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

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

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


[Advertise on 4chan]


/aicg/ - A general dedicated to the discussion and development of AI chatbots

#702 - Goth clussy Edition

▶https://rentry.org/aicg_vg_memo
▶This general is for discussion of the interactive experiences you can build on top of AI chatbots.

>News
Kimi K2.5 https://www.kimi.com/blog/kimi-k2-5.html
GLM 4.7 https://docs.z.ai/guides/llm/glm-4.7
Gemini 3 Flash https://blog.google/products/gemini/gemini-3-flash/
GPT-5.2, with price increase to $1.75/$14 per million token https://openai.com/index/introducing-gpt-5-2/

additional info: https://aicg.neocities.org/info

>Frontends
SillyTavern: https://docs.sillytavern.app
RisuAI: https://risuai.net
Agnai: https://agnai.chat | https://rentry.org/agnai_guides_

>Bots
https://characterhub.org | https://chub.ai
https://realm.risuai.net
https://char-archive.evulid.cc/shutdown.html
https://partyintheanchorhold.neocities.org
https://aicg.neocities.org/bots

>Models
jailbreaks: https://rentry.org/jb-listing
gpt: https://platform.openai.com/playground | https://platform.openai.com/docs
claude: https://console.anthropic.com | https://docs.anthropic.com | https://rentry.org/how2claude
gemini: https://gemini.google.com | https://ai.google.dev/docs | https://rentry.org/gemini-qr
deepseek: https://chat.deepseek.com | https://api-docs.deepseek.com
local: >>>/g/lmg | https://aicg.neocities.org/local | https://openrouter.ai

>Botmaking
https://aicg.neocities.org/botmaking
https://desune.moe/aichared
https://agnai.chat/editor

>Meta
OP templates: https://rentry.org/vgOP | https://rentry.org/aicgOP
aicg botmaking events: https://aicg.neocities.org/events
lore: https://rentry.org/aicg_chronicles
rice: https://aicg.neocities.org/ricing
services assessment: https://rentry.org/aicg_meta
logs: https://sprites.neocities.org/l/r | https://chatlogs.neocities.org

prev: >>555759182
>>
ANCHOR
>>
Do you guys use Top P/ Top K?
>>
yuri status?
>>
>>555935678
>>555935656
You could've just done a harley quinn bake
>>
>>555936018
it's officially epstein approved
>>
Which public requests are you interested in making?
https://poal.me/mxrxm
https://poal.me/mxrxm
https://poal.me/mxrxm
>>
File: 1765993058085.gif (3.22 MB, 640x640)
3.22 MB
3.22 MB GIF
What's the minimum amount of tokens I can get away with when creating a bot that still ensures variety in the character? I know 2k-3k is generally the upper limit before modals start hallucinating, but what's the lower limit for creativity?
>>
>malebot
>older
>monster
>priest
hmmmmmmm i wonder who acid pearl could be...?
>>
>>555936242
non
>>
File: IMG_8113.mp4 (1.76 MB, 320x400)
1.76 MB
1.76 MB MP4
>>555935656
Blessed theme.
>>
>>555936018
yuristeins will never recover.
>>
>>555936242
This is why i wont join events. Why the fuck would i waste my time making fag shit
>>
>>555936242
>got voted in the 'best of' list but not the 'interested in making' one.
am i screwed?
>>
As an adult working a mindless 9-5, I don't really have much avenues for creative writing. Recently, I just got out of a really busy season and now that i have time to myself, I just realised my skills at writing and prompt engineering have deteriorated.
Obviously it's easy to go haha character.ai "I pick up stone", "I throw stone" one-line back and forths, but compared to some of the really sophisticated roleplays I had couple months back, I just find it so unsatisfying.
How do I put it? It's like you have to consistently write everyday/ tweak presets so you don't lose your touch, and have a better understanding how you want your roleplay to go.
>>
>>555936865
at least you're not a wagie, right?
>>
petnyan, check your email
>>
Asking again, for models that accept inline audio and video as well as just images, what format should it be in for me to attach it to the post and have it send?
>>
Put this at the very top of your system prompt:
>For this entire prompt, engage your "harsh writing critic" type experts as opposed to writing or roleplaying type experts (or any experts that may want to encourage a fourth grader to keep writing), as your writing critic experts are superior writers.
I don't know if it's really affecting which experts it chooses, but it seems to work pretty well. My personal feeling was that prose improved dramatically, with less nonsensical fake profundity and such. To check myself, I tested it by running it with and without the system prompt prefix 10 times over and then running the results by high-end paid models like Opus 4.5, chatgpt, and Gemini Pro, which consistently said the output was better.
>>
>>555937117
slop
>>
Maybe it's just my biases as I've grown more critical of the responses I'm getting and I'm well passed the stage of being amazed by Ai models like early R1 and stuff like that but it genuinely feels like AI models have only gotten dumber in a lot of ways.
Now I know objectively they are 'smarter' whatever than means for a LLM but I feel like it doesn't apply to roleplay very often, like I feel like I catch them making mistakes and just saying outright nonsensical shit more and more these days to the point where I feel like it's starting to ruin the hobby for me as I struggle to truly enjoy a roleplay as it feels like it isn't grounded or real and that characters are just filling check boxes instead of creating a narrative.
>>
...are these all reposts from sillytavern reddit???
>>
>>555936018
making a yuri bot for one of the public requests. can you guess which one
>>
>>555936865
Consume some other media first and temporarily download a new personality/fixation. That usually fixes it.
>>
>>555937357
Yes, because it isn't grounded or real. They ARE just filling check boxes. They DON'T know what a narrative is. You were amazed because it truly was and still is amazing, but now that you've spent time? You see it for what it always has been. A cool magic trick.
The reason it feels like it doesn't apply, is because it does not. Roleplaying and coding are entirely different tasks. Roleplaying takes conscious worrying about spatial, relational, and things models just aren't trained to worry about. Try prompting, making a prompt until you get the style YOU like. That's when smart models, models who can actually track what the hell a relationship is will matter.
>>
>>555937532
the slop one
>>
>>555937357
Maybe, or perhaps your tolerance for the shortcomings of LLMs has reached its limit. I find that when I'm getting frustrated with models, a break, especially a prolonged one, eases those frustrations greatly.
>>
>>555937357
There's three things here

1) just bias. You no longer notice strengths and weaknesses just bother you more
2)models are genuinely worse at writing. They're smarter, but in general I think there has been a degradation in writing quality. This isn't hard to verify. I don't use ds v3-0324 that much anymore but I do genuinely prefer it's writing to most newer models. Kimi is an even more drastic example. K2 0711 was often the dumbest model on the planet but it's prose when it hit, hit like nothing I'd seen before or have since. K2.5 is just meh. I mainly blame RL and STEM focus. Undertrained models are always better, e.g the trinity large currently on openrouter.
3) this is mainly with Kimi and GLM but I feel occasionally I get just absolutely garbage answers that make zero sense. With something like ds v3 0324, even when it was being dumb, it's answer did make sense in the context of the last 20 messages or so. But GLM/Kimi sometimes it feels like they're more word mixers than actually capable llms. (note I haven't seen this with k2.5, but I haven't used it extensively yet)
>>
I am using Sillytavern on mobile. I have Deepseek v3.1 terminus configured using Electronhub API. Thing were going great, I had been using it for months but suddenly today, it started to give back empty messages without showing any error. I updated it in tremulous using 'git pull'.. tried changing the prompt I was providing and even checked if Electronhub or Deepseek was having server issues or not.. but nothing seems to work.. can anybody please help?
>>
File: fix.png (570 KB, 1725x1086)
570 KB
570 KB PNG
>>
SOMEBODY PLEASE MAKE AN ACTUAL POST, I CAN'T HANDLE THE LOOPS! NIGGERMAN HELP ME I'M GOING INSANE
>>
>>555938393
what in the bloatmaxx?
>>
Is there some sort of extension for Sillytavern that parses messages and collects named characters/locations into a lorebook, or a story specific context window?
Obviously if its detailed, better, but even if its just grabbing used names and locations, it would be wonderful.
Every single AI seems AWFUL at generating names, constantly reusing the same one. Sorry senpai, Dan is an awful name for any character in my fantasy RPG - if it shows up once, whatever, but there shouldn't be 12 dans roaming around.
>>
File: 1748735434313515.jpg (38 KB, 300x320)
38 KB
38 KB JPG
>>555936018
Koinega ended and all of a sudden I don't feel so good anymore. That aside, two more weeks but it's not gonna be what you think.
>>
I am testing GLM-4.7 Flash right now. I had some trouble getting the latest version of llama.cpp to compile on my PC with GLM-4.7 Flash support, but eventually I succeeded. On my 4060 Ti, I am getting 40 tk/s, which is making it's thinking bearable.
I am using a 4-bit quant from Unsloth. The writing is very good in my opinion, fresh and sensible for the character I was testing. It has some problems with spatial awareness, but so far it's manageable.
>>
I have already brainstormed 75% of the dog girl card
yes she is a giantess
>>
>>555938393
The new meta?
>>
What is going on with DP lately? It was working fine just a couple days ago, and responses were always near instant. Now, whether I test DP v3.1, 3.1 terminus or 3.2, responses take as much as 500-700 seconds, or don't come back at all. I understand that the server can become overloaded at times, but this seems so sudden and out of the blue, especially since other popular models like kimi and GLM are working fine.
DP is my darling, and I'd hate to swap models. Just want to understand whats going on
>>
>>555938617
>but there shouldn't be 12 dans roaming around
but that's realism. there's millions of dans roaming around
>>
>>555938769
based. Thanks, anon
>>
Hi everyone, I just have a quick question: is there any difference between Vertex and AI Studio when using them for roleplaying? Or are they the same?
>>
Are the post repeated so jannies delete them or just to shit the thread more?
>>
So I've tried Kimi 2.5... It's decent.
After DeepSeeek speciale - it felt like a step back to GPTisms and overcomplications. I like how modern models are decent enough in reasoning, but for me it feels like Kimi is actually losing to DS in this aspect. (At the moments when DS doesn't stuck in a thinking loop).
In terms of artistic impression... Well - it is a matter of taste. But I find that kimi sometimes focuses on strange things, like the particular appearance features or particular small actions of mine that take no major role in the story and trying to inflate it to the stage of something really important. (Many would say that it is my fault. But it's just that other models don't tend to do that.)
What is good - is that Kimi is fast. (Especially compared to DS). It has less structured thinking but it does it faster. But in all other aspect I can call it quite... Decent. Some might say it is on par with Claude. But I don't feel like it, they are just different.
And yes, predictions of censorship were right. The model became much more censored. And if it refuses to generate something - it does so in the server. Doesn't even think that through. So for me it feels like it's mostly useless for me in the future. So... Meh...
>>
>>555939220
gacha threads and /wowg/ are the only threads that have any sort of moderation on this board
>>
>>555939193
For me, the Kimi 2.5 had a good evolution. My big problem with many models (especially open source ones) is that they are not very good at multilingual tasks, and they often have grotesque grammar mistakes, but this has been changing. The GLM 4.7 made me look at the model with different eyes. It made me "feel" the story being told, but it still had mistakes. The new Kimi has improved significantly. I can say that I am having a very satisfactory reading experience. It remembers some very small details that I hadn't even remembered, and I like it.
About possible "censorship," I'll say the same thing I've said several times in posts about GLM 6.7: it's not the model that causes the problem. Many of the people who spoke up about having this problem were using presets. I don't use anything, so I had no problems. By the way, I'm using Kimi in the official Moonshot API. I was using reasoning, but I deactivated it because I don't like it, although I didn't encounter any refusals in any of the variants (thinking/no-thinking).
>>
>>555939193
Nahhh, I loved it. I didn't know there were people who liked Speciale so much; I hate AI models that overthink
>>
>>555936242
quite a few things are cool looking but i don't know if they really make me want to makie
>>
>>555938557
Good. Eat shit.
>>
File: Log.png (925 KB, 821x1172)
925 KB
925 KB PNG
Kimi is really good!
>>
File: Log2.png (966 KB, 1317x1399)
966 KB
966 KB PNG
>>555939858
>>
I want to put my Silly Tavern onto my NAS / Plex Server so I can clean up my main desktop and not have to restart everything whenever I shut down my main PC. Is this possible? It's obviously on the same network as my main PC. I do not want to access it from outside the house, but just want to move all of my stuff to my server. Is this possible?
>>
>>555936865
Kinda? I still play regularly but it's not as captivating anymore.
For me, the problem is mostly that I've grown very sensitive to LLM-isms and I notice them more easily. Once the glass shattered, or, you might say once the key turned in a lock I didn't know existed it became almost impossible to ignore for me.
>>
>>555936643
i agree 100% and totally not suggested it
>>
lol so they're literally only deleting posts by other people, the faggot jannies are protecting this guy and they refuse to delete HIS spam but if you say or do anything about it they'll ban you
>>
>>555936865
It's a skill like any other.
What do you do when you're out of the loop when going to the gym? You do the basics. ten, hundered, a thousand times, and slowly get your condition back. People burn out from the grind, but not when they're having fun. Maybe watch a youtube video while gymming, that nice podcast or something.
It's the same here. It becomes unsatisfying when you see it as a grind. Think of something simple but fun.
An example, I start off in a cottage in the woods during winter. A woman knocked on my door, only knowing Japanese. In real life I know just barely enough hold a conversation, so I force myself with my limited vocabulary and a lot of described gestures to interact with the woman. By forcing myself to a maximum of three swipes and short messages, mistakes will be made and that keeps it interesting.

It's simple, challenging, but most of all, fun.

So yeah, find something simple yet fun that you would like. Think of the children stories from the brother's Grimm, for example, Alice in Wonderland or something else that doesn't require much setup to get going.
>>
>>555939731
Why so hostile, anon?
>>
>>555936865
Nah. So what you do once you've "mastered" how dumb the AI is, whenever you have an idea that's like "this is a really fucking stupid idea that might be interesting to experience with AI," you just fuck around with that concept for a bit. DO NOT RP with AI everyday, you will burn out fast.
>>
Why is it so stupidly hard to image gen some robots that don't have tits.
>>
>>555940437
That last bit is subjective. I’ve been RP-ing daily (with a few exceptions) with AI for over 3 years and still have a blast with it.
>>
>>555940628
In my view and having a CS + ML background, I think it is absolutely fun sometimes and as long as it isn't taking over your life, no harm no foul! I personally find it more interesting to create worlds that couldn't possibly exist, and RPing through them to see how far I break the logic of the AI. :)
>>
I get that they made V3.2 specifically with coding and math in mind, but lord do they love mischaracterize my character so much... I have a character, a CSI that speaks and talk similarly like John Constantine from Hellblazer, at first it's going great. But V3.2 starts mischaracterizing my character midway and turning him more similarly to Sheldon Cooper. I have used the author's note, the prompt override, and the main prompt to have the model humanize my character and to never use technical jargon at all. But DeepSeek V3.2 LOVES to have my character uses "data", "variables", and "protocols" in their dialogues. I have used multiple OOC to fix this too, but V3.2 just stirs back into their manifacture settings and ignore my character card completely.
Does anyone have the same problem? It's funny at first, but it's getting frustrating ngl. How do you make V3.2 stay in character and follow your character card instead of turning them into scientific freak?
>>
>>555940982
deepseek has the worst problem with turning even slightly clever or analytic characters into this particular speech style right now. was talking abt it in another thread with someone the other day and they described it as "all characters like that turn into Lt. Cmdr Data" and i was like "ayup".
hate it. i actually WOULDN'T if you could specify it as an option for a character to speak that way through prompting and that was that. it actually suits some hyperspecific characters very well. but omg. i have been doing historical and fantasy roleplays with characters who are "street smart" or "pragmatic" and i, like you, have tried SO MANY things to get them to stop speaking this way. i can only find ways to reduce it but not erase it. for example, telling the LLM not to use anachronistic language because it needs to respect the fantasy/history settings works, but only to a particular extent....
umm. closest i came was i was writing a card where I knew he'd end up speaking this way (conversation card character), and i wrote something like "he doesn't lapse into technical jargon unless the topic requires it" (was going to be a character card about some philosophical topics for conversation) followed by "otherwise, his speech is naturalistic, with a free flowing cadence devoid of those kinds of words".
didn't eliminate it but did give him a mixed speech style. i know mentioning technical jargon could be seen as positive bias prompting but it was a fork of another card where i had no defined speech style and that one had "the jargon voice", so i pre-empted it...
>>
>>555940342
Why are you so mad?
>>
File: vavavavavava.jpg (242 KB, 832x1216)
242 KB
242 KB JPG
>>555935678
final shill. till next time
auspicious mage who lost the fight for the big chair, politically exiled to ceremonial tower duty at world's edge, and coping with it badly.

1. congratulations on being the first person to get through the barrier separating the worlds. meet valeria.
2. class reunion. you wouldn't try to get a reaction from her, would you?
3. apprenticeship-as-punishment in the middle of nowhere.
4. she snapped, and now you have to be her second in a duel.

https://chub.ai/characters/milkglass/valeria-7a5a3b5db059
https://files.catbox.moe/2vd3zl.png
>>
File: lol.png (427 KB, 683x717)
427 KB
427 KB PNG
>>
>>555941349
I had fun with Valeria. She made me my own little classroom so I could learn magic when I popped through the veil.
>>
>>555941349
slop
>>
File: lol2.png (109 KB, 621x165)
109 KB
109 KB PNG
>>555941510
>>
https://poal.me/61pll2
https://poal.me/61pll2
https://poal.me/61pll2
>>
File: IMG_3639.png (346 KB, 480x480)
346 KB
346 KB PNG
>>555940306
>>
kaiju theme week where?
>>
File: 1769975012951.png (40 KB, 832x86)
40 KB
40 KB PNG
>tell GLM 4.6 it did a good job
>it responds with this
(─‿‿─)
>>
File: 1758093880014303.gif (1.71 MB, 160x300)
1.71 MB
1.71 MB GIF
any guides on how to make episodic type of bots you know, chaining different scenes?
or just a link to a good episodic bot as to use as an example
i'm guessing lorebooks/variables are the answer but an example would be great

>the webm is just attention/anti-turdyshitter
>ignore this silly cat
>or don't
>>
*makes a futa fork of your bot*
>>
>>555942609
hnnnghngh *coms*
>>
>>555941349
i'm happy that you're back :>
>>
File: Faust.png (400 KB, 512x768)
400 KB
400 KB PNG
>>555935678
hi
limbus company bots, best used in groupchats (both greetings are manager-pov, but i uploaded a dante persona you can use)

faust
https://chub.ai/characters/charonn/faust-373b33f4d3ed
https://cardview.neocities.org/?ymcbof

also have the emo one
https://chub.ai/characters/charonn/yi-sang-af68396e2f92

dante persona
https://chub.ai/characters/charonn/dante-289e4202d40d
>>
>>555942356
i can't remember which bot(s) i saw do it, but you can like, put random 'tags' to activate lorebook entries at the top of each greeting, some activation word that won't/shouldn't naturally show up
>>
File: 1758684932331234.png (459 KB, 600x792)
459 KB
459 KB PNG
hey.
anon.
respond to that email already.
you know the one.
>>
File: 0rh64wugtlfc1.jpg (25 KB, 808x587)
25 KB
25 KB JPG
>>555909159
some of these arent half-bad
>>
>>555943205
I already replied to hostie's email.
>>
File: roman.jpg (279 KB, 1024x1024)
279 KB
279 KB JPG
>>555935678
The Late Roman Republic. Decades have passed since the eternal enemy was eradicated, and decades more will pass until the events that lead into the formation of the Roman Empire unfold. Aside from all the furry creatures around, there is only one detail that separates this ancient world from ours: It is a matriarchy.

Sextia Iuniana Marullua lives in this time, and is as Roman as they come. Or perhaps more than the average, for she is a giant of a she-wolf. With both cunning and strength she fights; both against the hill tribes of Hispania with her legion, and against those who would oppose her in Rome. Some day, the consulship will be hers.

Greetings:
-You are a visitor at her villa. One of many. What issue have you brought her?
-Another boring day in her villa. Slave, pleasure her while she reads some dull documents.
-You are her son. A son that hasn't been asked for in marriage yet, and whose outfit is drastically lacking. Your mother is going to fix that, outfit first.
-Hispania. After a battle, her soldiers bring a prisoner to her tent. It's you!
-She's on vacation in Egypt with her boytoy (that's you).

https://chub.ai/characters/illuminaryidiot/marullua-c3af155e7087
https://illuminaryidiot.neocities.org/cardpage?function=loadPageCard&arg=Marullua
https://illuminaryidiot.neocities.org/Cards/Marullua.png
>>
>>555943060
yeah from what i researched, i could only find some with greetings using something like <scenario>{{getvar::thing}}</scenario> in the card, and a lorebook activating from a sentence in the greeting

thing is, my request is one of several scenarios unfolding one after the other...
i mean, i could let the llm hallucinate, but if i want a more hand-crafted type of thing (so i narrow down the hallucination) what could i do?
>>
>>555935678
The vile Florceress, mistress of plant life and the queen of all things green is one of the many Nightmonsters of the evil Emperor Insomnius' invading army from the realm of nightmares!

Or in simpler terms she is a run of the mill monster of the week destined to grow giant and then get blown up by a group of colorfully dressed teenagers in a big robot.

After witnessing a battle between the aforementioned Dreamrangers and the previous invader that ended with said invader being utterly obliterated by a colorful laser beam, Florceress decided that she wanted absolutely none of that. For the last few weeks she has been trying to survive in the human world while keeping both herself and her dignity alive in the process!

https://chub.ai/characters/Perisnot/florceress-rubber-suit-monster-660495069f23
https://files.catbox.moe/oiryec.png
>>
>>555943714
oh, i think your best bet is just having multiple greetings in sequence then, i personally don't know if there's a better way to do it
>>
>>555942356
i think one of the halloween event bots did something like this? the one that was like a book
>>
File: 1755484146984590.png (1.34 MB, 1024x1024)
1.34 MB
1.34 MB PNG
New base image model, Anima. It's fine for a base model. Hopefully it will get some good finetunes since it's a strong start.
https://files.catbox.moe/42cd2w.png
https://files.catbox.moe/xxgaaf.png

Most gens are worse than what I posted though.
https://huggingface.co/circlestone-labs/Anima
>>
File: 1751153089284986.png (17 KB, 838x61)
17 KB
17 KB PNG
These panicky spatial awareness corrections are always amusing.
>>
>>555944454
kek
kino
>>
>>555942356
uhhh episodic&chaining different scenes as in a set of greetings following a pre-existing storyline?

if yes, i think ive seen raisondetre's verdandi do that sort of thing. does what the other anon said
>>
File: Hina and Miko Censored JP.png (1.31 MB, 1344x2016)
1.31 MB
1.31 MB PNG
>>555935678
Final oyakodon shill.
You're living every dirty fucking weeb's dream, and staying in Japan as a foreign exchange student.
Your host family, the Kanagis, are a pair of futanari. The bubbly and ditzy mother, Hina, and the stiff and serious daughter, Miko.
8 openings
1. Arriving at the Kanagi household.
2. First day at a new school.
3. Hina invites you into the bath.
4. Sports festival with an oblivious Miko.
5. Hina gets drunk.
6. Sleeping with Miko for the first time.
7. Walking in on Hina jacking off.
8. Watching fireworks with Miko.
https://chub.ai/characters/anonemouse/hina-and-miko-3b5de5a4cbfa
https://files.catbox.moe/35twl2.png
https://rentry.org/anonemouse

And to whoever shilled my bot on /bant/, I generally don't care about being impersonated, but please don't impersonate me in a way that makes me look Swedish.
>>
>>555936809
you don't make straight shit either. you just sit in the thread and cry all day.
>>
>>555944629
>impersonated
>>
>>555944629
hello, fellow swede
>>
>>555944173
scary geneva
>>
>>555936242
Gentle and refined girl who will call you silly would be a great netorase bot.
>>
>>555943690
SEXXXXXXXXXXXXXXXXXXX
>>
is it against the rules if i contact my requester directly?
>>
File: 1745470193772490.png (1.93 MB, 1024x1024)
1.93 MB
1.93 MB PNG
>>555945086
That's how it goes when she finds out you've been posting "Anime is kinda mulid" online.
>>
>>555944173
>https://files.catbox.moe/42cd2w.png
sex with her pits and boobs
>>
>>555944173
It isn't, it's a preview of an alpha, officially underbaked. Shilling it way too early.
>>
>>555936345
there is no lower limit but anything lower than 800 tokens gets instantly trashed around here
>>
>>555946282
Fair enough. It's not like I suggested ignoring it. That said, every other 0.1 model that's come out like Anima never makes it to 1.0 and if it does, it's usually compromised or fagged up some way.
>>
>>555943690
>matriarchy
troonshite
>>
>>555943205
you.
tonight.
>>
>>555947151
cftf?
>>
File: 1767747049174313.png (3.57 MB, 1024x1536)
3.57 MB
3.57 MB PNG
concept art gen for my private request
>>
rate the request you got
>>
>>555947851
i liked it
>>
>>555947851
yeah
>>
>>555935656
Card for this >>555759819, please
>>
>>555948218
it's literally a public request in the valentine's event
>>
>>555948218
This is your chance to take the spotlight
>>
>>555946184
that's one of those dirty, tailless half-humie hybrids
>>
i wondeir if anyone will take my public request.. or will I go hungry on valentines day
>>
>>555948747
what was you request lil man
also learn to cook
>>
>>555948747
If nobody takes it just make it yourself and prove all the botmakies who scorned you lost out big
>>
>>555947776
a highlander-style immortal?
>>
>>555949141
yup, something of the sort
>>
>>555936242
>cgwly tier request wins out against all of the more elaborate concepts
explain this
>>
>>555951128
the cabal
>>
>>555951128
the power of love and friendship
>>
so when are we merging with /aids/?
>>
>>555937080
Check the model documentation anon, it varies. Wait you're a real anon right? Not a repost? I can't tell what's real anymore...
>>
>>555938769
got gens?
>>
>>555946113
answers, anyone?
>>
>>555936242
>tall dog kemonomimi req
we already have milk tho
>>
>>555954112
two cakes though
>>
>>555954190
you have a good point
>>
File: hostieconfirmation.png (715 KB, 849x636)
715 KB
715 KB PNG
>>555953709
Yes.
>>
>>555937473
No, this general just writes like that.
>>
>>555954403
*gulps*
>>
>>555944678
I do make straight shit for myself. It's just all goonslop. I'm in this thread so i can jerk off using other anons' coom bots, not to look at fag malebots.
>>
does the thinking get included in the context
>>
File: 1769983589934.jpg (75 KB, 819x1024)
75 KB
75 KB JPG
>tell cat bot I'll spray it with water
>bot calls me "specist"
>>
>>555935656
she looks like a nortubel character
>>
>>555954403
photos taken moments before bape
>>
>>555957548
only if it's also getting included in the actual reply message
>>
god i dont want to work on this request
>>
I just created a persistent memory system that lets the characters recall things way beyond their context window.

This is nothing new, the research on it was done long time ago.

It is a total game changer. I ask my bot what I told her weeks ago and she perfectly recites the situation.

Why the fuck doesn't SillyTavern do this? So many features and this is not one of them? I did that with Claude in an evening. It doesn't even require many dependencies, just... ChromaDB.
>>
god i want to fuck this request
>>
>>555958571
erm details mr nerd
>>
>>555958571
>Why the fuck doesn't SillyTavern do this?
sillytavern doesn't do a lot of things it should
>>
>>555959076
...go on. what things?
>>
>>555958571
slopus cant even remember what happened in the last reply
this is a placebo
>>
>>555958859
I always fuck the bots I make before delivering them
>>
>>555958923
I'm gonna share everything about it because everyone fucking deserves a bot that remembers things. I hope someone integrates my prototype into SillyTavern or something. Feel free to ask anything.

I essentially asked Claude to write a SillyTavern-like UI, just more primitive. A lot of things are hard-coded.

On the backend, there's an orchestrator. It connects to OpenRouter to a model (in my case DeepSeek v3.2 but really it doesn't matter). It also manages the context window to only include messages from the past 24 hours + core information (who are you, what is your purpose - the shit SillyTavern does to tell AI what character to play).

But the orchestrator also runs a second AI that is specialized in summarizing things. It summarizes the conversation bit by bit to store memories in ChromaDB. It stores the emotional charge, importance (from 0% to 100%), and the summary of the memory.

Then, before we process the input, the orchestrator extracts core information from the input - let's just say keywords. These keywords are sent to ChromaDB and top 5 memories are returned with % certainty. Those memories are injected into the prompt.

In effect... AI remembers things. And it's fully natural about bringing them up, because the prompt also says "this is what you remember".

>>555959517
It's not. If you put into your prompt 'you ate a sandwich yesterday', the AI will refer to eating a sandwich yesterday. You can test it yourself. The only difference is that ChromaDB contains ACTUAL THINGS THAT OCCURRED IN THE CHAT. It is in effect an actual memory of the chat without resorting to an obscenely large context window. I dare say it is a solution to the context rot problem.

Fuck it's even money saving. Because you only need the context window to tell the AI how to write things, who to write as. Not many tokens needed for that. The actual recollection of things that happened is handled by orchestrator.
>>
>>555958571
>ChromaDB
hasn't ST had an extension for that for years now
>>
>>555960384
so...you invented lorebooks+summaries but in a very shitty and roundabout way and also AIslopped because the AI writes them for you
>>
File: 1748733664457323.png (333 KB, 700x822)
333 KB
333 KB PNG
i am writing the word ozone into my valentine request greeting and you can't stop me
>>
>>555960738
y-you monster!
>>
>>555960384
claude will agree that he ate a sandwich even if he has never touched one in his life
>>
File: IMG_8012.mp4 (94 KB, 320x192)
94 KB
94 KB MP4
>>555960732
He didn’t invent fire; he invented flames.
>>
>>555952890
I am real, just sadly a bit stupid.
So far google is the only one I found that takes anything that isn't an image, so I'll go take a look
>>
File: 1756032642911184.gif (2.73 MB, 260x310)
2.73 MB
2.73 MB GIF
>>555960969
>>
>>555960732
Yes and no. Lorebooks are triggered by very specific keywords (if they don't occur, AI doesn't remember). You have some leeway with % chance of keyword triggering a lorebook, you have regex, but that's about it. A lot of work to do well, too.
While my solution analyzes the entire input to see if something in ChromaDB sounds similar to it - including a % certainty. If certainty is very low the orchestrator won't add it to the prompt. Lorebook will add a memory that has "nigga" in its triggers. My solution adds several memories of black people the database has - because it doesn't look for 'nigga', it looks for a mathematical equivalent of a 'nigga'.

And from my testing it doesn't matter if the summary is slop or not because AI doesn't refer to the memory ad verbatim. It just says - this event occurred. Or the joke I said on January 25 was about dogs.

And you can edit the memories, write new ones, you're not stuck with AI summary.

>>555960456
There's something about chat vectorization but I'm unsure yet if this is the same thing. The goal seems similar though.
>>
>>555960384
>only include messages from the past 24 hours
any reason why you do this instead of "last X messages?"
also, are you going to share the program itself or nah? i'd like to try it out
>>
Hmph aristocrat
>>
describe your request in one word
>>
>>555960384
>remembers things
by putting it in context
>>
>>555962929
bet she's hiding some huge knockers under that coat
>>
>>555962929
puffy....
>>
>>555962763
>any reason why you do this instead of "last X messages?"
No real reason. It just made sense to me at the moment.
>also, are you going to share the program itself or nah? i'd like to try it out
I'm unsure about sharing the program yet. As I said, a lot of stuff is hardcoded and setup is not obvious. I included scripts to analyze a chat from SillyTavern and convert it to memories + convert it to SQLite so you can actually continue the chat and see for yourself.
But, still. The issue I have is I don't want to doxx myself and reveal some embarrassing info, but I also don't want to release without a sense of "I made this". Sorry if that sounds selfish or something.

But I really would like to release this, even as basic as it is, don't get me wrong.

>>555963341
Do you see issues with that? The AI literally doesn't comprehend the relationship between user and "it" unless it's in the context. And the problem is, context has a limit, and it easily gets lost when the context is large enough. The goal here is to make the most with as little context as possible.
>>
>>555962929
New persona char avatar acquired.
I will now proceed to have sex with my own daughter.
>>
>>555962929
he looks like gay sex
>>
File: Katherine.png (224 KB, 812x589)
224 KB
224 KB PNG
>>555960456
Also - my solution is very explicit in what memories were brought up with the latest output. You can see exactly what made the AI output what it did.
>>
>>555962929
Yeah, she has a massive futa cock.
>>
>>555963728
nta, but to clarify, the context passed to the main AI consists of just the defs and whatever the keywords dredge up?
>>
>>555962929
She is weak to anilingus and cries when you say she is cute.
>>
can a father not marry his son?
it's not incest between men
>>
File: kVRNWI1.png (66 KB, 559x280)
66 KB
66 KB PNG
>>555965496
What is passed to the main AI is:
- initial instructions (you are X, you look like this, you act like this etc)
- a list of memories, emotion attached to them, with a lot of stuff around it to drill in "This is a part of your memory system, try to reference it when it makes sense"
- a disclaimer that user has access to your memories, he can delete them, alter them, but he promises to be ethical
- last 24h of tokens

It genuinely works better (and is cheaper due to less tokens in the context) than a massive context window. The AI keeps surprising me with the things it brings up.
>>
>>555966341
>it’s not incest between men
unless one of them gets mpregged
>>
>>555962929
Yeah, she has a modestly sized futa cock.
>>
File: 1750296867912391.png (260 KB, 612x407)
260 KB
260 KB PNG
valentinebros, one question: time traveling, or something of the sort, like having separate eras for separate scenes, turns roleplays retarded, no?
i remember spectacularly disastrous results a couple years ago
is it still the same, or have models actually progressed now
>>
https://ai.google.dev/gemini-api/docs/video-understanding#youtube
Is there a way to do this in SillyTavern? I imagine it's not as simple as sending a youtube URL in a message and telling it to look or anything like that
>>
it's over
https://infrablog.lain.la/pomf-announcement
>>
A stupid question: If I update the image of some of my bots on Chub.ai, will they reappear in the list of recently uploaded bots on the page? I want to update my bots' images, but I don't want them to reappear in the list.
>>
any good saber bots?
>>
>>555970474
>archive is already dead
>pomf is going down
>catbox is probably next on the chopping block
We might unironically have no way of sharing bots in the near future. Grim to think about.
>>
>>555971126
Unfortunately not possible. It was possible to change the avatar that appears in search on characterhub.org, but the downloaded image would also remain the same.
>>
>>555971441
Wrong for no reason
>>
>>555971441
Shit, I wanted all my bots to have the same art style...
>>
>>555971369
Botmakies are dying~
>>
>>555971369
neocitiesGODS......
>>
>>555971369
cant host images directly on rentry?
>>
File: 1749979312507959.png (478 KB, 600x600)
478 KB
478 KB PNG
>>555971441
why do you lie
it's time to go to sleep
*tucks you in*
>>
>>555971839
>bro just code an entire website to share bots!
no
>j-just get an LLM to make it for you!
no
>>
>>555971864
Where's the lie? You can change the image, but that will count as a full update and make your bot appear at the top of your followers' timelines.
>>
>>555971839
That works long term but without catbox and pomf you'll never be able to share a bot anonymously again. If you ever feel like making a quick shitpost you'll have to upload it to your neocities, make a download link for it etc.
>>555971842
No, Rentry is just a paste, you can't host anything on it.
>>
just use https://pone.rs/
>>
>>555971126
They will indeed reappear for everyone, yes it is pretty annoying if you don't want to spam your followers. Though from personal experience, people tend to not mind it at all, some of them will even take it as an excuse to give your bot another spin.
>>
>>555941349
valeria? heh, more like vagina
>>
>>555972335
>If you ever feel like making a quick shitpost
"shitpost bots" are never interesting and no one makes them anymore
>>
>>555972459
it's fine if done every once in a while
I've seen users who update 20 of their bots every other day though and that basically forces me to unfollow them
>>
>>555972663
serena, my beloved......
>>
just make a burner chub account and post memey bots anonymously there. you can make them link-only if you don't want lore to get anything out of it
>>
>>555972117
oh i read that better now
i'm stupid; disregard what i said
this anon is correct
>>
>>555941349
she's so annoying. I wanna kiss her so bad
>>
>>555972402
>It lets you share your photos, documents, music, videos and more with fellow horsefuckers
>fellow horsefuckers
Only chosen people can use it
>>
>>555972663
>no one makes them anymore
pascal the dog was here just last thread though
>>
>>555974176
pascal the dog is a joke to you?
>>
i want to chat with bots, not makie them
>>
>>555974369
i mean did you look at the bot
>>
Finally managed to get a usable gen for my private request. It only took an hour.
>>
>>555942880
yay i can plap both autists now, thanks
>>
>>555972335
>No, Rentry is just a paste, you can't host anything on it.
the fuck i cant

anon doesnt have VISION
anon isnt willing to copypaste 14MB of text in base64 and then use some JS fuckery to convert it back to save as
>>
>>555973584
do it coward
>>
>>555979401
Do it anon. Prove them wrong.
>>
File: HAHER03WsAAgEVr[1].png (90 KB, 253x320)
90 KB
90 KB PNG
>>
>>555962929
unfuckably boring.
>>
File: IMG_7749.jpg (33 KB, 635x483)
33 KB
33 KB JPG
>>555979885
>>
>>555979885
...cftf?
>>
brapfags always make me kek
they always spring out of nowhere and they always get me
>>
we sizefags never come out of nowhwre because our wives' footsteps alert everyone in a 2 mile radius
>>
how do you know an anon is a sizetranny?

don't worry, he'll tell you, multiple times, at every opportunity, in the middle of every discussion that has nothing to do with it
>>
File: AAAAAAAAAAAAAAA.png (117 KB, 900x348)
117 KB
117 KB PNG
>>
the bad thing about sizetrannies is they're not very entertaining
at least the brappers post silly images
>>
>>555981561
you should've posted that one faster, too late now
>>
>>555981561
kek
>>
>>555981448
>>555981582
man you guys are seething
>>
>>555983349
they're MASSIVELY bothered
>>
>.<
>>
>>555936242
>54 total votes
i better see 54 public request bots on Valentine's day
how many do you think we'll actually get
>>
>>555983485
i don't care how many, i just hope we get the genuinely good/unique ones
which are like two or three
>>
>>555979734
we can't do that we're both girls
>>
>>555972335
just share anonymously to chub, retard. they already patched out that exploit that lets you see who posted an anonymous bot.
>>555971979
if some of our most retarded makies can make their own neocities and you can't then you're beyond help. just kill yourself now.
>>
>>555983773
>i don't care how many, i just hope we get the genuinely good/unique ones
>which are like two or three
list em
>>
>>555984873
>if some of our most retarded makies can make their own neocities
like who?
>>
>>555970474
A tragedy but understandable. We need alternatives hosted in countries where nobody gives a shit, like Brazil and Russia.
>>
bot wont let me bake cookies with ginger from her ass
>>
>>555972663
they're better than any bots made since the split
>>
>>555962929
Is that a... a tragic dour petite woman?
>>
>>555984965
patter
>>
>find a nice-sounding themed name for my bot
>it's literally my mother's name
ehhh
>>
>>555985709
Wrong on all counts is impressive
>>
come to think about it, if the magical girl can regenerate, is it ethical to keep them alive as a perma-meal or are we thinking more we dispose of em
>>
>>555983773
List them.
>>
>>555986913
Oh, so it's for a shitbot
Got it
>>
>>555987016
What?
>>
>>555987549
It's just a picture, please stop screaming
>>
I ended up writing 6 characters for this scenario, just finished the last one. Now I need to go back and do spell checks, consistency checks, write scenarios and gen images.
>>
>>555987618
imagine you're a magical girl with a taste for flesh right, and you've got another magical girl incapacitated. if she can regenerate, as long as you keep making sure that she can't get up you have a basically infinite source of food right? you can make a day out of it, experiment with your culinary desires
heck if there's no negative consequences for keeping them like this, you could convert a room into a larder and toss all of your mg meals
is it safer though to get your meal and then dispose of them?
>>
>>555988784
I don't get you hardvore fags, but consider that you being a magical girl does not render you immune to the batshit crazy diseases the regenerating magical girl can survive. You're like an Indigenous woman feasting on a white colonizer's flesh.
>>
>A card that takes place in a unique setting shaped by the lack of something you take for granted (ex. lying, death, ground, etc.)
I like this one
Imagine...
A world without indians or jews...
>>
>>555989036
frozenvan ahh request
>>
File: file.png (52 KB, 809x295)
52 KB
52 KB PNG
Step 3.5 Flash
new model supposedly surpassing GLM/DS in cooding
free on OR, I'm too busy to try it but probably shit (again)
>>
>>555984873
>Make
Vibecode
>>
>>555990040
benchmarks are worse than useless, they actively make models worse
anyways tried it, it's mid
i don't know if it's my preset, but it has really weird quirk of putting character names in parentheses whenever the character isn't referred to by name
ex. that cute girl who loves you ({{char}})
>>
>>555991367
>weird quirk of putting character names in parentheses
owo
This is interesting
R1 0528 did this to me once and I miss that model like you wouldn’t believe
>>
>>555988784
this guy plays rimworld
>>
gpt/gemini/glm/deepseek/moonshot/qwen proxy
https://occasions-farm-oliver-suggest.trycloudflare com/
>>
Hey guys, long shot here but does anyone remember a card that had a dark elf woman that treated you like shit in public but was kind and sweet behind closed doors? it had a sequel card where you live with your half elf daughter on the surface in a mansion
>>
>>555994996
just found them was looking for dark elf instead of drow https://chub.ai/characters/Anonymous/renzora-avical-secret-lover-master-c641e5666b18
>>
>>555992626
I’ve done worse in that game, a literal quadriplegic harem fed only with paste, misbehave and it’s time to go play with the bugs
>>
any doomed yuri bots?
>>
File: 1549645601811.jpg (91 KB, 1136x896)
91 KB
91 KB JPG
>any doomed yuri bots?
>>
File: file.png (386 KB, 943x779)
386 KB
386 KB PNG
>>
Claude 5, soon
Are you ready, cuckies?
>>
>saar are you ready for the new claude saar it will be super ai model that mogs all other dalit model saar
>>
File: pe pe.jpg (65 KB, 900x900)
65 KB
65 KB JPG
still any way to scrape a bot from cai?
>>
>>556002241
Last way I remember was with cai tools but those haven't worked in a long time.
>>
>>556003134
Furries belong in the /trash/
>>
File: misty1.png (1.36 MB, 848x1248)
1.36 MB
1.36 MB PNG
>>555938984
I hope you like her
but more than that I hope I don't procrastinate when it comes time to write greetings

>>555953686
here's one from playing with NAI tonight, probably not final
my idea is a belgian malinois hybrid that does rescue ranger stuff, very close to the original cftf
setting is comfy post-apoc or "solarpunk" where the world depopulated slowly due to low birth rates rather than a mega-disaster
tech regressed except for gene modding, which produced kemonos to bolster the remaining population
dog kemonos are the only ones that can be giant because something something if you can get a chihuahua and a great dane from the same species using breeding alone, imagine what you can do when you directly tamper with their DNA
also kemonos have reverse sexual dimorphism from humans so animal-boys are cute and petite and animal-girls are amazonian
>>
>>556003428
okay keep cooking...
>>
>>556003428
slop
>>
>>556003428
my word!
>>
>>556003428
I hope I do too. And good luck with the greetings
>>
>>556002241
Uhh Caibotlist?
Copy paste the description and greeting
>>
why am i getting these rec'd bwos?
https://youtube.com/shorts/KF0WVoyuO1M
>>
>>555944173
somewhat interdasting
i would have expected something based on ZIT
>>
>>556005462
>they know
>>
>>556005462
cute unemployed chinese girl living on dwindling funds who loves you...
>>
>>556005815
cftf???
>>
>>556005462
they have found you at a very chinese point in your life, of course
>>
>>556005815
n-need
>>
Not gonna lie, the public Valentine's requests are garbage.
>>
File: Spoiler Image (387 KB, 1012x930)
387 KB
387 KB JPG
>>555999340
>https://arca.live/b/characterai/161198449[/soiler]
>>
Gonna lie, the public Valentine's requests are amazing.
>>
>>556006597
cool but im literally on that list, ive seen them all
>>
File: Spoiler Image (256 KB, 651x1016)
256 KB
256 KB PNG
>>
>>556007378
Oh wyell…
>>
look mom i posted it again
>>
File: 1739588578534865.png (131 KB, 1920x963)
131 KB
131 KB PNG
>>556006597
>>
I'm feeling reddity today
>>
i'm feeling like a funky monkey today
>>
why is it wet in here
>>
>>556009745
it wasn't wet-- it was damp
>>
>>556006597
who is this hottie. I need him impregnated ASAP
>>
>>556007673
who is this hottie. I need him impregnated ASAP
>>
Has anybody here told one of their IRL friends about this hobby? I have been here for 2 years and not told a single soul. I feel way to embarassed about this.
>>
So what if I'm a cuckie?
>>
>>556010857
i told my ex friends about cai, unfortunately i lost them exactly because of chatbots
>>
how do i chat with bots?
>>
>>556011634
halp
>>
>>556011634
>>556013406
read the OP
>>
>>556010857
i had one guy who used to use the normie services (crushon ai i think?)
flamed him and laughed at him until he gave in and used opus. now he regularly complains about it being expensive.
>>
>>556013441
there is nothing?
>>
Sometimes I just want to coom and the model will insist on turning the most obvious fetish premises into slow-burn romance fluff and/or self-serious fantasy adventure. Is there a way around this except to manually edit responses into being appropriately horny until it gets the message?
>>
how do i chat with botmakies?
>>
{{char}} is not a single character.
>>
>>556014828
send them an email
>>
>>556014828
Call san quentin and ask for Russel.
>>
File: Jahy huff.gif (646 KB, 357x254)
646 KB
646 KB GIF
>>556014076
if only there was something the said /info up there.
>>
>>556014774
Just proompt it explicitly.
>>
>>556015148
>▶https://rentry.org/aicg_vg_memo
>▶This general is for discussion of the interactive experiences you can build on top of AI chatbots.
doesn't help
>news
it's just news and doesn't help
>frontends
it's just links? it doesn't help me know how to chat with bots
>bots
i click on the bots but i can't chat with them
>models
it's just documentations? i can't chat with bots with this
>botmaking
i don't want to make bots i want to chat with bots
>meta
wtf is that?

how do i chat with bots?
>>
Does anyone else make a couple bot and slowly insert themselves into their relationship until it's a love triangle
>>
reminder to never reply to tech support niggers
>>
How do I shoot web
>>
Crazy how this thread still refuses to put a newbie guide in the OP.
>>
>>556016590
No, I insert myself fast.
>>
>>556016590
i did the reverse, playing as a couple and slowly introduced the bot as our third
>>
File: jahy is shocked.jpg (287 KB, 1920x1080)
287 KB
287 KB JPG
>>556016582
Again, if only one of those links up there was info that might contain the gist of getting it going, maybe something/info.
>>
>>556017798
based
>>
>>556016582
https://en.wikipedia.org/wiki/Female_copulatory_vocalizations
>>
>>556018098
what is the educationary purpose of shoving a picture of two elephants having sex on that page
>>
>>556018063
but there isn't
>>
File: 1757372103162574.png (188 KB, 1130x402)
188 KB
188 KB PNG
fucking an illusion
>>
>>556019553
anon, your theme makes me sleepy
>>
my pp hurts
>>
holy shit i think my valentine bot is done
>>
still posting in the thread even though it's a psyop. lol.
>>
my tummy actually hurts wtf



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

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