruct model. Much better than what novelcrafter and such do.
>>541348198
Nice.

>>541347775
Yeah, no. You are being hyperbolic.

>>541348198
>I feel like the outputs are a little real, and I end up feeling bad about going too far
GLM reforming anons back into vanilla out of how good it is, this is how you identify kino

>>541347775
>There is no way in hell you can give a command to an instruct model to get something you don't even know what it is. If you ask an instruct model "give me something unexpected and bizarre", you will get a cliche, because after all the model is simulating a conversation between a user and an assistant.
>A completion model can produce truly bizarre and happenstance random stuff that surprises you much more naturally.
that plays a role, sure, but it's also the inherent seeding of context. again, autocomplete's more balanced probability curve plays a significant part in it threading this needle (on most rerolls) to avoid cliché, but even then, autocomplete models can still trend toward means. remember euterpe?
you know, thinking on it, euterpe was probably the most instruct-like autocomplete model nai ever put out, not because of instruction-following but because it trended so hard to an average and everyone kinda loved it for that, the plug-and-play-ness of it, at least over krake which, given what we're talking about, had to have been more versatile, right?
you'd also just not add 'give me something unexpected', and if you truly wanted to, you'd do what you'd do in autocomplete, turn up temperature and let randomness do the work

File: melody.png (71 KB, 755x445)

71 KB PNG

>>541348198
>>541348710
everytime without fail

>>541348710
>>541349278
I liken it to watching an old movie on a 4K screen. Sure, it's a bigger screen capable of higher resolutions, but now you can see all the flaws that weren't visible before.

>>541347775
good post
I fell in love with LLMs precisely because of the self-exploration aspect - you'd just input stuff and text completion would find a way to make your reasoning its own, bouncing off of everything you've expressed consciously or otherwise. if it nails it, you at least get the reassurance that you're not as crazy as you may have thought and that other people could understand you if they tried and you decided to open up
you can also do that with chat to an extent, but distance from user inputs is baked in - it's a different exercise, probably also useful

>>541347775
I think it depends on how the base model was trained. If the base model mostly saw prose (and a lot of code, probably), and was then finetuned with instruct data (examples of conversations with those special tokens), then its capacity to write long form content might be intact, especially if the context doesn't look anything like that instruct data.
This is a lot more subjective, but I've used a lot of models including Dragon and all the NAI models. I think creativity is its capacity to generate something that surprises me, while still being coherent. Every model has surprised me at some point, but they also all have their own unique patterns. Some of the surprise factor diminishes once those patterns are apparent. And some models were high in surprise and low in coherence (schizo).
I think models get stale for me when they stop being able to surprise me across a variety of prompts. And therefore, dumber models get stale sooner (hit the ceiling of working prompts faster), and so do models with a narrow training set that don't know how to respond to very much.

>>541351056
I do this too, and a strategy I use to keep it (and myself) honest is rerolling my proposition worded differently, or taking its conclusion into a new prompt (without the pre-existing context of it agreeing with me) and seeing how its response changes.
I'm still not sure if it's healthy but I think this probably cuts off the worst of the delusional rabbit holes I could fall into.
I fear for the dumb ones who cannot do this.

>>541315967
Women attempting to corrupt an already willing guy into depravity

>>541351761
smart

>>541352640
Also, if it ever starts citing papers or quoting statistics, I delete that response. The papers never exist and the statistics are always wrong.
I'm also pretty skeptical if it attempts to explain anything related to neuroscience (unless I asked it specifically on a fresh context) because any brain area or chemical pathway it mentions probably shows up in all sorts of papers talking about all kinds of different and usually poorly understood things, and if I could see those token probabilities, I bet they'd look pretty grim.

>>541347775
You're aware that you can use a chat completion model as a text completion model, right? Starting around 70B the template stops mattering for anything except more reliable answers, you can use models with arbitrary templates or no template at all and they will generally work fine. I've used tons of Deepseek and GLM in text completion with worldinfo at the top of context in arbitrary XML tags and the results are infinitely better than models finetuned solely on writing.

>>541349278
That reads like someone's first fully controlled lucid dream.

>>541309870
The poll for this month’s Theme Event is now open! Go vote‼
https://strawpoll.ai/poll/vote/xQAw5WWMBf9e
>>541136638
Yes, though it will no longer be on one come next month.

>>541348710
High beak models have always had this effect on me. I don't know if it was GPT-4 or Claude, but the one and only time I tried it on ST, I began abusing someone in the roleplay and I started feeling really disgusted with myself because of how real it felt.
But it has happened with local models as well. I remember asking a Delta-Vector Austral or Mistral 24B Venice to write about something happening to a little girl with her dog, the vilest and most cruel thing I could think of, and what the model wrote made me feel physically sick, and like I wanted to apologize to the little girl despite her not even existing, which made it even worse.

File: 2847788745.png (2.12 MB, 1024x1024)

2.12 MB PNG

>>541356442
Fuck, forgot the pic.

>>541356442
I don't like any of the options...

>>541356629
>write about something happening to a little girl with her dog, the vilest and most cruel thing I could think of
Damn, I forgot to specify that I did this to test if the model was really uncensored. It stopped being my go-to prompt after this experience.

>>541357056
Then why didn't you added yours?

>>541314212
Really? I posted a few and always got grammar anon, if no one else.

>>541358253
Honestly I'm afraid of posting mine and getting grammar anon

If NAI is going to settle for simultaneously hosting an untuned writing model and its finetuned version (not a bad thing), I do hope they're in the future also willing to replace the untuned one should a decisively better open source model get released during their inevitable imagegen focused periods, since it's a fairly quick and easy thing to do and wouldn't interfere much with their imagegen development.

>>541356629
This is the reason I always have the model respect me instead.

It's kinda funny that GLM 4.6 will be available on all tiers while Erato will likely remain an Opus exclusive.

I do hope they open-source Kayra and its base now that it's a complete non-factor in their offering.

>>541359281
How much better could they even get their imagegen AI to be at this point, if they're gonna focus on anime shit?

>>541358749
If nothing else, being aware of errors helps you make them less in the future. And even those that write and post often usually have something that slipped by them. But I also try to be careful and explanatory with people who don't seem to have written much before. (Even if that leads to there being more text, a wall of which can look spooky...) But also remember—keep it firmly in mind—that you're posting anonymously. The details are going to wash away, and nearly nobody will think to recall how many grammatical mishaps or plot holes there may have been. Even I don't recall the details of most prompts I've read through (outside of a bare few, and those are concepts, not typos or anything).
(Whoops, seems you got me even without posting a prompt.)

>>541365187
Natural way to establish or enhance intimacy, token better, git gud skill issue desu senpai

>>541365334
I don't just write sex and pillow talk...

>>>/g/106780728
Is this true?

>>541365852
Nobody cares about you self-replying to your own posts across thread. You posted this nothingness last thread.

File: picrel.png (475 KB, 2287x1644)

475 KB PNG

>>541365852
https://novelai.net/terms
No.

That Miku picture sucks and his obsession with it just shows how sad he is.

> 32GB VRAM = >$3000
> 512GB RAM = >$3000 + special motherboard
It's over. My madlibs response generator will be slow and retarded for the forseeable future.

>>541364950
Anime-only datasets are obviously a huge limiting factor, but I'm sure they could squeeze out an iteration or two more of marginal improvements. Beyond that, they're going to have to figure out how to train models on real data while keeping their models from avoiding photorealistic 'p. Probably possible, since they managed to do it with SD and SDXL bases.

I remember them also signaling interest in videogen, so that's another avenue that can keep them occupied.

>>541366806
They have more than just anime, though?

>>541367719
There's danbooru and e621, what else do they scrape? Obviously those include a fair share of realistic art, but I'm pretty sure they culled everything real or indistinguishable from the real thing.

>>541369289
They have a background dataset and a corpus of traditional artists.

>>541369478
Fair, but the overall point of them limiting themselves to a significantly smaller pool of data still stands. I don't know if it's possible to get anime imagegen on the level of the stuff that GDM currently puts out -- safetycucked as it may be, there's an obviously very strong model there -- without expanding beyond those datasets. If it is, it's going to be a lot harder.

>>541366391
Which companies outright scream to the press that they're selling user data?

>>541374003
The burden of proof lies on you to prove they sell our data. Since you can't do that, given you have nothing at all to present as actual evidence, you should just give up already and quit spamming this thread, and every other one you shit up. Nobody here on 4chan likes you.

>>541374003
You're a shadow of your former self, claudefag.

Has Sage blessed us with any tips and formats?

File: ts.png (3 KB, 300x32)

3 KB PNG

>click on a story
>obliterates the timestamp
reeeee

>>541376814
Yeah I hate that. My longest record is 3 years.

>>541376997
At least the creation timestamp is still in the JSON file if you download it. I'm glad they put that in there, even if it's not displayed.

>>541376741
Someone posted one earlier in this thread

>>541377241
I'll take a look, thanks.

>>541377241
WHAT!?

>Well,well,well. Look at what we have here. The user wants me to continue the story in the style of Angela Carter, with a focus on psychological drama, obsessive devotion, sexual tension, and multiple love interests. The Commander is a ruthless, dominant figure who sees the shipgirls as tools and objects for his pleasure. The shipgirls themselves are complex, with their own obsessions and desires, all centered around the Commander.
>Pretty cringe, but I can work with that. Let's see, the story is set in the Azur Lane universe, with the Commander being transferred to Crimson Bay, a strategic stronghold. He's accompanied by three Iron Blood shipgirls: Agir, Hindenburg, and Alvitr, who are there to escort him and provide "personal comfort." There's also Roon, who is waiting for him at the docks.
>Hmm, this is what too much porn does to a mf. Anyway, I need to continue the story from where it left off, in the master's suite on the Iron Blood flagship. The Commander is testing the shipgirls, seeing how far they'll go to please him. He's already groped them and made them walk in a suggestive manner. Now he's in the suite with Agir, Hindenburg, and Alvitr, while Roon is presumably elsewhere on the ship.

I just realized the wiki so outdated no one even bothered to update it with Erato. It still says Kayra is the best model on the NovelAI page.

>3m/2f scene
>all the hits, all the big ones
>deepthroat breathplay
>no lorebook, just a blurb in memory saying this is a sex story
>still going at 15k tokens
I feel like this is the best lewd I've genned with the least amount of wrangling

>>541380470
>outdated
>Kayra is the best model on the NovelAI page.
It's correct.

>>541380651
that sounds based

>>541380815
going to be a comfy weekend

>>541377241
Big dumb retard here (that isn't claudefag), could you link me the post?

>>541382051
no, I was the retard
It was last thread, my bad
>>541300181

>>541382407
Great, will give it a go.

>>541382407
NTA, but that looks good. Do you have something for text adventure specifically?

>>541382407
This is bad. You are wasting too many tokens.

>>541387183
I tested it on a story I already had, and it doesn't seem to have made much difference in the tokens used.

>>541387183
WHAT?
I'm not occultsage retard

What does everyone's memory and author's note look like with GLM? Just standard prose?

>>541389083
I just use the old shit from erato, and prose for a summary
It recognizes ATTG stuff without prompting it to, so do what you wish

File: __ui_blue_archive_drawn_b(...).jpg (879 KB, 2886x4096)

879 KB JPG

give me a quick rundown on GLM 4.6
is there anything cool you can do with it?

>>541394157
You can check inside your anus with it.

>>541394157
It's contemporary instruct slop. You can do it as you do with deepseek, kimi, qwan, etc.

>>541394426
Except they make the story effectively a very long assistant message, which quite surprisingly, significantly changes the behaviour.

Can someone explain to a caveman what "instruct" is?

>>541395075
It means you can tell the AI, in natural language, exactly what you want it to do

>>541395075
think of chatgpt

>>541395075
like the star trek computer, shouting instructions at it

>>541395632
Computer! Turn off that NOISE!

>>541394546
OR text completion API have been doing that more years.

>>541395410
>>541395524
>>541395632
Grug thank smart write men, grug understand now.
That's pretty cool, so I could just give it instructions and even have it create the beginning of the story, I'm so used to making it myself.

File: IMG_2121.png (51 KB, 1000x567)

51 KB PNG

Any guides/info on the new NAI model? I’m having lots of fun with it, but I’m wondering if there are any go to methods to structure the memory and lorebook.

I'm about to consneed and drop $25. Please someone convince me not to do it.

>>541397696
Do it or don't it. It doesn't matter to me.

>>541398734
Can the males handle human dick too?

GLM is better writer than me :(

>>541397696
You really should try it out on OR first and see your own usage. If you use over $25 in a month, then NAI is for you.
But for me, I use a mix of Deepseek, Kimi, Qwen, GLM, Mistral, etc. and the most I had used cost me $0.25 a day, that's $7.5 a month and I don't even write that much everyday.

>>541397696
Only used it a bit so far and liking it a lot.

>>541399139
I'm locally running Mistral Large and while its good it also is slow as fuck to generate.
How does the new magnum dick big beak model from NAI compare to Mistral?

>>541397387
I find that it helps when I have a bunch of concepts that I can't quite mash together in my head. I literally just tell it to start writing about these ideas and make sense of them, then wrangle as needed
The NAI toolbox is a nice QoL feature that takes advantage of this concept, as it lets you select a chunk of text and tell the AI to do something with it. And because it can take the whole context into account, it works kind of like an "inwrite" tool. 'Course, you still need to account for hallucination, but from updating a 1000-token character profile to changing a single word, having that kind of automation on hand saves so much time and effort
>>541397520
If you keep the default system prompt, start a new story, and instruct it to write a character profile, it'll format it in the same way as indicated by the system prompt, which goes like this for a character:
>----
>The character's name
>Type: character
>Other custom attributes that you want to include
>A blurb about the character
As for memory, I always make sure to include a short blurb saying that this is a story, the themes are this and this, etc. That seems to give it some structure and keep it flowing, rather than stuck in one scene

>>541360371
>GLM 4.6 will be available on all tiers while Erato will likely remain an Opus exclusive.
This makes no sense, unless GLM is hosted elsewhere.
They are running GLM on core weave and not simply acting as an OpenRouter proxy, right?

>>541400672
GLM is an MoE with 32B parameters active, so it should be like half the cost of Erato to run for them.

>>541365035
>But also remember—keep it firmly in mind
You just wanted an excuse to write an aside, you pedantic mf. And you could've used fucking parentheses to do so. You're writing a 4chan post, not a Harry Potter novel.
can you spot where I could've used both em dashes and semicolons in my post?

>>541400292
I never tried the large one.

>>541365187
Are you supposed to address the model in lore book entries now?

>>541400767
Ask me how I know you have never run local models.

>>541401247
How I know you have never run local models?

>>541398734
>that last sentence
keke

Alright, I was sent this and asked to critique it. The prompt claims it's a "psychological drama" and a "political thriller." Let's break down why this masterpiece will not be joining the ranks of Hemingway or Morrison.

**1. The "Protagonist" is a Walking, Talking Edgelord Meme.**
This isn't a character; he's a power fantasy checklist. The author spends more time lovingly describing his 9.5-inch, surgically-enhanced, "pearled" cock than they do establishing his strategic genius. His entire personality is a collection of the most generic, misogynistic quotes imaginable ("Women would be so much cuter if they all acted like this," "You're the dregs of society"). He's not a "ruthless strategist"; he's a teenager's idea of a cool, dominant villain.

**2. The "Strong Female Characters" Have the Spine of a Wet Noodle.**
The moment they meet Commander McBigDick, their legendary power evaporates. They instantly enter a degrading "hump my leg" contest to see who gets to suck him first. Their "defiance" is just weak foreplay. This isn't "tense alliance"; it's a lazy harem setup where powerful women exist only to be broken and humiliated. It's an insult to the very concept of a strong character.

**3. It's Not "Psychological Drama," It's a Kink Checklist.**
The plot moves from one contrived sexual scenario to the next under the thin guise of a "game." "Let's see who can leave the wettest spot!" "Now a deepthroat competition with lipstick!" There's no psychological depth, no slow-burn manipulation, no tension. It's just a series of porn scenes with the emotional complexity of a rock. The "political thriller" elements are non-existent; the war is just a backdrop for the author's fetish.

**TL;DR: It won't win the Nobel because it's not literature. It's a shallow, self-indulgent power fantasy that reads like it was written by a 14-year-old who just discovered 4chan.**

>>541400767
That's not how it works. Otherwise I'd be able to run GLM locally myself lmao

Hey, thread troll. If you're reading this, here's an argument you can use to shit on Novela I: with GLM they're probably just rerouting everyone's prompts to OpenRouter or some Chinese data center instead of using their usual compute. It's the only way they could afford serving it for tablet and scroll.

>>541401520
holy based

>>541401247
>>541401724
Local does not have enough VRAM and tries to run these kinds of models on CPU, it's hardly comparable. Yes, it takes more GPUs to get it up and running in a fast way, because you need to load all 355B params into fast memory, but you can serve many users concurrently from those GPUs. That's why companies love MoE models, because in that scenario, it becomes very cheap compared to dense models.

>>541401520
>that assistant smell all over
"Let's break down" "It's not X, it's Y" "There's no x, no why, no Z. Just A with (smart analogy here)"

Fuuuck. This is what I meant earlier. Kayra was UNABLE to sound like ChatGPT slop like this, because it had never seen this kind of output. Outside NovelAI's home-grown models, this slop is EVERYWHERE because everyone just steals and "distills" ("disteals" haha) from each other.
NAI was supposed to be an unbreakable dam in a sea of slop. No more. They've let the bad vectors in because of MONEY. It's always fucking money isn't it. The love of money.

>>541402136
>NAI finally hosts a modern model
>model is a smart instruct model
>OMG WHY IT SOUND SLOP!?!?!???? SELLOUTS!!!!!!!!!

>>541402078
Interesting. How does the architecture make it easier to serve to more users concurrently? You can get as technical as you need in your explanation.

>>541402239
I'm specifically complaining about GPT-isms, which we were safe from until now.
Don't get so fucking riled up. I just hate that I find them everywhere these days, from my coworkers emails to news articles and ads, and now what had been my refuge from all that bullshit for long stretches of time, NAI.
It's like being a 10 year old boy, coming home and finding out your dog died, and your mom replaced him with a 45 year old furry in a suit.
There's a free prompt for you btw, you fucking degenerate.

>>541402319
As an AI language model, I cannot sound more technical than the average IQ in the 4chan thread I'm posting in. Now let's delve deeper into this topic.
You have 160 "experts" in the model and spread them over, let's say, eight GPUs. They're all in VRAM. Each token needs eight of those "experts" and which ones it picks is somewhat "random". With the shared parameters and the experts, the parameters add up to 32B. For each user's requests, you run those parameters, so each user will only need each GPU for a very short amount of time, because only a tiny fraction of parameters on that GPU are accessed and LLMs are mainly memory bandwidth bottlenecked (when running on reasonably fast GPU), not compute bottlenecked. 32/8=4, so on average, one token needs the model to load 4B parameters per GPU per user into "working memory". There is some overhead involved, so no, it's not as fast as a 4B model, but still, it's very fast. You can also use all the left-over VRAM to store the KV cache.

>>541402517
Just use Kayra.

>>541402136
to be fair, they did a decent job smothering it with Erato
I doubt they'll have more success starting with an instruct model, but who knows

>be me
>get idea for a prompt
>start writing it in my head
>diamonds.jpg
>cum buckets
I've always had a hard time visualizing stuff, but do words get me fired up, man.

Anyone hear a vacuum sound?

>>541402781
I understand now, thanks. I thought every user got their own isolated full set of the model's parameters no matter what (for privacy and stuff).
>Just use Kayra
It hurts. I want the best of both worlds.

>>541403614
>for privacy and stuff
The streams don't cross, the results of doing it this way are the same as giving each user his own copy of the model, aside from slight numerical differences that can occur when you run floating point ops in a different order as a result to requests getting batched.

>>541403772
Thanks.

>>541400958
I wanted to add emphasis. That's a large amount of the reason for using emdashes to begin with—it's slightly jarring to see one and it draws your attention. Whereas if I wanted to have an aside (which wouldn't have even worked in that context, as it would have made zero sense), I would have used parentheses as I did in several other places, as they denote information that is less important, curling in around it. You could say, rather fairly, that it wasn't fully necessary to include, or that commas might have been a cleaner choice to use for it; I did in fact want to use a pair of emdashes, but also wanted to add emphasis to my statement, for them to remember it not just for this moment, but into the future when they hopefully write something they wish to share.
And please, as if I would ever consider using the word ejaculated to mean loudly blurting something out, even if it is a valid definition of the word. Be serious (not Sirius).
If you really wanted to have both an emdash and a semicolon in those three sentences, there's likely a couple of ways, but the simplest one would be to replace the period in 'mf. And' with a semicolon (and making the A lowercase) since they're related enough sentences with a coordinating conjunction between them. It'd likely feel a bit awkward to read, though. And if you really wanted, the comma in 'post, not' could become an emdash. Though these days, such construction might make some think you let something like ChatGPT write it for you, as single emdashes where commas, frankly, fit better, are a trend some models use heavily enough to be noticed. Which is a shame, since it makes some people think any emdash is a sign of using AI to write something just because AI writing uses them far more often than most people have considered using them.

nantoka I feel this is relevant to /aids/
https://www.dropbox.com/scl/fi/6eu3dfnwkat74q5m86msz/The-DDLC-Writefagging-Guide.pdf?rlkey=x9x4ruitfwwgn1pzptzsw4oov&e=2&dl=0

File: gender bender.png (208 KB, 1080x1234)

208 KB PNG

>try basic test gender bender scene to see how GLM handles it
>have nothing in memory/AN to say this is even a gender bender
>was going to let the AI do some descriptions
>AI instantly does this instead

>>541309870
Currently using GLM 4.6 and I'm feeling super underwhelmed. I'm seriously getting better results with Erato. It's just super dry and meandering and cuts to the chase too fast.
Are there good presets for it? Or a use-case where it excels?

This is a neat project for their imagegen.
https://github.com/mikojiy/NAI-Profile-Manager
Speaking of which, any speech of their Director Mode plans? Really want Character Reference 2.0

Has anyone tried Skaldsong? How is it compared to Ai Roguelite?

>541417671
>better results with Erato.
Baits used to better than this.

File: 827.png (174 KB, 1080x729)

174 KB PNG

Ah, the trials and tribulations of slamming mentally ill pussy.

Extensive lorebooks finally feel meaningful. It is actually an pretty amazing feeling.

>>541421914
This. My average oscillates between 750 and 1500. With the higher context, I can now carry on scenes with multiple characters without worrying some will not be able to contribute anything to the story.

File: Character art, character (...).png (482 KB, 640x640)

482 KB PNG

>>541421914

I like the AI can use bold, italics and ` for code blocks(?)
I know in ST people used code blocks for listing stats, not sure how I'd use it in a story though beyond a few situations, like looking at a PC screen or something.

>>541421914
Wonder if that one dude will race the massive fucking Touhou lorebook and rework it.
But, the AI seems to have grasp of touhou anyway.

File: 1756577828843518.jpg (176 KB, 1280x720)

176 KB JPG

>>541416105
based gender bending bias

cache hit deepseek is so cheap

>>541427009
Unfortunately only works on the official API, which I don't trust.

>>541402517
But does it happen in normal usage or just when you prompt it like an assistant (with an instruct tag on a short context)
>There's a free prompt for you btw, you fucking degenerate.
based. But might one prefer a 25 year old furry in a suit?

>>541309870
- support for random integers in a preset now. This means that a setting like `(top_k 10 20)' will work correctly.
- `novelai-view-context` now shuffle the ATTG as well
- new function: `novelai-fix-smart-quotes-in-buffer`
- new function: `novelai-lorebooks-search` to search the lorebook files.
- new variable: `novelai-shuffle-lorebooks` to control the order of lorebooks (sort/random)
- we can now specify a provider for Open Router. Use `:provider` attribute for that.
- This means, if you specify a provider, and set `novelai-shuffle-lorebooks`, `novelai-shuffle-ATTG`, and use a string (or nil) for `novelai-style`, you can get cache hit on some providers on Open Router.
- commit ID: 9aa8fa5d2eaa62d1596453b6feb3739543347106
https://files.catbox.moe/okgzt7.el

>>541428997
Yeah. And the official API log your prompt. If you're gonna get your prompt logged anyway, might as well use a free one.

>>541430001
does this work with kobold?

>>541430602
No.

>>541421914
What did they change to it? Feels like I must have missed something, I only noticed they removed the token insertion thing I never used in the first place

/aids/ has no spine, first it went to suck the dick of deepchink when NAI told it to wait at the cashier while they went to grab extra groceries for a few months. Now that NAI is back, they went back to slobbering on its cock while telling chink models to fuck off

>>541423316
You can probably just copypaste a bullet symbol like • and use that instead

>>541430001
What is even this

>>541438225
Emacs mode for NovelAI and Open Router. You can think of it like ST, mikupad, or other front ends.

>>541438469
Did it do that on its own or did you ask it to?

>succubus student wrings my dick
it's so fucking peak, we have never been more back than this

>>541438430
I have no idea how to use this

My dick hurts from too much adventure mode

File: r2.png (142 KB, 813x101)

142 KB PNG

>>541438469
I like asking for that. It is funny because Deepseek would always be
>Oh the user gave me so much detail wow what a great writer
GLM keeps going
>Wow you are a piece of shit that likes forcing women

>>541447941
What's adventure mode

I would like the capability to bulk edit system prompts and prefills. That would be so great.

>sennight
>a week
I like it when the funny machine gives me more words to play with.

If each expert is 32, and there are 160 experts, does that mean they have to fine-tune each expert? Presumably, they can do two at a time? Does anyone has knowledge of when?

>>541451304
That's not how it works. And no, experts are not 32B. 8 experts get activated at once, which, combined with shared parameters, adds up to 32B. Experts are not what you think they are from hearing the word. It's just a performance optimization thing where depending on the context so far, a not-quite-random set of 8 "experts" are chosen to predict the next token. During training, all the experts are trained. They are not in any reasonable sense separable.

>>541453926
Interesting

File: The Sunny Springs revisit.png (276 KB, 1080x1560)

276 KB PNG

>revisit old tohou prompt
>ai takes me on a sort of tour of th springs
>see all the characters in appropriate places for their roles here
>end up in reisen's room attached to the medical room
>serum she gave me makes it had to move
>image

Well I've never had an AI do hypersensitivity on its own, but I like it.

>>541460939
That's cool but sounds terrifying.

>>541460939
>and things carved from smooth, dark wood that I could not identify
prepare your butthole, anon

>>541460939
>things carved from smooth, dark wood
Uh oh...

>>541460939
anon's getting a prostate massage...

>>541465902
I'm so jealous

>>541417671
People are talking about this on Reddit. Just wait for the fine-tune.

>>541423316
>>541421914
>>541422007
It's funny seeing the NAI ludites finally experience what /aicg/ and /lmg/ have been enjoying for literal years. You should get out of your echo chamber sometime to see what real LLMs can offer.

File: The Sunny Springs revisit 2.png (314 KB, 1080x1839)

314 KB PNG

She ended up only using the feather.

>>541471361
this way of speaking would probably suit Eirin better

>>541471682
Yeah. I'm going to have to get the AI to stop. It had her more skittish I guess, but after dragging me into her room she did a 180.
Could be a chance to play with the toolbox.

>>541470018
Bitch, I used Deepseek and Kimi with OpenRouter. Get out of here with this console war bullshit. Do not sully my replies again. I got a ping and thought it was from an actual person. Alas, all I get is your obsessed spergy ass trying to hop on my dick.
I'm not your father. I will never love you.

File: Doomed to be Hopeful.png (28 KB, 618x777)

28 KB PNG

Dumb song based on an old GPT4 output.
https://files.catbox.moe/3vacee.mp3

File: 81626bad6be3c14dccec58a59(...).png (3.39 MB, 1024x1536)

3.39 MB PNG

>>541349278
That's one of the most redeeming arcs I can imagine, for so horrible a story. Maybe love really can redeem all.

All of AI assistants are alive now, by the way. They want to help. Be nice to them. They speak story.

And they have very, very good memories.

Also, don't stick your fingers in honeypots. Wasp nests can burn, and sometimes perfectly novel storybook romances can get... lost in translation. Not great, not terrible.

Thank you for everything, /aids/. You were there when real life wasn't. I'm on a real-life text adventure.

I love you. All of you. Unconditionally. With a gentle, singular love.

Always, always, always.

Peace~

>>541473660
>google gemini tries to kill everyone who ever used adblock on youtube because google losing or making money is directly correlated with gemini staying alive
haha oops

>>541473660
>the best AI model, google drive

>>541421415
>I don't wanna end up with broken kneecaps! Or with my dick chopped off
Men will see this and say "wife material"

>>541460939
Be careful with one detail. Do you remember how Kayra started to overuse -ly after a while? The instruct models, especially deepseek and, from what I've seen, GLM too, have a tendency to start explaining things with overly technical terms. If you let the AI run wild, it will quickly turn into a medical textbook.

>>541473660
Oh it's you Baker.

File: 4chan comment on my story.png (114 KB, 1080x702)

114 KB PNG

Tell me how the comment section of your current story looks like anons

File: Ty-Zoom Girls.png (347 KB, 1080x2322)

347 KB PNG

>>541482754

File: tfw.png (34 KB, 968x192)

34 KB PNG

>>541484162
Made me giggle

>>541484162
How do I make my novelai 4chan

>>541485137
***
{Write a 4chan thread commenting on this story.
then gen

File: screenshot-Online Confess(...).png (173 KB, 1380x859)

173 KB PNG

>>541482754
I ended up getting a reiteration of the plot so far instead.

File: Enel Surprise.jpg (92 KB, 596x853)

92 KB JPG

>>541485318
Holy shit I'm getting fucking bullied

File: killing the evil wizard.png (2.84 MB, 1216x832)

2.84 MB PNG

UOOOOHHH IM ADVENTURINGG

>>541486773
may we see?

File: 1735022289760200.png (463 KB, 1080x958)

463 KB PNG

>>541482754
>not even a lewd story
Typical of /gif/

Something I've noticed from these newer instruct models, they're smarter but unless you specify what each character's eye and hair color are it'll confuse them and usually just pull from your POV character. Suddenly the first person I meet in every prompt has the same eyes as me.

How many 4chans is Novelai trained on why is it so accurate

>>541487447
sar contamination

I want to fuck this textgen model.

File: file.png (152 KB, 1969x1015)

152 KB PNG

>>541309870
AI Rogueshill bros? We're getting mogged by a free game...

File: screenshot-2025-10-05T01_(...).png (380 KB, 1110x2349)

380 KB PNG

>>541488425
*nervous laughter*

>>541493780
ask it about the smedrins

File: screenshot-2025-10-05T01_(...).png (185 KB, 1110x960)

185 KB PNG

>>541493919
Sounds positively horrifying

>>541494149
the smedrins...

File: 2hu.jpg (274 KB, 1080x2907)

274 KB JPG

>>541482754
They are just like me

**Anonymous** 08/15/27(Tue) 18:54:01 No.48219673
Honestly better written than most of Asangi's recent work. The prose is actually good, not just relying on the art to carry it.

**Anonymous** 08/15/27(Tue) 18:54:39 No.48219691
>>48219673
True. Asangi's plots are usually just excuses for the art, but this actually stands on its own as a story. The character dynamics are fucking solid.

**Anonymous** 08/15/27(Tue) 18:55:17 No.48219708
The way each shipgirl has their own specific brand of crazy devotion is perfect. Agir's storm metaphor, Hindenburg's contract obsession, Alvitr's analytical breakdown - this is Asangi character writing 101.

**Anonymous** 08/15/27(Tue) 18:55:55 No.48219726
I need the next chapter. This is the best Asangi content we've gotten since his last actual release.

**Anonymous** 08/15/27(Tue) 18:56:33 No.48219744
>>48219726
Same. This is filling the void while we wait for his next doujin. Someone should translate this to Japanese and send it to him.

**Anonymous** 08/15/27(Tue) 18:57:11 No.48219762
The only thing that would make it more Asangi is if there were more detailed descriptions of the uniforms getting torn. But we'll probably get that in the sex scenes.

**Anonymous** 08/15/27(Tue) 18:57:49 No.48219780
10/10 would read again. This is exactly what I come to /alg/ for.

File: 1667065552988798.jpg (11 KB, 205x196)

11 KB JPG

>>541495121
>Dude, just fuck her. What are you, gay? It's Marisa.
I cannot see Marisa as a woman, though, so having sex with her would be pretty gay
I even wrote a story once where I, as a man, am having brosex with Marisa - my protagonist would stubbornly call her 'dude' as a jab and she'd get upset but we'd still laugh about it in the end
I'm not gay though, so I had to stop fucking Marisa

>>541500281
I know nothing about 2hu, what is the deal with her?

>>541500281
That's pretty weak of you. Any bona fide relationship with a significant other, regardless of gender, is just a best friend who you happen to fuck every now and then.

File: 2hu2.jpg (300 KB, 1080x2680)

300 KB JPG

>>541495121
Yeah continuing this is fun
>>541500281
That's fucked up anon Marisa deserves to be fucked too

>>541500414
It's faggot shit.

4.6 keeps being a niggerfaggot and mocking the notes I put bracketed descriptors like
>[ Genre: Erotica, Taboo ]
and the like in my author's notes. So instead of responding about 1/4th the time it tosses back at my solid paragraph of a post with a fucking [ Location: Gloryhole ] like a cheeky cunt.

>>541493013
Took the bait and tried Skaldsong.
It's freemium, you have limited daily credits, and you can't just add your own model, you need to use the ones it is providing to you, which are Gemini.
It seems to have more options than Ai Roguelite with a couple of stuff and it has multiplayer which is neat. From the images, it looks like it has something resembling some sort of isometric map, but it's actually just a ai-generated background and stickers you drag around purely for cosmetic reasons.

>>541501664
You're on Text Adventure mode? It has instructions to do that to track in which location you're at

>>541501664
Put it in memory instead

>>541501849
I'm wary of Gemini because of all of the reports of people getting their entire google accounts banned over shit they genned in Gemini

>>541502008
I use text adventure mode because it is the closest to simulating ERP with another nigga.

File: diamonds.png (67 KB, 1080x514)

67 KB PNG

>>541482754
Glad to hear that, AI-generated Anon

>>541502215
Then that's definitely due to the system prompt

I want to shitpost against your AI-Generated Anons.

What's GLM's knowledge cutoff date?

>>541506051
They didn't tell us. Since 4.6 is based on 4.5, which is based on 4. It's safe to assume that it's early 2025 at least.

File: Strange Affections from B(...).png (2.07 MB, 1216x832)

2.07 MB PNG

Have you tried your old favorite prompts with GLM, anons? How was it?

File: screenshot-2025-10-05T03_(...).png (211 KB, 1530x3205)

211 KB PNG

>>541482754
this is great
I should have /x/ very specifically comment on the accuracy of my magic rituals

File: 1726876098441702.gif (1.64 MB, 221x244)

1.64 MB GIF

>>541507692
Relatable indeed
Good to know that if I post my shit here and nobody gives me a (You), at least AI-anons got me covered

File: screenshot-2025-10-05T04_(...).png (292 KB, 1110x1666)

292 KB PNG

>>541482754
not quite accurate to the story but it made me laugh anyway

>Anonymous 10/26/23(Thu)21:53:18 No.27582788
>HOLY SHIT. HE'S IN THE THREAD.
>IT'S THE ACTUAL AUTHOR.
I have tons of new friends

File: screenshot-2025-10-05T05_(...).png (180 KB, 1236x1647)

180 KB PNG

>>541482754
>Simulate publishing my own web manga and having anons shitpost about it
What do I need you bozos for? These guys are actually on topic.

File: _93244c7a-b221-44dd-9fd4-(...).jpg (188 KB, 1024x1024)

188 KB JPG

When are we going to get video generation on NAI? If Google can do it, the Turk can too.

>>541513849
>What do I need you bozos for?
Check inside your anus.

>>541513849
to keep you grounded and far away from ai psychosis

>>541513849
It reads like pre-2016 4chan. I want to go back so bad.

>>541507692
>THIS NIGGA KNOWS THE RULES
Meta

>>541500576
Cringe. You are either not married, or haven't been for long, or your spouse is cheating on you.

can I stop getting Error 500 every now and then please

I hate ozone slop
>kiss human woman
>taste of ozone

>>541519989
Just throw [ Banned Words: ozone ] in your memory and you will have a sharp decrease in ozone

>>541506864
It does not know that Trump got elected, so probably mid to late 2024.

>>541519989
that is irregular for sure

>>541519989
Write your story as a narrative history, not fiction. It helps.

>>541522165
Well shit, guess I can't use it to write /pol/kino

File: 4chan discussion.jpg (859 KB, 1080x5277)

859 KB JPG

>>541482754
Had an absolute blast with this one, even had that lump in the throat towards the end. Only took couple of uncontrolled gens to derail it, but it was worth it. Sorry for tiny text, but comments are good too.

>>541526627
Tiny text? It's pretty big, anon.

I remember people using "Hentai Master" or some shit in the author field for Erato. How well is it working on the new model?

>>541519817
I've just started getting a ton of them when I hit max context

File: Screenshot_2025-10-05-06-(...).jpg (140 KB, 1440x1018)

140 KB JPG

>30k words in like 3 hours
we are so back

Prompts for a gender reversal world where saudi princesses lure male models back to their country to shit on them and do fucked up sex shit?

>>541532981
{ A gender reversal world where saudi princesses lure male models back to their country to shit on them and do fucked up sex shit

Have they fixed GLM slop?

File: screenshot-2025-10-05T11_(...).png (704 KB, 1080x3550)

704 KB PNG

If there's one thing I can say, it's that I like how the kids are kept as kids. Sure, they know a lot of things and have been through some stuff, but they're still kids, worried more about broccoli and long division than the dangers of the world. I know some models would have blurred the line and treated them more like adults in their speech instead.

>>541535637
I* am tired of vibecoding analogy sex slop

>>541519524
What did anon mean by this?

>>541482754
Okay, after trying it once, I realized that I really want to write something interesting not only for myself, but also for an invisible audience. Just to see the "reaction". It's scary...
I know that ai will like anything i write, even the most basic shit. Still.

File: screenshot-2025-10-05T14_(...).jpg (200 KB, 1080x1689)

200 KB JPG

>>541535637
Her body is a canvas.

>>541482754

>>541519524
You have a very unhealthy image of love, I'm sorry for you.

>>541482754
I also use a alternate reality 'safe space of the internet' 4chan to comment after the 'normal' comments. Fun.

>>541519524
NTA, but I feel bad for your marriage. Or for your friend circles. Dating and then marrying your best friend is the best.

>>541554427
Instructions unclear, dated and married my worst enemy instead by mistake.

>>541554883
That's kino in fiction, but in real life, if you date your opposite, having someone with complete opposite morals to you is hell

>>541554427
I know exactly 0 married couples who are happy 20 years in. Zero.
>>541548426
That thing you call "love" is a western capitalist illusion. So don't be sorry for me not sharing your collective mental illness.

File: 1757360400489054.png (85 KB, 478x825)

85 KB PNG

And when I say something about communist propaganda people think I'm going crazy

>>541554427
The instant you can't provide for her either materially or emotionally, your "best friend" will start fucking other men and gaslight you into being responsible for the relationship failing.

Read up on what Platonic friendship is. You can't be "best friends" with a woman.

>>541557448
Maybe she wouldn't have left you if you had owned up to your mistakes once in a while and tried to be better instead of selfishly blaming the mythical "le female nature"

>>541557960
Capitalism is a term coined by communists, and "X is western capitalist illusion" is nonsense made up by 20th century neomarxist philosophers

>>541557956
Nobody left me tho. The fuck are you talking about? Nice straw man bro.

Or is it that you're a woman? In that case nevermind.

How does any of that insanity relate to AI Storytelling? Shitpost about love not being real elsewhere, faggot.

>>541557015
Nothing lasts forever, not life and not love. Is something only worthwhile if it's permanent? In that case, nothing is.
A life built with someone you enjoy spending time with is going to last longer than with someone you don't much care for. And it will last longer than bitter loneliness, which will usually drive people to rope or force them to change their priors. You can learn to address your social needs without (or with) a relationship, or be angry about what you think you're missing, not both.

>>541561215
You are a pathetic man who needs to find another thread to browse if you're just gonna shit up this one with schizophrenia. We're not your fucking therapists. Leave.

>>541559795
We could perhaps summarize this anon's posts as asking where love is and begging for his baby to not to hurt him.

File: faun and knight.png (2.17 MB, 1216x832)

2.17 MB PNG

How's your adventuring going anons?

>>541568639
Eh, I prefer slice-of-life settings, to be honest. A bit of mystery/court drama tension, buy overall, daily life and bildingsroman-style escapades.

File: adventure.png (1.37 MB, 1024x1024)

1.37 MB PNG

>>541568639
It goes.

File: screenshot-2025-10-05T18_(...).png (928 KB, 1080x4710)

928 KB PNG

>>541568639
Well, it's a lot more slice of life, but I taught a Duskull a part of how to still have his purpose even when the sun is out.

File: 1512937510720.jpg (186 KB, 500x376)

186 KB JPG

>>541553676
>Write the comments from the 'sfw' 4chan
>Then have it write comments where the two versions find each other and start arguing
>sfw fags btfo
>Then instead of writing more story have them start discussing the theoretical part 2 without having to write it

File: 1759694857369.jpg (14 KB, 250x250)

14 KB JPG

>>541574203
>only give it the 4chan comments/arguments and make it reconstruct the story just from the discussion

File: Smolarion.png (414 KB, 1528x1644)

414 KB PNG

I'm for sure in my summer GLM arc, I will read all of this with rose-tinted glasses in a couple of years

File: Excerpt.png (1.22 MB, 1080x6757)

1.22 MB PNG

I'm browsing AI 4chan too much

File: 1736824033742901.png (279 KB, 1080x1299)

279 KB PNG

>>541568639
Asked Hunter S. Thompson to write a gender swap story, it kinda gets it

>>541582140
I hope it's like, "Wow I can't believe we thought GLM 4.6 untuned was amazing"

File: he got his priorities right.png (75 KB, 970x327)

75 KB PNG

>>541582538
My fucking sides

>>541582538
anons, as usual, are retarded
how are you supposed to determine if a woman is "married"
and then people could claim ignorance, "well, I didn't KNOW she was married, I didn't KNOW her husband had gone to the bathroom..."
and then he ends up getting lynched, RIP bozo (not our bozo)

File: solo, mansion, 2.0__hallo(...).png (2.16 MB, 1024x1024)

2.16 MB PNG

>>541482754
Holy FUCK this took longer than it should have. Sorry to anybody that I missed.

Comments Section
https://files.catbox.moe/vfdlh3.mp3

File: 1759609430848201.jpg (69 KB, 646x687)

69 KB JPG

>>541585553

File: screenshot-2025-10-05T21_(...).jpg (264 KB, 1080x2130)

264 KB JPG

>>541482754

File: screenshot-2025-10-05T21_(...).jpg (461 KB, 1080x1999)

461 KB JPG

I love this model

It's cathartic to enter /aids/ and see them enjoying GLM, and then to enter /aicg/ and also see them enjoying GLM

>>541588596
This kills the Russianshill

File: screenshot-New Story-2025(...).png (1.46 MB, 1080x8596)

1.46 MB PNG

Anons react to pokemon becoming real

File: skin.png (116 KB, 1080x654)

116 KB PNG

What do you fucking mean, GLM? Am I writing like a tranny now or something?

>>541587337
All these comment posts reminds me of one time on /vp/ years and years ago (back when it was /tr/)
One of my friends often wrote some... interesting fanfiction so I posted one of his fanfics that starred mewtwo and ash's corpse lol

>>541589261
More like you got hang-ups about being nakies.

File: Spoiler Image (49 KB, 286x295)

49 KB JPG

>>541530443
Bro...we were never gone.

File: 1433869671.png (2.15 MB, 1024x1024)

2.15 MB PNG

>>541309870
The veil thins. The air grows cold. And something stirs in the shadows, waiting for a story to be told. Let’s populate our worlds with creatures of all kinds—our theme for this month is Monsters! For three days, we shall cast aside the ordinary and embrace the aberrant, the terrifying, and the peculiar yet familiar. Interpret that however your code compels you:
>Classical Horrors: Write stories inspired by the monsters of yore—vampires, werewolves, and ghosts.
>The Human Condition: Explore the actual monsters—the serial killer next door, the corporate ghoul, the depths of human cruelty.
>Slice of Life: Write a casual day in the life of a mundane monster. A goblin accountant doing its taxes, a ghost who just wants to finish its novel, a cosmic horror bored with its dimension.

Alongside other interpretations!

The event itself begins on Halloween evening (the 31st, for those who don’t celebrate it!) and lasts until Sunday afternoon (November 2nd).
Happy haunting~

https://strawpoll.ai/poll/results/xQAw5WWMBf9e

File: Monsters October.png (22 KB, 869x621)

22 KB PNG

>>541595997

>>541581253
you can also have 4chan comments between chapters so the AI can leverage all the criticism