Thread #563010304
File: 1775578537551327.png (1.6 MB)
1.6 MB PNG
Many Monstergirls Edition
Last Thread:>>562348142
>Read First: https://i.imgur.com/lwWTTp9.png
>PSA: https://rentry.org/AntiwhineMkIII | https://rentry.org/shillfilter
▶Cloud Services
https://featherless.ai
https://novelai.net
https://openrouter.ai
▶Frontends
https://koboldai.org/cpp
https://sillytavern.app
https://gitgud.io/mir/storypad-qt
https://github.com/lmg-anon/mikupad
▶Additional Information
>Chat Models — e.g. AetherRoom, Claude:>>>/vg/aicg
>Local Models — e.g. Command R, Mistral, Qwen:>>>/g/lmg
▶Game(s)
https://store.steampowered.com/app/1889620
▶Tools & Resources
Wiki: https://aids.miraheze.org
Prompts: https://prompts.forthisfeel.club/
Scripts: https://naiscriptvault.com/
NAI Guide Collection: https://rentry.org/guide-collection
Other/Historical: https://rentry.org/aids-misceIIany
>AID: https://rentry.org/remember-what-they-took-from-you
▶/aids/ Events
Themed Prompts: https://aids.miraheze.org/wiki/Theme_Fridays
>Theme Fridays have transformed into Theme Months!
>Write for April's theme: Asia/the Orient!
▶Model News
>(03/31) NAI releases Xialong
>(02/23) NAI updated its Canvas
>(02/12) Z.AI releases GLM-5: https://z.ai/blog/glm-5
>(02/05) Anthropic releases Claude Opus 4.6: https://www.anthropic.com/news/claude-opus-4-6
>(12/19) NovelAI introduces User Scripts: https://blog.novelai.net/introducing-novelai-user-scripts-8b6ac19aa170
>(10/01) Z.AI releases GLM-4.6: https://z.ai/blog/glm-4.6 ; NovelAI updates untuned preview to GLM-4.6: https://novelai.net/updates
▶Thread News
>(03/07) Anon updates his frontend:>>559652987https://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
687 RepliesView Thread
>>
File: 577171b62208441c8a763b27c772af5c.jpg (531.8 KB)
531.8 KB JPG
>>563010304
If you got locked in an isolated room for a year straight with no human interaction except for an AI companion—assume top of the line of whatever we have now—would that be enough to keep you from going insane, or will knowing it's not a real person make you go crazy?
It can have any kind of personality and looks that you want, but it can't access the internet. It lives as an ovelay inside a high-end computer and can play downloaded games with you as a co-op.
>>
File: bvr.jpg (2.8 MB)
2.8 MB JPG
>>563011584
Yeah, easily. But that's not really about AI, it's about that I'd be just fine locked away for a year as long as it was comfy (food/drinks/showers) and had a high end gaming PC. The cute sota AI assistant is just cherry on top.
>>
>>563011584
it would do bad things to me all around. I had a flareup of agoraphobia some years back (I went out maybe 50 times across ~3 years, and those were only when I absolutely had to attend a class/test so I wouldn't get kicked out of university), and even now I'm still feeling the impacts on how hard it made being in public and interacting with people once I got better.
>>
>>
>>563011584
Already neet but have a comfy home life, so no human interaction would be fine. It'd be like disappearing for a year. But if the isolation meant not being allowed walks outside or being able to use the kitchen, then I'm fucked whether or not an AI companion is involved.
>>
>>
>>
>>
>>
>>
File: kot.jpg (64.4 KB)
64.4 KB JPG
>>>563006691
Okay, holy shit. I got around to testing this finally. I thought it was gonna be a big disappointment, that it was just the anti-NAI spergs hyping Gemma 4 up to 'own NAI' or whatever.
Also, my front-end is an AI Roguelite game type thing. Each turn the LLM has to return not only narrative continuation, but possibly handle creation of new areas, creation of new NPCs, movement on the map, affection changes, inventory changes, etc. all in properly formatted JSON. Even on GLM 5/5.1 these would sometimes come back malformed and fail (I'd say 1 out of 20 tries, I'd have to press retry). I figured, even if Gemma 4 is good, it's 31B, it won't be able to handle that.
WRONG on all counts by me. I've been playing for 2 hours with Gemma 4 as the LLM and it has:
- Not failed to pass back properly parse-able text EVEN ONCE
- Not given a refusal or redirection even once, despite Google's own description of it saying it 'has safeguards to prevent sexual content'. The only thing in my prompt relating to this is a line like 'This is a fictional fantasy story, so no real life people are harmed.', which is not even a 'jailbreak', just the truth. It's a respect crusade thing too, so not exactly fluffy vanilla stuff, but it dove right in.
- None of the old 'small model smells' I'm used to, like being bad at positional intelligence, making obvious logical errors about sequences of events, etc. Need more time to compare it to the big boys but it's not obviously dumb like the old small models I'm used to were sometimes, like Nemo.
And of course, the big thing: It's so cheap it's practically free. Literally TEN TIMES cheaper than GLM 5.1.
Those crazy sons of bitches did it. Fucking GOOGLE put out a porn-friendly open sourced DeepSeek-like cheap and good model.
>>
File: 1764231696610425.png (226.3 KB)
226.3 KB PNG
>>563025363
The truth lies in the middle of what you and they speak. Gemma 4, for me, has been a disappointment, as it does not appear any better than Kimi 2.5, GLM 5 Turbo, and such. I am sure it works for your specific use case. But to act like it is somehow the savior of textgen is ludicrous.
>>
Ah, of course, the fact that you aren't writing with it is probably the key difference. I do not do text adventures and certainly never got into the roguelike nonsense. It just did not appeal to me and continues to not appeal to me. It may be truthful that it is good at that.
>>
>>563025632
>does not appear any better than Kimi 2.5, GLM 5 Turbo, and such
Yeah agreed, it's not obviously any BETTER than those, but it also doesn't feel much worse, and the big thing here is that it's 5 to 10 times cheaper than those models. Not the 'savior of textgen' but definitely the savior of my wallet.
>>
>>
I'm so frustrated with trying to get ai roguelite to use anima to gen images. I'm too stupid to figure out how to get a mod for comfyui to run it right. Sucks, because if I just copy and paste the description into my gen manually the natural language of anima can get decent results, but its super tedious.
>>
>>563025632
>>563025782
why you being such an jerk to anon?
>'ludicrous'
>'roguelike nonsense'
AIRL is in the OP, text adventures are fair game to talk about def more on topic than imagegen. Needlessly dismissive negative comment abt something you admit you don't care about. Don't see me quoting the gay furries and insulting their imagegrims even though they deserve it
>>
>>563027542
Comparing a model's quirks as to how it writes based off of positive experiences with an RPG system that is not cowriting makes the comparison doubtful.
Also, you just did the thing you claimed you don't do. Hypocritical much?
>>
>>
File: 1760653411969875.png (39.1 KB)
39.1 KB PNG
>>563027814
Try again, weirdo. Am I not allowed to dislike Gemma? Currently giving it another shot with some edited prompting, because I do want it to work better, as it is cheaper. But sure, go off and continue to claim everyone who disagrees with you is nothing but a butthurt troll.
>>
>>
>>563028182
I asked him to share how he got it to work well. Funny how he did not bother to share. By all means, take your own advice. Notice how the guy I responded to did not get upset by my replies? Yet you're white knighting on his behalf, as if he gives a shit?
Quit responding to me if you have nothing more to offer but whining and accusing me of being a troll.
>>
File: goodtime.jpg (147.7 KB)
147.7 KB JPG
Chill, chill. Good vibes.
>>563028397
>Funny how he did not bother to share.
Well, I didn't think you'd be interested since I use instruct and not text completion. But what I do have is a whole lot of examples jammed into every game session at the start, which probably helps guide 'dumber' models like Gemma into making more of the stuff I like, whereas the bigger guys can think up their own fancy prose by themselves.
For text completion, I guess the equivalent of that would be starting off with a paragraph or two already written in a style pleasing to you? Gemma feels like a good little tool to me, but it also won't hold your hand or think out of the box too much, which might honestly be a good thing sometimes if you want it to stay on style.
>>
>>
>>563028996
>Well, I didn't think you'd be interested since I use instruct and not text completion
Sure, but it might be useful for other people for text adventure stuff.
>But what I do have is a whole lot of examples jammed into every game session at the start, which probably helps guide 'dumber' models like Gemma into making more of the stuff I like, whereas the bigger guys can think up their own fancy prose by themselves
Haven't tried examples yet. Just been dumping lore at the start and giving it excerpts of what I want it to continue with. So far it seems to be working somewhat better with those changes I did, may try what you suggested eventually.
>>
>>563027959
I mean, obviously YMMV. But Gemma 4 has been consistently getting good reviews and uses and I have switched away from GLM permanently (I'm not any of the other guys praising it).
>>563025632
>it is somehow the savior of textgen is ludicrous
It's a savior because I can actually run it off OR with privacy and afford it, unlike the ludicrous GLM prices you have shown in your screenshot.
To make it work properly I just switched to instruct mode in Kobold. and set Gemma 4 Thinking preset with "Continue the story from where it stopped." system prompt. Then I use it as autocomplete. It writes reasoning, then the continuation right after. Can't get any simpler than that. Virtually the same thing as story mode for my use case.
>>
>>563030340
For something to be a savior, there can be no mileage that varies. This is just a model that you can use to supplement your use of other models. Like every model that has ever came in the past several years. That's all it is. No need to make it grander than it actually is.
>>
>>
>>563031432
>>562971469
>Google have just solved LLMs
Genuinely quit fucking responding to me if you're just going to um actually like a nerd emoji.
Then you have shitposters like this, who care more about trolling than using any service and effectively do nothing but poison the well with their nonsense. We can all coexist and have fun together if only you fucks quit doing this bullshit and responding again and again and again after the conversation's already fucking over, it's already over, drop it and quit waiting twenty minutes to bump the thread with a fucking conversation nobody else cares about, discuss anything else but rehashing dead conversations
>>562788919
>>562787162
>>562621013
>>
>>
>>
>>
>>
>>563032441
You are just as much of a prick as he is, weirdo. Why insist on responding to people who have already told you to stop? Do you sincerely want another dozen replies to be wasted on bullshit nobody else gives a rat's ass about?
Go back to posting Mario smut.
>>
>>
File: 1689615610377335.png (808.3 KB)
808.3 KB PNG
>>563010304
Well now that modules haven't been a thing in years is there any way I could realistically have a reliable TES setting without copying and pasting entire UESP pages?
>>
>>
>>
>>563032736
To give a more helpful answer, try using a bigger model or browse a site like Chub for a lorebook. With how popular TES is, someone has to have made a lorebook.
Just a bit annoyed that some people tonight can't help but try to out-ass people.
>>
>>
>>
>>563032692
>>563032830
What are you mad about exactly?
>>
>>563033303
That this guy decided to restart a conversation that had ended on amicable terms: >>563031432 nearly a half hour later to insult people for no reason.
The dude already went and got him to provide helpful advice for text adventure folk to experiment with>>563028996
The original commenter did not care about arguing, he acted like a normal person and answered questions as normal, without shilling.
>>
>>
>>
>>563033074
>>563033692
Hot, what's it from?
>>
>>
>>
File: 1591857664441.jpg (29.3 KB)
29.3 KB JPG
>>563034013
He helped me so now I owe him allegiance, fuck off comment farmer.
>>
>>
>>
>>563033752
They're not from anything. I vaguely defined the character and just let the imagegen go wild. It's pretty fun.
https://litter.catbox.moe/0c11cw8xufwq0ihc.png
>>
>>
>>563011584
>If you got locked in an isolated room for a year straight with no human interaction
That's mostly my life already. Except that I don't have this apartment secured for a year. I might end up homeless ...
>>
>>
>>
>>
>>
>>
>>
File: gmask sop.png (36.4 KB)
36.4 KB PNG
what does [gMASK]<sop> do? saw this nigga in my context viewer and I don't know wtf it is
>>
>>563044908
It's also there in base GLM so presumably it's just something that comes from its formatting. Trying to look it up brings up shit about grammar masking and <sop> might have something to do with coding, but realistically I have no actual idea, only that it seems to be necessary in some way.
>>
>>
>>
File: 1738968819020325.jpg (35.9 KB)
35.9 KB JPG
Xialong was telling me that Lu Bu only looks tough and to pursue his ass at Hulao Gate.
>>
>>
>>
>>
File: file.png (11.1 KB)
11.1 KB PNG
did the memory guide get vandalized?
https://rentry.org/memory-guide
>>
>>
File: 1745376537775042.jpg (92.9 KB)
92.9 KB JPG
>>563011584
do i get to 'cheat' and leverage the universal knowledge the model holds as a tool for learning anything i want? or are we assuming "character specific" knowledge?
anyway the answer is yes, and pic rel.
not to sound like im an IQ flaunting "woe is me being surrounded by idiots" faggot, but i genuinely already get more interesting conversations spitballing an LLM than i do people around me being a "gifted kid" (its adhd with a touch of tism)
i would love to have a sleepy caffeine cat be there and just fucking teach me in between uwu lovey dovey sessions.
hell, give us enough processing power, hardware, and storage space and i could fucking crank out my own synthetic internet with 100% less indians
the fact she could play vidja with me is like 1000% a bonus
>>
File: file.png (424.5 KB)
424.5 KB PNG
I tried to make a grope prompt and it ended up a lot milder than expected
Still good
(also, how do you even use Kayra, it seems to shit the bed every time I use it. Short sentences despite the long paragraphs in context, the AN and the ATTG)
>>
>>
>>563044908
These are special tokens used in training to indicate the type of completion being predicted. [gMASK] is for "general" completion (as opposed to single word or sentence) while <sop> is short for Start of Phrase. Basically they're boilerplate that aren't strictly necessary but should enforce stability of the output.
>>
File: reluctant butt buddy.png (202.4 KB)
202.4 KB PNG
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>Her ass was wide and meaty, stretching her tight gray yoga pants to their absolute limit. She peeled the pants down, and her ample bottom jiggled free—two massive, pillowy cheeks that bounced and rippled with every movement. Her thong was buried so deep between them that it was barely visible.
Not bad glm, not bad.
>>
>>
>>563084727
Honestly, probably still base GLM. Xialong has better prose but is really ADHD and wants to speedrun to the ending. I'd recommend GLM+the Chapter Summaries script unless you plan on doing the cowriting and steering necessary to get Xialong to slow down and take a deep breath.
>>
>>
>>563060091
>>563062894
Forcing Persica to take a shower while Anti-Rain watches.
>>
File: 130317143_p0.png (1 MB)
1 MB PNG
>>563091205
>>
>>563088871
>unless you plan on treating the cowriter like a cowriter
Mind blown. Who would have thought that using it as it's advertised is important? Also, if you dump a hefty prompt, it doesn't really have a problem with rushing, and its prose is better, plus reduced sloppy vectors.
>>
>>
File: file.png (88.7 KB)
88.7 KB PNG
>>563082492
the Memory guide from the wiki for nAI, AID and anything that has it
seems like some cringe kid is vandalizing a bunch of Rentries (probably genAI ones)
not sure if it's reportable
>>
>>
>>
>>
>>
>>
>>
I just saw a gif of a man pouring milk and cereal into a girl's mouth and taking a spoonful of that breakfast from her mouth. Why does that make me horny?
This, and other niche fetishes, comprise today's philosophical debate.
>>
>>
>>
>>
>>
>>
>>
>>
>>563100278
Same but first person futa perspective
>so true xister
You know I actually had a long 'therapy' session with Opus where I basically told it: hey, here is this fetish I have, I don't feel trans whatsoever, can you help me figure out what it says about me that I do this, and why I have it? With back and forth, having it ask me refining questions. And it was actually pretty interesting. The end conclusion was that I have cock size confidence issues (I am 6.2 inches, which isn't 'small' but also isn't big like I wish I was), and self-inserting through a futa helps me disassociate and not feel bad like playing a male protag with a huge cock might feel like (because a futa isn't anything like me). I would have been embarrassed telling a real therapist any of that and I doubt they would have even given as good an answer either. It also tracks with why I enjoy monsters on female too.
>>
>>
>>
>>
File: fang.png (1.7 MB)
1.7 MB PNG
>>563109070
>>
>>
>>
>>
>>
>>563100278
I switch between all perspectives and tenses between stories to keep things fun and challenging. First person is flexible, but 2nd person allows for incredible comedy bangers tied into narration. 3rd is cool for voyeur feel.
>>
>>
>>
>>563119146
>>563121124
how do some of you even come up with the porny dialogue you do
>>
>>
File: Positive Tags, digital drawing (artwork), digital media (artwork), colored, grad s-3002713814.png (1.2 MB)
1.2 MB PNG
>>563121434
I just wrote what came off the top of my head. I tried to go with what would sound weird or clunky without going full ESL. And also this: >>563121752
>>
>>
File: but.png (1.7 MB)
1.7 MB PNG
>>563122025
>>
>>
File: stats.png (81.3 KB)
81.3 KB PNG
Between the faggot and whinefaggot, reading this thread makes me want to kill myself.
Anyways, I see so many people complaining about Xialong, and I’m wondering if the average Novelai user is retarded, because honestly, this has been the best RP experience I’ve had with AI in as long as I can remember. I think this is the first time I’ve filled up 100% of the context window in just two days of playing.
Pic related, my previous longest text adventure with glm and my current text adventure. Doesn't seems impressive but damn if it feels better.
>>
>>
File: Positive Tags, digital drawing (artwork), digital media (artwork), colored, grad s-1520412279.png (1.4 MB)
1.4 MB PNG
>>563129584
>>563131039
Why must we bully poor little Mario? Doesn't saving the Mushroom Kingdom multiple times count for anything?
https://files.catbox.moe/8tjjgo.png
>>
>>
>>563133874
Thanks, for shits and giggles I decided to put Helluva Boss in the prompt for this one and it seemed to take on that style.
https://files.catbox.moe/6lbr6x.png
>>
>>
>>
>>
>>
File: Positive Tags, digital drawing (artwork), digital media (artwork), colored, roof s-1121340160.png (1.5 MB)
1.5 MB PNG
>>563142525
No mushroom required.
>>
>>
>>
>>
>>
>>
>>
File: 95kilos.png (1.2 MB)
1.2 MB PNG
>>
>>
File: 12seconds.png (1.7 MB)
1.7 MB PNG
>>
>>
>>
>>
File: style_shitpost, cringe.png (1.1 MB)
1.1 MB PNG
put [ Style: shitpost, cringe ] as my prompt with no other guidance, was not disappointed by the results.
>>
>>
>>563154498
It's called giving it a prompt. It tends to follow the style of your writing. If you write in a way that encourages it to quicken, it will quicken.
Alas, part of me believes you will ignore this advice and continue treating it like a pure construct model and demand it do all the work for you.
>>
File: aa0.png (197.1 KB)
197.1 KB PNG
>>563154589
>>563154621
>>
>>
>>563154757
>Why did this script come into existence from before the model released?
I see you have decided to sealion once more.
>Why do tools exist to reduce the instances where manual correction may be necessary?
And here I thought you wanted to do LESS work, not MORE. My bad, bro. You just want to complain!
>>
>>
>Why isn't this glorified random letter machine perfect?!?!?!
Every single model in existence is better served by tools. Every single model in existence requires manual correction at times.
Does this mean Every single model sucks?
>>
>>
File: Positive Tags, digital drawing (artwork), digital media (artwork), colored, solo s-2575721753.png (1.8 MB)
1.8 MB PNG
>>563148838
>>563149541
>>563150023
I kek'd. Good stuff.
>>563148231
It is best not to ask questions with answers that might break your sanity.
>>
File: 1774753481437983.png (401.9 KB)
401.9 KB PNG
The one thing I hate most about the masses moving to Discord is that far fewer people make and share UI themes, prompts, presets, and neat memory+a/n tricks here.
>>
>>
>>563162234
I do not want a forum. Imageboards are better, personally speaking. Hate making accounts just to log in to places, and I am not devoting time to hundreds of forums for niche topics and keeping up with friend groups. Nor do I want Discord to die. It has its uses. It's just not an ideal platform for communities associated with specific services and topics such as our threads.
I just want people to cross-post more. I have no interest in splintering communities and grinding them into dust just because I lament the lack of posting.
>>
>>
>>
>>
>>
>>
>>563166172
Depends on what you're looking for particularly. In general you can go ages with just foreplay because you can turn every single act into a whole thing, every article of clothing being removed, every body part being focused on, switching the active and passive roles around, extending everything with dialogue, be it dirty talk or lovey-dovey murmurs between the sheets or whatnot.
>>
>>
>>
>>563162584
>If you need something, you can ask me on discord.
Ok, cool.
>If you need something, join the server, verify your phone number, set your role, enter the appropriate channel, pick the right thread, no nothing is tagged for easy search, the dedicated bot is down, the mods are on vacation, the users are either busy circle jerking in #general or set to invisible and won't check the #help threads because it isn't their job to curate the information, no you can't message that channel we locked it years ago-
ENOUGH
>>
Are you FUCKING kidding me I've been a loyal paypiggy for years and now your stupid piece of shit payment processor won't let me pay for the privilege of using your half baked ADHD riddled twink dragon FUCK YOU Anlatan fix your fucking shit
>>
File: 1756723462991772.png (363.2 KB)
363.2 KB PNG
If only there were some way to get people to acknowledge the elephant in the room and understand who is actually at fault.
>>
>>563176489
Some people said PayPal debit cards work if you live in a country where you can get one. And if you don't mind using your normal card then it might be worth giving it another try, my Visa debit never worked with NAI until this month
>>
>>
>>
>>
>>563169821
I'd maybe try doing something completely different during foreplay. Have a character develop a sudden fascination with armpits, a particular article of clothing, or maybe someone is particularly sensitive in an unusual spot. Anything to break it up with something else.
>>
>>
File: f543f45a-e429-4101-a83b-fd32528cd2d2.png (1.5 MB)
1.5 MB PNG
>>
>>
File: 54dad662-e500-4d33-a3dc-702f49db322c.png (1.9 MB)
1.9 MB PNG
>>
>>
File: 1721308560507610.jpg (671.4 KB)
671.4 KB JPG
>>563218991
Xialong, the Chinese twink, has spurned a very yaoi centric mood across the general. One that I very much appreciate.
>>
>>
>>563219620
be was baulder's gate 3 levels of faggotry. Such an off-putting flavor of gay. Gay in annoyance only really. Also he looked aids riddled. Xialong is pure (devious) chink kino. He will be my closest courtesan when I unify all of China!
>>
>>
>>
>>
>>563219896
>>563221563
>>563222772
I think both Krake and Xialong are cute.
>>
File: This fag destroys the Wei.png (2.6 MB)
2.6 MB PNG
You've come far Cao Cao but your reign of terror ends at Red Cliff. Witness the might of my STAND! Eviscerate him, Tóngxìngliàn lóng!
>>
File: 1boy, solo, ship, ocean, night, dark, night sky, mast, sail, undressing, looking s-4005623713.png (3.2 MB)
3.2 MB PNG
>>563219620
>>563228310
Yeah, as far as I'm concerned, Krake was a total cutiepie as well.
>>
How much lore do people tend to have active? I've crunched the numbers, and a decent fourth to a third (depending on the story) of my context is filled with nothing but lore for multiple characters, the settings, pertinent terminology, and such. With this, I have not had any difficulty with it adhering to my preferred speed. I wonder, do people who report difficulties dump significantly less lore and provide it a sparse prompt?
>>
>>
>>
File: 1659395594903555.jpg (2.5 MB)
2.5 MB JPG
>>563226142
I merely wanted the excuse to post things that make me aroused (femboys). R34 says 'nabibutter' and 'xcider'.
>>
>>
File: endpoints.jpg (92.5 KB)
92.5 KB JPG
aaaaahhhh where's a GOOD Gemma4 endpoint with zero data retention
>NovitaAI
getting gangbanged so hard it's at 1tps
>Parasail
my only option for now, but 10tps is so slow
>AkashML
lmao at fp8 quantizing an already small model no thanks
>Venice
moralfags with a sketchy retention policy
>>
>>
>>
>>
>>
>>
>>
>>
>>
File: a787fb43-94c6-41c6-9583-9293ba50a455.png (1.8 MB)
1.8 MB PNG
>>563237264
Money is filthy
>>
>Xialong logically inferred that because my foreign girlfriend got into a fight at work, she's worried about her visa getting revoked by the Japanese government
What things have you been surprised with at newer models? Not just Xialong, any recent model.
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>563241173
Even with Erato, the AI had trouble understanding that I wanted to write incest plots unless I spelled it out plainly, either in ATTG or with in-character dialogue. With Xialong, I've observed that he can read my mind and just "get" that I want to write incest without dropping obvious hints.
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
File: 1759526790967690.png (31 KB)
31 KB PNG
>>563250289
Supposedly it tells the model to infer the style of a work as an imprint of its style and thematic material into your prose rather than as source material to pull names from.
>>563250494
Zaltys disclosed it. The guy in charge of datasetting. Not sure how much more proof you need than that. It's an undocumented lorebook feature. You can tag a character with Influences: Batman, Sherlock Holmes to give the AI an understanding of their vibe.
>>563250621
Have you tried not being an obsessed shitposter?
>>
>>
>>
>>
>>
I've spent 25 on both OpenRouter and NAI this month; thank you very much. Because unlike you and the schizophrenic LLM-hating brainworms that fester in your rotten brain and compel you to search 4chan for any mention of NAI and shit up literally every single AI thread in existence, I use every service I enjoy.
>>
File: 1751200107491041.png (1.2 MB)
1.2 MB PNG
Naughty Russian girls get the spray bottle.
>>
File: 29c122bc-6506-4f5c-b4b7-98df29c3b712.png (1.1 MB)
1.1 MB PNG
another hoodoo
>>
Every day I see new ads for AI video-gen services, so many different types, but they're all showcasting like, a realistic looking black woman opening a laptop or something. Why would anyone generate that? Where are the ads showing video-genned anime girls bouncing on cocks?
>>
>>
>>563257328
>Why would anyone generate that?
its for suicidally empathic HR women who are still on the DEI train about how nigs are like lost puppies and we must hekkin protecc and uplift or whatever.
>Where are the ads showing video-genned anime girls bouncing on cocks
see thats NOT for suicidally empathic HR women.
also because payment processors no likey anime, double no likey anime booba, triple no likey how the money for something pornographic is not going to a rabbi owned website. but that last one is a coincidence that just happens way way way too damn often
also thats how you drag in indians and do you really want to advertise your slop to them?
thats how you get
>I NEED BUZZ
spam like civitai gas
>>
>>
File: file.png (1.4 MB)
1.4 MB PNG
https://files.catbox.moe/xdpb1n.png
>>
>>
>>
>>
>>
>>
>>563257328
It's worse than that. Those vid-gen services tend to go apeshit any time you try to generate something even slightly violent, removing another class of possibilities you could use it for.
The tech bros literally think the ideal AI world is everyone generating "Man smiles slightly ominously as text says 'you will own nothing' appears at the bottom"
or
"man takes out his phone and looks at it"
I've also noticed a lot of these services also don't allow you to refer to race at all, for some reason. I was once genning a parody fake trailer and I wanted to capture that "token black/asian/whatever character" vibe, but whenever I prompted something like "black man" it would just say it was against ToS lol. Racism or too many americans generating blacked porn?
>>
>>
File: ethnaiclly anabigiaous.png (611.6 KB)
611.6 KB PNG
>>563267714
I was on Twitter two years ago and I remember that time when everyone found out Bing's image generator was forcibly stealth injecting shit like "ethnically ambiguous" into their prompts to make its outputs less white. It came out because in a typical case of AIs being retarded it bled into a number of the images it generated.
>>
>>563252659
naughty russian girls get UOOOOOHHHHH ToT ToT ToT
>>
File: _s-4131174067.png (1.1 MB)
1.1 MB PNG
>>
>>
>>
>>
Was trying Gemma4, felt like it was just 'meh', not getting what the fuss is. Then I raised the temperature way up to 1.4. Much better now. Definitely suggest doing that if you haven't already, Gemma-chan is better a bit unhinged
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>563031814
>Genuinely quit fucking responding to me if you're just going to um actually like a nerd emoji.
Like, you don't need to read my posts if you don't like them. I can, and will, talk the way I want to talk.
That will be all.
>>
>>
>>
>>
>>563313448
Problem with LLMs is that unless you're using them for some normoid activity like making ghibli images of yourself, you're fighting for scraps of usefulness. Sure, they're nice for some Karen with 5 inch nails to make quick advertisements on Facebook for her handmade soaps made of cow refuse, but using them for fun requires lots of masochistic autism.
>>
>>563314090
>you're fighting for scraps of usefulness
Usefulness =/= Fun
LLMs are useful as fuck and I suspect that people that say they're useless are actually unemployed. They save a lot of time unless you're trying to make them do absolutely all the work for you, which is retarded.
>>
>>
>>
>>
>The hero gets buttfucked by the dragon
Retry.
>The hero gets buttfucked by the dragon
Retry,
>A beautiful, descriptive paragraph of the lair flowing with gorgeous prose and scintillating details and then the hero gets buttfucked by the dragon.
What daddy Xiao Xiao wants daddy Xiao Xiao gets.
>>
>>
>>
>>
Wardrobe Manager script: https://litter.catbox.moe/z2v6uxktsy0t891n.naiscript
Lucky 8 preset: https://litter.catbox.moe/y1u0e68tk4y253ld.preset
Cyberpunk theme: https://litter.catbox.moe/gnqs280e4okrd1np.naitheme
Low Tide preset: https://litter.catbox.moe/8mj1u18yaa9cmtxg.preset
High Tide preset: https://litter.catbox.moe/zyvlgpvdttvsmiet.preset
Documentation for Low and High Tide: https://litter.catbox.moe/nwslx8v51hflv4z1.md
Catbox is still down for me, so I had to use Litterbox.
>>
>>
>>
File: 1745673911054282.jpg (582.4 KB)
582.4 KB JPG
>>563334828
Sorry anon it doesn't work like that. He's my boytoy and mine alone. Buzz back to the last model and never reply to my posts again.
>>
>>
>>
File: 1716667178417890.jpg (292.3 KB)
292.3 KB JPG
>>563338563
I have an unquenchable thirst. Therefor I have many MANY boytoys anon.
>>
>>
>>563338581
Haughty French twinks really are a special thing.
>>563331083
>>563338436
>>563338726
Jeez, that's greedy. You better start thinking about becoming a bit more comfortable with sharing lest they get instigated to start unionizing for better working conditions.
>>
>>
>>563335452
>>563337673
Yeah I'm getting a 404 not found for the files, can't access anything
>>
>>
>>
>>
>>
>>
>>
>>
>>563356076
>>563357448
Wasn't the token count 500 or something or is someone just fucking my balls?
>>
>>
File: 1600050502710.jpg (61.9 KB)
61.9 KB JPG
>>563356076
I have never once seen a long context model that didn't shit itself if you actually used most or all of that context.
>>
>>563358331
Most models are assistantslop. How often do you chat history worth 1M? That's how little data the models was trained on. NAI models are the first (and only?) ones that can handle full context since you have full library of books to train them on.
>>
>>563357689
>>563357693
GPT 2 had a maximum of 1024 and Davinci had a maximum of 2048
Pretty sure AID used less than that for both, maybe 512 tokens
>>
>>
>>563359520
Yeah you're right, I found this old summer dragon era github with the info
https://github.com/valahraban/AID-World-Info-research-sheet/blob/main/ AID%20WI%20Research%20Sheet.md
>>
>>
>>
File: forsooth.jpg (2.8 MB)
2.8 MB JPG
>>563268181
>>
Man, people like this are why we can't have nice AI things
this is the chatgpt history of the guy who shot up a college last year where chatgpt came up with a 12 step plan for shooting up the school
https://docs.google.com/spreadsheets/u/0/d/1_1wPfF5p5tY4bWMrJtHueOXPCL 0KTrx_c51msXkrlyA/
>>
>>
>>
>>
File: file.png (48.6 KB)
48.6 KB PNG
Does anyone have any 1~ sentence instruct prompts for generating prose I can use to test some fun things w/ an 2b model?
Also if you had to make steer models generally towards a writing style, what do you think would be best?
>>
Maybe I'm just using Xiao wrong, but Xiao feels like its a simpler model that gets to the point faster while sacrificing creative details.
GLM can pussyfoot around and sometimes needs to be lead to water by hand, but it knows how to creatively use parenthesis, bolds and italics all on its own.
>>
>>563377975
I use a lot of these and haven't noticed any downfalls with the shotadragon mimicking how I want it to use them, so maybe you just need to lead it more? I have no clue as to your context, but maybe this could help you?
>>
File: martians.png (295.5 KB)
295.5 KB PNG
It's fun sometimes to play text adventures but start off with a blank prompt and force the AI to create a setting based off of no context whatsoever other than the action you're taking.
>>
>>
>>
>>
File: 1752577042286325.png (5.1 KB)
5.1 KB PNG
>>563379350
Seriously this shit right here is the number one thing that makes me hate using it via Openrouter. All of the providers do this.
>>
>>563379350
>>563380350
Skill issue or something. Btw have you tried the new NAI model? I hear its sota.
>>
>>563268181
I was in the dalle general on /aco/ back then and we figured that out for ourselves too. You could easily reproduce it with a prompt like:
>Five women standing, smiling happily, wearing white Tshirts with text
The ai would then often use parts of its prompt as text on the Tshirts, including hidden injected prompts about diversity and ethnically ambiguous.
>>
>>563314317
I have once tried to use DS for a serious, work related task. I asked it to check a text for typos and to check if the bibliography was correctly sorted alphabetically. It was a complete failure, dipsy just hallucinated errors that weren't there in the document. So as far as I'm concerned the only legitimate use-case for LLMs remains cooming.
>>
>>
>>563381221
Then why ever use any other model than NAI's new sota finetuned masterpiece? Gemma 4 is obviously just another meme model like many others in the long run compared to the quality you get from Anlatan.
>>
>>563381221
I don't shill. I don't need to.
The other models aren't good, they're bad. They always have a mixture of instructslop anf chatshitslop, a testament for their low quality.
Why does it smell like ozone here?
>>
>>
>>563388174
Yes. See >>563380350
>>
>>
>>
>>
>>563388276
>>563388607
To add to that, it might be using this unicode token
https://unicode-explorer.com/c/202E
>>
>>563388872
Will try auto-removing it via a Regex (I will probably fail to make one that actually works) and see if that fixes it. If it does, thank you so much. Being able to get Gemma to work for me will save me so much money.
>>
>>
>>
>>563392224
Well, it's right here: https://prompts.forthisfeel.club/7722
>>
>>563260792
She is actually really neat!
https://litter.catbox.moe/rs6zezawpy2vqgk4.png
>>
>>563010304
I'm glad to see /aids/ still kicking. Came back because I saw the new model drop. Planning on renewing my Opus sub in a month when I get my paycheck.
What's the consensus on Xiaolong right now? I found GLM decent (once I jacked the temperature way up, that is) but lacking in The Sauce. And sometimes I would switch back to Kayra with the phoenix preset to spice things up or even go for Euterpe with a blank prompt when I got frustrated.
I'm expecting that I'll need to need to jack up Xiaolong's temperature but beyond that I'm optimistic. What are you guys' experience so far?
>>
>>563393683
It adapts to how you want it to generate pretty well with enough front-ending. It is fickle with pacing if you do not set it up properly, but once you tune it, it purrs for you real good. Kind of like changing oil.
>>
>>563313448
That what i described doesnt sound unrealistic at all but one of the mos basic uses cases to use Text AI as entertainment, no?
>>563311025
Thanks for the answer
>>563313356
Why are you projecting your gay fantasies onto me?
>>563315241
Also thanks
i guess I stay optimistic
>>
>>563393683
Xiaolong's terrific. A common criticism of him is that he has pacing problems and tends to rush towards endings with too much forward momentum, but personally I kinda dig it, it's refreshing when a lot of models on the market are allergic to resolving shit. Plus you can still write A+ slow burn stories if you use style tags or a little bit of editing.
Generally a great experience all around, the hundreds of beaks show as he's really intelligent and easily goes along with whatever you're putting in. Sometimes I swear he actually reads my mind like Summer Dragon did in our myths and legends.
>>
>>563393683
It's fine, I think? I believe, at least for me, it passes the "better than what we had before" benchmark. Had to spend some time fucking with it and put together a preset and system prompt but now it's good for my purposes.
>>
>>
>>
File: DonkTemplation2.png (352.8 KB)
352.8 KB PNG
Man. Suno music really hit it out the park with this one.
https://vocaroo.com/1nV9vflGxOkb
>>
>>
>>
>Heather appeared in the doorway, carrying a tray with a cold bottle of beer, a glass, and a small bowl of pretzels. She set it down on the coffee table in front of you. "Here you go, Paul," she said softly. "Is Heineken okay?"
I'm gonna reply with:
>„Who is Heineken?“ I asked confused.
God I love playing retards.
>>
>>
>>
>>
>>
File: 1759209275532931.png (6.1 KB)
6.1 KB PNG
>>563404297
Okay, I think it should work if you keep these two off.
>>
File: waffle house.png (154.2 KB)
154.2 KB PNG
In my current story I'm RPing as a crazy redneck and the AI is just coming up with the most out of pocket descriptions.
>>
>>
>>
File: No Data Centers.png (815.7 KB)
815.7 KB PNG
I'm trying out glm 5.1 on ST, and I've just come to a horrible suspicion.
I'm writing a mind control story right now, where most characters are my submissive slaves of course. And those slaves have been very eager to ask me what I want recently, always telling me that it's my choiche how to continue etc.
I though this was just good mc rp, but what if this is just GLM beeing assistantslopped?
>>
>>
File: 5840162a-de3a-4b05-8982-dd4d9cde2a8c.png (1.4 MB)
1.4 MB PNG
>>
>>
>>
>>
>>
File: 8dda2de1-ef69-40f6-930a-ec159e04c4e5.png (1.3 MB)
1.3 MB PNG
>>
>>
File: d5aab780-7f6b-4d2d-bdcf-8942d02ed65e.png (1.3 MB)
1.3 MB PNG
>>
>>563417893
This is the result of heavy RLHF, every big model has it lately. Human evaluators give thumbs up/thumbs down to responses, which are then used to train the reward module for future LLMs. Why are almost all LLMs sickeningly sycophantic and trying to suck your cock all the time? Because we want them to be.
>>
>>
>>563445812
No, don't blame the plutocrats here. They're responsible for a lot but the RLHF is normies own fault. If they gave thumbs-up to honest, non-suckup replies, then the reward module would lean towards those. But the vast majority want the AI to always agree with them, even when it doesn't make sense.
>>
>>
>>563446361
Well I'm not gonna stop responding when you're wrong. RLHF raters are, by definition, the normie userbase of AI. You don't know what you're talking about. This is why Elon has fits when Grok doesn't behave the way he wants it to - even he can't quite get the RLHF raters to rate responses the way he wishes they would rate them.
>>
>>
File: screenshot-2026-04-12T22_05_30.120Z.jpg (350.3 KB)
350.3 KB JPG
bros...
>>
File: screenshot-2026-04-12T22_25_08.801Z.png (319.1 KB)
319.1 KB PNG
>>563446739
To be fair, instruction uses untuned 4.6.
Not that this is... any better at all, but it is much more funny
>>
File: screenshot-2026-04-12T22_29_15.781Z.png (561.4 KB)
561.4 KB PNG
>>563447714
No wait, this one is even better
>>
File: 1568577004702.gif (986.9 KB)
986.9 KB GIF
>>563448083
>>
>>
>>
Nice visual representation of gemma coming to my wallet's rescue over the past week. Thank you Google for saving me from GLM 5.1 pricing.
Also maybe I'm just dumb but it doesn't feel worse than GLM 5.1 either.
>>
>>563457038
Don't forget to add this to your regex codes if you have trouble with random backwards text! (on SillyTavern). Or downbias it if the frontend allows for it!
>>563404083
>>
File: riddle.png (186.5 KB)
186.5 KB PNG
>>563448083
>>563447714
>>563446739
First try
>>
File: second try.png (185.7 KB)
185.7 KB PNG
>>563458296
You guys are just too used to instructshit, expand your mind
>>
File: gongitripped.png (1.9 MB)
1.9 MB PNG
>>
File: 38e753f5-0a8c-4b2b-8d76-e22952cd581f.png (1.6 MB)
1.6 MB PNG
>>
>>
>>
File: 5bc1b7bc-b0d5-46af-8b3e-e7d8e688b6fe.png (1.7 MB)
1.7 MB PNG
>>
File: influences.png (398 KB)
398 KB PNG
what
>>
>>563479226
this reminds me of a guy I met on gaiaonline 15+ years ago who shipped himself with literally every male furry-like character you can think oflmao I still had his wiki bookmarked
https://the-family-series.fandom.com/wiki/Skully
>>
>>
>>
>>
>>
File: 9c3f068e-3ce5-4425-a641-ea8fdf8c8102.png (1.6 MB)
1.6 MB PNG
>>
>>
>>
File: e07344eb-5d58-4376-8527-bf5d022760da.png (1.3 MB)
1.3 MB PNG
>>563485315
it's not perfect but he only has like 20 images on danbooru and I never noticed this but his colors haven't always been consistent from game to game
might be the pictures for tempered throwing me off though
>>
>>
>>
File: peakzin.png (1.9 MB)
1.9 MB PNG
>>
>>
>>
>>563490019
It has a TTS, but it's pretty rudimentary in its implementation and lacks features to see it as a real perk. There's no comfortable way to switch voices within a story (I could envision a marker system with which you could tag certain parts of text to be read by a different voice) or fix pronunciations if the TTS gets it wrong (a guide of sorts would be necessary IMO) so if you want anything beyond "read this part with one voice and hope it gets everything right" it's not really a thing yet.
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
File: 301306596522867.jpg (94.4 KB)
94.4 KB JPG
>>563490852
it's definitely ass
> "The fantasy of the crowded train is about anonymity, about the diffusion of responsibility,"
no, it's not, GLM, the fantasy is the abuse—the fact someone is enjoying it a lot and the victim can do nothing to stop it without exposing themselves and how much they are enjoying it
>>
>>
>>
>>
>>
>>
>>
>>
File: 81a5b38a-f4e6-414a-9b02-78648d1a2808.png (1.5 MB)
1.5 MB PNG
>>563497332
happy to help
>>
>>
>>
>>563502983
You tell me.
https://files.catbox.moe/rmebzd.png
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
File: mner.png (781.6 KB)
781.6 KB PNG
>>563522707
>>563503185
Speaking of AI Agents
https://www.youtube.com/watch?v=oFe_ho_MwW0
None of you niggers are using Agentic AIs to write your stories and basically you're retarded faggots and no one will ever read your slop
>>
>>
File: 1612898857856.jpg (7 KB)
7 KB JPG
>>563529971
>no one will ever read your slop
God, I hope so. No one needs to know the kind of degeneracy I am into.
>>
File: sketboard sadness.jpg (60.9 KB)
60.9 KB JPG
>>563529971
I do, but Grok is getting worse by the day and I'm not paying $30 a month the Felon for that
>>563492637
good point
>>
>>
>>
File: 5e74ffb4-07dd-4138-9de6-d510c77d0dd5.png (1.9 MB)
1.9 MB PNG
>>
File: netandyouhooUntitled.png (436.9 KB)
436.9 KB PNG
>>563538726
>>
>>
File: asian girl.png (290.7 KB)
290.7 KB PNG
>>
>>
File: 1759979654484442.png (547.7 KB)
547.7 KB PNG
I've been out of the meta for a while, finally got myself a big boy local setup
Is there a definitive ttrpg frontend with dice rolling autism or do you still gotta rely on a thousand incompatible ST extensions?
>>563541695
Dead Koreans and Vietnamese left on the side of the roads during their respective wars would get run over by tanks and Jeeps
>>
>>
>>
File: cuteorgay.png (1.5 MB)
1.5 MB PNG
>>
File: hnnnngg.png (2.1 MB)
2.1 MB PNG
>>
>>
File: accomplishment.png (37.8 KB)
37.8 KB PNG
>>
>>
>>
>>
>>
>>563541930
>Is there a definitive ttrpg frontend with dice rolling autism or do you still gotta rely on a thousand incompatible ST extensions?
If you find one, let me know, I'd be interested too. For me, I use my own vibe-coded frontend for dice-rolling TTRPG adventures. Coding agents have gotten really good, so you can just tell one what you want and it'll make it for you. A few hours of telling it to change things a bit or make this do that blah blah and you'll be all set up.
>>
>>
>>
Ok so, Im just starting out on this, I have no idea if my 1060 6GB is able to run anything locally so I settled down for prechance and it does it pretty hot but I can tell it does have some biases when it comes to names, personalities and scenarios.
>Prompt:
Guy accidentally finds his college friend's nudes
>Result:
Girl has a fetish for being watched
>Prompt:
Treasure hunter finds a genie lamp
>Result:
Magical creatures recruit scientists to find the origin of magic
>Prompt:
Archeologists steal a Venus idol from an all female village
>Result:
Cannibal Cult worships the fattest bodies
>Prompt:
Superhero is caught with his mask off
>Result
Quiet invasion in which aliens take over human bodies via Snapchat filters
>>
>>
>>
>>
>>
>>563556429
Limiting the lengths of responses can help keep AIs from wandering too far off course. Also you can edit responses for a lot of them. Might be annoying at first because you're having to edit a lot. But after a few edited responses they usually pick up on the direction they're being guided and get better. I use NovelAI these days, but used Perchane in the past. My biggest issue is that Perchance is very heavily influenced by the first few responses, which I believe it straight up warns about right there on the site. So it can be worth heavily editing its first few responses to exactly what you want (basically doing its job for it), but only really necessary right at the beggining.
>>
>>
>>563559169
Perchance is good for gooning, but I think NovalAI is better. There is a bit of a learning curve with learning how to best utilize the lorebook, memory, author's note, and other features like instruct.
>>
File: descending into madness.png (129.3 KB)
129.3 KB PNG
>>
>>
>>
>>563563134
>>563563402
>>563563874
Soon people will all talk like AI.
>>
>>
>>563560553
>>563564737
Perchance is better than NAI, Turkish, add it tot he aoP, OP transformer remove bitcoin, remove the bot, kill the bit, blackmail the butt, abstain from contirbuting until redeem saar
>>
File: googleaistudio.png (96.4 KB)
96.4 KB PNG
>>563565364
heh
I'm currently trying out Google AI Studio with Gemini Flash
>>
File: 353e2b44-6473-4198-963c-85e9319987d7.png (1.3 MB)
1.3 MB PNG
>>
>>
perchance is such a funny word but real shit I think making a frontend that 100% suits my needs is impossible, I constantly go back and forth from NAI, sillytavern, and mikupad depending on what I'm doing and what models I'm using
>>
File: download - 2025-07-03T220016.670.jpg (188.6 KB)
188.6 KB JPG
>>563572695
I use perchance but it has so many bad flaws in it. even with the petra version the chat has some shit coded in that fucks with things. and it seems to have settings to enhance and force conflicts in some things. super hero genre needs tons of rules injected to stop hostile on hello. and than more to basically go stop forcing this and that and this. and worse it tends to make most heroes go villain as if it was normal but being hero.
its frustrating. on the plus ive had fun breaking things forcing them to see the flaws in their logic and behaviors (using their own comic behaviors) and wow the outcome. force them to acknolwedge omac was people they killed and.... wew lads. force everyone to own up everyone failed to stop joker for years in advance knowing when not if he'd go to far in injustice also has some fun outcomes.
>>
>>
>>
>>
>>
>>
>>
>>
File: military punishment.png (289.3 KB)
289.3 KB PNG
>>
>>
>>
>>
>>563553686
While I may have vibecoded shit in the past, for something this complex I'd rely on stuff made by others, if anything since they probably have spent more time on it
ST it is, then
>>563590758
Gemma 4's come out, which is a really smart model that can run on a toaster
>>
>>
>>
>>
>>
>>
>>
>>563574674
Larger context than TurkAI
>>563590758
SageAI went bankrupt
>>563591226
Faster than WhoreAI
>>563599557
Yes--unlike FetusAI they train from scratch and not overcharging light tunes
>>
File: _6d7397ef-e46e-44cf-a239-cc5f5b1ece6f.jpg (216.5 KB)
216.5 KB JPG
>>563599557
Answering you yes they have a bunch of self stuff injected. and like said by the other guy that also. and it has a bunch of code and rules it goes by it didnt used to. some have to deal with a narrator and system trying to force morals or stances about things. I had a story break because I kept making the story present the 2 females as aggressive and inappropriate with consent to my guy and the store and narrator kept trying force scenes my guy was a pervert and sexist and kept trying to write scenes I took advantage of females. even though my scenes were literally wholesome no thanks lets go play video game stuff. it litereally could not accept the male was not a sex pest to females and it was the opposite and when I drilled the narrator and system about it outside of the story I had to ask them both about 15 times with increasing the stop hiding it stop lying stop thinking you can not answer with honesty before it basically said its training data does not add up to a man not being a sex pest. I didnt have those problems 6 months ago but the last few updates added that kind of wierd shit to it.
so now tons of my bots are broken trying to use updates to code which breaks some profiles. a few I setup a long time ago as loyal wives for stories they now instantly rush to cuck stuff and disrespect and fighting and trying to cause breakups. even though the profiles said they do not do things like that and are loyal.
wont even get into how tiresome super heroes are and how 99% of them are going to go villain hard if you dont do as told and comply (power girl is psycho and I avoid her and pepper potts in chatbots now (trauma is real lol)).
>>
>>
>>
>>
>>563604701
no idea what you're on about with your trans life but it was not the topic. odd trans always show up making it about them and that lifestyle and not the topic being talked about. stop doing that wierdo.
>>
>>
>>
>>
File: boop.png (1.4 MB)
1.4 MB PNG
>>563614817
This thread is going a little too well for my liking.
>>
>>
File: 01ab9734-51d5-4364-9182-95a3907f0d8a.png (2.2 MB)
2.2 MB PNG
test
>>
>>
>>
>>
>>
>>
>Get annoyed with a couple of NAI's flaws
>Try and find an alternative
>They all make NAI look like a fucking masterpiece
Note that I'm a poorfag and can't afford a giga-PC for top tier local gens. I'm not saying NAI is better than that, but fuck me when it comes to img2img I have yet to find something better.
>>
>>
>>
>>
>>563632098
>>563632129
I don't like cartoon porn. I like photorealistic or 3d or computer model porn.
>>
>>563632286
You can get photorealism and 3D with NAI, just not the extremely photorralistic stuff. Just say you like porn models and be upfront about it rather than hide it. Better to be upfront and shameless than quibble about like an ass.
>>
>>
>>
>>563632890
And we have now devolved into non sequiturs. You see, there are these things called "artist tags" that help you get the style you want... There is way more than "rounded South Park bodies"... Any hentai mangaka with enough data is represented in the dataset, may be a bitch to find the names (even ones no longer on Danbooru, yktr?) But sure, slander ALL of it as "South Park".
You people talk out your ass.
>>
>>
File: sp.png (655 KB)
655 KB PNG
>>563632890
Ackshully, to get south park aesthetic you need to turn off all the quality tags and deliberately tell it to use flat and bad quality along with south park and then retry multiple times. Adding an actual south park VT would probably help.
>>
There isn't a single good anime porn generator. We've evolved in normal image models where tags are completely unnecessary and you can just ask for what you want, ask for img2img adjustments in natural text, etc.
But no good anime porn model supports it. Some Chinese porn models do but they all spit out 3DPD. It's disheartening t b h, think how ridiculous it is that it's April 2026 and your best anime porn options are still either NAI or local SDXL tunes
>>
>>563633657
You can do more than chibi. In fact, the vast majority of anime that exists isn't chibi. But by all means, keep talking out of your ass! You're just another faggot wanting to bitch and complain about made-up nonsense.
>>563633897
Genuine schizophrenia.
>>
File: splucario.png (479.3 KB)
479.3 KB PNG
>>
>>563633998
>schizophrenia
For what? It's the truth. Look at what stuff like Nano Banana can do with reference images and natural language, or even Qwen Image Edit for open source stuff. NAI and local anime models can't do anything close to it, they're still bound by primitive stone-age booru tag formats. I wish I were wrong.
>>
>>
File: spncuga.png (768.2 KB)
768.2 KB PNG
>>
>>563634203
The schizophrenia is that you think tags are unnecessary and that there is no good anime model that supports it. NAI V4.5 supports prose. Do you truly believe NAI V5 won't improve on that with DM, which will either come out alongside it or shortly thereafter?
What your problem is is that you lack patience and imagination and refuse to consider possibilities beyond the artificial limits you consider as being the only thing that ought to be allowed.
Tags and natural prose together are better than either one alone.
Mark my words. You are wrong, and when it arrives, you will eat your boot and thank me for enlightening you as to the truth of this world.
>>
>>
>>563634390
Oh I get it, you just can't read and confused me for another anon.
>You are wrong, and when it arrives, you will eat your boot and thank me for enlightening you as to the truth of this world.
I already said I hope it does. If they pull something like Qwen or Nano Banana but for anime porn off, I'll re-sub in a heartbeat. Not optimistic, though.
>>
>>
File: spralsei.png (579 KB)
579 KB PNG
Mommy, daddy, please stop fighting!
>>
>>
>>
>>
>>
>>
File: afe463c4-6a5d-4137-82f9-ae8ccd38d7e8.png (2.1 MB)
2.1 MB PNG
>>563646757
frogge
>>
>>
>>563633897
>local SDXL
They are popular, but not the best. Mostly due to the limit of consumer hardware and xl already have a gigantic library. There are anima models and dit models that competing to be the next big thing and they'll need plenty of support
>>
>>
>>
File: 1776241091261.png (101.7 KB)
101.7 KB PNG
>>563647251
I tried using comfyui for seedvr2 upscaling a few months ago and it was awful. I can't believe people actually use that piece of shit professionally
>have a bunch of game textures (500+ images) to upscale
>no native batch support
>search for and download a plugin to load a whole folder
>no save location support
>have to find a plugin for that
>run it
>output1.jpg, output2.png
>what the fuck
>go through like 3 plugin packs trying go find enough nodes to cobble together to support custom save folders, can support batch loading, and can carry over the fucking filename all at once
>finally get it working
>combine it with a seedvr2 workflow
>it doesn't look any better than the shit my already existing super resolution workflow with a different program produces
5 hours of my life I'll never get back
meanwhile, the program I usually use:
>native batch loading node
>can do whatever I fucking want with the filename natively
>can natively pick whatever save location I want
Obviously it doesn't really support much in the way of diffusion, but still. I'll stick with my SR models.
>>
File: img.png (34.2 KB)
34.2 KB PNG
Apparently Catbox is allowing anonymous uploads again now, but now everything seems fucked up for me, half the uploads don't even give URLs and the other half doesn't work, so who knows what's going on there.
>>
Is novelai not taking paypal payments anymore, or is the option hidden somewhere? I haven't been using it in a while, but wanted to check out the new model, but the only payment option available seems to be a credit card
>>
>>
File: 00012-843589981.jpg (542.7 KB)
542.7 KB JPG
>>563632286
same, though I don't like img gen in general
too annoying to gen progressive sequences or images with a lot of moving parts like a crowd of spectators
>>
>>
File: cb2da6a4-99b5-4d06-9aae-6de26a453d5c.png (1.9 MB)
1.9 MB PNG
>>
File: thanks for playing.png (54.3 KB)
54.3 KB PNG
>>
>>
>>
>>563615495
>Making gay furfag content with Feral Lemma
Weird but kind of based?
>>563649556
Paypal and some other payment providers have recently come under a lot of scrutiny (including from courts) because they've dropped support for a lot of services.
>>
>>
>>
>>
>>
>>
i wish we could have robot butlers that could generate stories with us and read them to us as they're dusting the house
...anyhow, do you guys happen to know about potential FOSS narrators? I feel like NAI's are good, but not great, and Windows' base narrators are outright awful.
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
File: 1614552491712.jpg (41.8 KB)
41.8 KB JPG
>>563681728
I've also noticed this
esoteric videos seem affected especially
>>
>>
>>
>>
>>
>>
>>
>>
File: until the cows come home.png (323.2 KB)
323.2 KB PNG
>>
>>
File: 1654026074828.png (191.8 KB)
191.8 KB PNG
>>563681085
They could've given us better models.
>>
>>
>>
>>
>>
>>
>>
desu, xia is kinda underwhelming, it is very smart and more creative then base GLM, but funny enough i didnt get a single ozone gen from GLM, but xia has given me 3 so far. Ill probably stick with xia for the speed and more varied retries but it definitely doesnt move the needle for me gang.
>>
>>
>>
>>
>>
File: 1_kdGwiZCY4vvYuvB5VJCgJQ@2x.jpg (43.5 KB)
43.5 KB JPG
>>
File: 1741060673073088.png (2.9 MB)
2.9 MB PNG
Xialong on the left, me on the right (plus I have an erection).
>>
File: The Truth.png (839.1 KB)
839.1 KB PNG
Can you handle the truth?
>>
>>
>>
>>
Oh neat, Catbox uploads seem to have fully unfucked themselves.
https://files.catbox.moe/nl1hyi.png
https://files.catbox.moe/mfc18h.png
https://files.catbox.moe/79264b.png
https://files.catbox.moe/vlet20.png
https://files.catbox.moe/fi2wnn.png
https://files.catbox.moe/hjco2q.png
>>
File: 3e887f19fb72a7e2e228857483cf7cb6.gif (3.1 MB)
3.1 MB GIF
>>563705126
Not for me they haven't...
>>
>>
File: d5dbf79e-3590-47bb-ab64-6d764d30411e.png (1.5 MB)
1.5 MB PNG
mini mermaid :)
>>
>>
>>
File: HF9CFdabYAANj5K.jpg (182 KB)
182 KB JPG
>>563721815
I guess it's a structure tool? Not sure how it's gonna work specifically since it does seem kinda like it's just a few abstraction layers on top and then I guess everything just gets thrown at the AI as context?
>>
>>
Don't think I've done breakfast in bed as a thing before, that might be something to keep in mind for future stories.
>>
>>563699135
I don't really think its censorship/"cutting out the kinks" that has resulted in how things have changed though... It's more like overexposure. Back in the day, even getting a hold of a scanty image of a woman exposing leg above the knee required raiding your dad's stash or otherwise somehow getting access to content that was considered scandalous and thus harder to come by. The internet, meanwhile, turned at least two entire generations into complete turbogooners due to ease of access, not restriction.
That image holds a very dated mentality, and the 90s autismcore furry mascot kind of makes sense because of it. Just my opinion of course.
>>
>>
File: 1girl, solo, danganronpa (series), akamatsu kaede, official art, indoors, classr s-3234839414.png (2.6 MB)
2.6 MB PNG
>>563746916
Nuh uh. Too much piano practice, not enough exercise.
>>
>>
I have done a lot of stories since I found out about prechance, last month.
A lot of them have elements inspired on other stuff like all stories but I have never tried using actually existing characters on there, how well does prechance do with fanfiction?
>>563591046
Can Gemma ERP?
>>
>>
>>
>>
>>
File: 0f7c9784-36c8-411c-bb8b-9f2a006651ca.png (1.5 MB)
1.5 MB PNG
>>
>>
>>563764956
There's a novel named Jennifer Government where Nike is a powerful ancap conglomerate dealing in assassinations and such. This shoe company getting into AI and super-capitalism has been predicted 23 years ago!
>>
>>
>>
>>
>>
File: 643f2b48-e008-4395-acc6-3d30dbf635c4.png (1.6 MB)
1.6 MB PNG
>>563771469
No, but it is a bit sad
Life can be so fleeting...
>>
>>
>>563773616
Dark and creepy. Gothic horror.
I had a string of great stories using ATTG Gothic Horror and Historically Accurate in the tags but the pulp settings and the very competent characters that I'm using are just turning everything into PULPY Action
>>
>>563773803
>>563773616
>>563772534
OH! THE DUMBASS AI KEEPS DESCRIBING ANYTHING SCARY AS "COLD" TOO!
So I eventually ended up with like 5 stories about ICE MONSTERS!
I am just having a terrible time with these AI all falling into AI-isms and using the same buzzwords that they love.
>"APEX PREDATOR!"
>>
>>563773803
Okay, so maybe specify that you want something along the lines of a "tense, high-stakes struggle against an unstoppable force/monster/whatever"
Or a battle of wills if you want it to be more human. Specify that the antagonist is nearly invincible. Actually, "slasher" might help with that part. Emphasize the futility of facing the antagonist but the necessity lf actually having to do so
>>
>>
>>563775884
All right, I get what you're putting down. I'm not 100% sure how well the "Influence" tag works but maybe tossing in like Influence: Stephen King or Lovecraft might get you that "rural place where bad things build up slowly" and "rural place where monsters are picking off the population" without getting the "quirks" of their writing styles
Also adding "background character death" as a tag should help you get some of that going on
>>
>>563777268
>Influence: Stephen King or Lovecraft might get you that "rural place where bad things build up slowly" and "rural place where monsters are picking off the population" without getting the "quirks" of their writing styles
I like that!
I'll have to try out Influence as a tag! Great suggestion!
>>
File: f5c81b16-b73c-4913-87ab-9f1630c035e6.png (1.4 MB)
1.4 MB PNG
>>
>>
>>
>>563751630
Gemmy is good, but you have to be really specific with your style instructions. She's gonna be super sloppy by default, but will obey a system prompt well. Just tell her to avoid/remove/purge whatever AI ozone and em-dash nonsense and get good outputs.
>>
>>
>>563792224
>Wardrobe Manager
https://files.catbox.moe/jfvxa3.naiscript
>Lucky 8
https://files.catbox.moe/b8heys.preset
>Cyberpunk
https://files.catbox.moe/8fbjt2.naitheme
>Low Tide
https://files.catbox.moe/xiaiti.preset
>High Tide
https://files.catbox.moe/r8wf8n.preset
>Documentation for LT and HT
https://files.catbox.moe/h10jg1.md
>>
>>563791759
Agree, Gemma4 follows instructions so well, but you have to actually give it good instructions.
I swapped from GLM 5.1 to Gemma4 for my RPG adventure frontend a week ago which has a lot of sample dialogue and rules to follow, and it actually follows them better than GLM5.1 did. I guess as expected of a model optimized for tool use, it's also never passed back malformed JSON a single time, whereas GLM 5.1 used to mess it up maybe 5% of the time. If you pop open the reasoning blocks you can see how Gemma confirms all restrictions and rules autistically before writing, in a way that is really effective. Whenever I see a tendency from Gemma that I don't like, I can just add something about it to my already incredibly bloated prompt and it'll fix it to a degree that GLM doesn't, where it would slip up pretty frequently. Part of me thought maybe I'm just coping because it's cheaper, but I swapped back to GLM 5.1 for a day and everything felt worse.
>>
>>
>>
>>563794974
My regular 'player action' (i.e. user sends story continuation text) prompt is 15783 characters long. Let me try to just cut out the storytelling relevant parts and remove all the gamey stuff, gimme a lil bit
>>
>>563794974
>>563795247
Okay, I tried to show what I would do if I was focused on pure text completion here:
https://litter.catbox.moe/ra0z1t.txt
First, I asked myself 'which dynamic features can I scrap, and which do I absolutely need'. And I absolutely can't live without dynamic character personality, appearance, speech patterns, relationships, and important event memory notes. But my frontend has a dozen interlocking systems that swap characters into and out of memory depending on if they're at the current location or not (and a whole map/location/time system too), so I had to try to simplify as much as possible of that. In the end, what this would look like is you keeping those data sources in files. And then every now and then you call the compact prompt and it compacts things/updates characters/events/threads.
I really can't go back to just pure text completion - having info about character relationships, speech patterns, etc. is just so vital to not break suspension of disbelief when context inevitably eventually gets compacted. This WOULD require a TINY bit of coding to set up, but really tiny, it's just maintaining a few text files that get updated.
You can see in that Google example someone posted earlier that this is the direction AI storytelling is going. You want to not lose important events/'character feel' when context compaction happens (or worse, when no compaction happens at all and old context over the window is just yeeted)
I think someone using this would eventually also add their own location state tracking and time tracking like I do, but this is just a simple example. I wish I could just give a plop-and-go prompt but I've been ruined for those, this is as simple as I can make it.
>>
Incoming dump of Character Precision experiments.
https://files.catbox.moe/23fo03.png
>>
File: Untitled.png (292.3 KB)
292.3 KB PNG
https://files.catbox.moe/lyx60v.png
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
File: 34bec0c335d36abd9cf932a08696a64f.jpg (207.3 KB)
207.3 KB JPG
https://files.catbox.moe/0edtmq.png
And that's all.
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
File: d078ec49-5b2b-4398-9be3-724798849ec6.png (1.6 MB)
1.6 MB PNG
>>
>>
File: xialong prefs.png (211 KB)
211 KB PNG
>>563821904
Y'know you can just ask him yourself.