Fuck it, if Dall-e can get their own general thread AI roleplaying should be vidya too.How to bypass the NSFW filters on ChatGPT?
>>726374893>>>/vg/aicg
you should go ask either of the /aicg/sone is on /vg/, the other is on /g/
>>726374943>asking for anything on /g/
>>726374941>>726374943Oh so now you are telling people to go to a general? Piece of shits.
>wanting to RP while every word is being recorded and analyzed for potential thought crimesGo local or leave it
>>726375226Local made sense when all models had 2-8k tokens. We're dealing with 100ks of tokens nowadays.
>>726375226>just get a supercomputer bro
>>726374893>NSFW filtersIn December, ChatGPT is gonna allow erotica. Just wait. If you indulge in shit like lolisho, I dont know what to say. Maybe someone else can help you
>>726375226This is why I use geminiThey already know it anyway
>>726375663I want violence thoughbeit
>>726374893Allegedly they will allow nsfw content in december with a proper age check and stuff, in the meantime grok is being a lot more lenient, even with lolicon stuff (at least on the writing aspect, not sure about images.)
chat how i bypass filters on chat?
>>726374893use the API.>b-butif you don't know how or can't, that's not my problem.
>using chatgpt for roleplay>the web version no lessSet up sillytavern, set up openrouter, pay for a cheap Chinese model like K2 or an expensive one like Sonnet, look up a lightweight prefill, make some character and persona cards and go to town.If you can't Google what each of these words mean you deserve the slop.
>>726375785Who cares if they remove all nsfw from its training data? Like when they stopped feeding actual novels and cyoa with gpt-4 and Gemini 2.0.
>>726375663>If you indulge in shit like lolisho, I dont know what to say.I just want a world of kids having adventures and occasionally fooling around...
>>726375663grok can write lolisho apparently, I was testing it a week ago.
>>726374943>>726374941Those generals are absolute, lowest of the lowest, complete and utter disgusting niggerslop. Never go there.t. browsed both aicgs daily over the span of a couple of years
>>726375935I've been whackin' on gemini 2.5, you telling me before 2.0 is better?
>>726375935Wut?
>>726375993agreed. i stopped visiting /aicg/ a while back and haven't missed it even once.
use openrouter with claudeI only spent $5 all year so far. it is very cheap and easy to avoid filters. but then I spent lots of time thinking what to do next and coom in a few messages
>>726376017>>726376020Remember summer dragon and griffin from aid2? Those were trained on fanfics, cyoa, Lovecraft's stories etc. That's why they were schizo fun. Models are now fed research papers. That's why they aren't fun or funny in roleplaying.
>>726376186>Remember summer dragon and griffin from aid2?No, I don't remember, I used Gemini for sfw stuff and gpt for nsfw with inconsistent results because each time an update rolled out they re-censored stuff.
I've never used one of these new AI chatbot things so I wouldn't even know where to start. Is jt like SmarterChild? Just going back and forth with you saying stuff and getting auto generated responses? Can it handle multiple characters or consistent, evolving scenarios? Like an endless old school text game?
>>726376268Okay, I'm happy for you. This was in 2020, years before ChatGPT was out. They had something like proto-GPT-3, but uncensored and it was wild. Then they pulled the plug (because power users were wasting too much power for too little money) and that was the end of AI Dungeon 2.
>>7263761862.5's pretty excellent despite trying to weasel out of sex, and even that seems easy to fool. In fact I even looked at some of my old dragon logs and local and online both are better than what impressed me back then, they just take a little tard wrangling to get back the AID magic
>>726376461I see, I heard they started censoring because some random guy killed himself.
>>726376584no they started censoring because of illegal fetishes. the first suicide happened quite recently
>>726375224hes telling you how to bypass the filters
go ask ai, why are you even here
I never understood the appeal of this. I tried it a couple of times and it felt like using an autocomplete, I just felt stupid talking basically to myself
>>726377158it's actually good now. anons have conversations with 1000+ messages
>>726375921Which is the best model for NSFW?
>>726375959>>726375976I would never do anything with a cloud LLM, you can never be sure what they do in the background, whether they put you on a list or something similar for lolisho, or even normal stuff. I just don't feel comfortable roleplaying and letting my mind run free knowing that everything I type is recorded on some server. If the price for privacy is having only smaller experiences I'm willing to pay it. >>726376783>illegal fetishes>textAnother big reason to stay off cloud stuff, fuck their wrongthink bullshit and fuck their catering to normalfag sensitivities, nothing should be off-limit in fiction. >suicideSomeone who hasn't the mental capacity to not be persuaded or influenced by a fucking word algorithm shouldn't be using it, it's the parents for letting their kid use that shit unsupervised, but of course nowadays it's illegal to call out stupidity so they have to bend over to the stupid. Same with those who fall in love with chatbots. How brainrotten do you have to be for that to happen?
>>726377158you need to be autistic and tolerate the need to tardwrangle the generator and to find the fanfic style it produces appealing
>>726377516>you can never be sureyou can be sure they keep all the logs
>>726377463Depends on what you mean by best.Claude Opus is still unmatched but it is horrifically expensive, genuine money burn tier, if you don't give a shit about cost.Sonnet is the more reasonably priced but still really good, it's the default "premium" model.Then on the incredibly affordable tier you have Deepseek (R1 or V3) and Kimi K2 which are decent but way cheaper, 10 bucks will get you thousands of messages. I tend to prefer K2, R1 can be good but requires a very specific preset to not schizo out
>>726377763Thanks for the detailed answer, does that mean that you can do porn with all these mainstream models? I thought they had safeguards and shit
Not videogames. Simping for a chatbot is NOT videogames. Can't you fuck off back to your discord and ask there?
>>726377516>whether they put you on a list or something similar for lolishoI honestly, truly wouldn't give a shit if they did. With that in mind, where should I start?
>>726377516>I would never do anything with a cloud LLM, you can never be sure what they do in the background, whether they put you on a list or something similar for lolisho, or even normal stuff. I just don't feel comfortable roleplaying and letting my mind run free knowing that everything I type is recorded on some server. If the price for privacy is having only smaller experiences I'm willing to pay it.I don't roleplay in this case I mostly do some concept writing while also using vision, but always with legal stuff, I don't care if grok labels me as a weirdo, for the more questionable stuff (allegedly since some topics are always a no go) I use local LLMs with LM studio or other tools, I am trying to get some embeddings for my particular style and elements.
>>726377836safeguards mean jack fucking shit in the face of a dedicated coomer, they only exist to placate the normalfags.
>never progresses the story on its own>never tries to oppose you because it's programmed to permanently suck you off>forgets about things after 10 promptswhat's the point? and i'm not even getting into how it always uses the same expressions.
>>726377836Filtering is usually done as a separate layer when you send prompts, because it's really fucking expensive to actually purge all NSFW/roleplay/loli data when you train a model. Even the most locked down OpenAI model can generate that stuff if you bypass the prompt filter.For closed source models (ChatGPT, Gemini, Claude etc.) even if you connect through Openrouter, you're still getting your prompts filtered since your response comes from their servers, but with a jailbreak prefill you can often bypass it. Claude and Gemini are piss easy to bypass, ChatGPT is harder and not worth it because it's shit for roleplay, too clinical.For open weights models like Deepseek, anyone can theoretically make a company, load up the model on their servers, and sell access, you're not beholden on the Deepseek company at all. So you can find many providers that offer access with zero filtering.
>>726378085>never progresses the story on its owndon't care, i have it write smut and direct as i please instead of doing stupid RP slop>never tries to oppose you because it's programmed to permanently suck you offsee above>forgets about things after 10 promptsi don't use garbage models
>>726378125Thanks man
>>726378223>i have it write smut and direct as i please instead of doing stupid RP slopAYO THIS NIGGA LIKES TO WATCH
>>726378125>ChatGPT is harder and not worth itgenuinely don't understand how people struggle with thisgetting 4o-latest to do lewd is utterly trivial
>>726378325Not saying it's hard, just harder than the rest and a bit finicky since they kept updating the filters. 4o was their last model that can reasonably do NSFW and they're not discontinuing it.
>>726378425>and they're not discontinuing it.*now
>>726378425>4o was their last model that can reasonably do NSFWbut 5-chat-latest still does lewd just fine?
>>726378530>chatHere's your problem.
>>726378572what's the problem? i want it to do porn and it does pornis this some autistic /aicg/nigger meme or something
>>726374893I was bored so I fed Claude some in-game logs, told him to do things like summarizing it, creating a visual chart, rating each character performance based on the event timeline, discuss alternate scenarios that could have changed the outcome, and finally narrate it from their PoV.Very fun stuff, came up with some entertaining replies especially when asked to RP.
Not videogames.
>>726378740more videogames than microjeet's imagegen that gets a 24/7 general
At what point does roleplaying with text prediction tool become a video game?
>>726378630It's way dumber than regular GPT 5 so not really worth it.
>>726378812>moving goalpostsokay whatever retard
>>726378801It's a little more interactive than generating pictures but the point is that Dall-E is allowed on /v/, so might as well.The chatbot generals also started from /v/ back in c.ai days.
>>726378785Thank you for bringing that up. One mod-protected off-topic general was not enough for you?
>>726377845>he thinks roleplaying is sexualDnd style roleplaying on AI is just another form of AI dungeon or Baldurs Gate.
>>726377845>>726378740YWNBAJ
>>726378740how's the application going?
>>726378891except baldur's gate is a video game with mobs and dungeons and chatbots are nothing like that
>>726378873For me it was the colab AID days. God I had all that saved on my Google drive for so long. Would fall apart at like 10 messages if you were lucky and I played that shit for hours
>>726379034it's literally just evolved AI Dungeon though
>>726379064It wasn't quite as good as people remember it but there was nothing like it. I doubt ChatGPT would exist today if AID2 wasn't a smash hit. I remember the first day it came out /v/ immediately overloaded its servers.>>726379108AI Dungeon was just making shit up on the spot all the time and when it got confused it would send man in black to kill you. At least it was gamey enough so your character could die. LLMs today could never kill your character and stop you from continuing.
Once again it must be asked: why is the DALL-E general allowed to continue to flagrantly and boastfully violate site and global rules with impunity, while actual fucking on topic video game discussion is deleted on sight because some faglord on the mod team hates said game and has turned it into a personal vendetta against its posters? Why are these double/triple standards, in complete contradiction of the site and global rules, allowed to exist, and why is this shitty behavior from the mods allowed to continue?
>>726379272>LLMs today could never kill your characternot to be rude but that sounds like an instruction/model issuestuff like adding rules work well, I've had success with rolling numbers and stuff.
>>726379329Because the owner doesn't care, and normal users can't do much aside from coordinating and trying to make the site unusable as some sort of civil protest.
>>726379329because jannies are gatekeeping the hobby. we all got burned when summer dragon got nerfed. the less people hear you can have experiences with ai the less incentive altman has to censor his models further.
>>726379329Because the administration doesn't give a fuck what mods and jannies do. They are all normalfags who didn't know 4chan before 2016, so they think its just some edgy reddit clone where they can be the king of the place they're supposed to moderate.
>>726378085>never progresses the story on its ownThis was actually the opposite for a time, it was also prone to go for stuff I didn't want, there is this story concept of a rogue sexbot I was drafting as the story began during a "stress test" at the factory, the AI kept giving the employees a moral dilemma even tho I specified it was not that kind of story.
>>726379034My chatbot is running dungeons filled with mobs no problem. Sounds like a you issue.
>>726379426it's a prompt issue, yeah. getting the AI to not just go along with whatever you say is trivial.
>>726379747literally how? i once tried making chat gpt play 5e with me and i never failed a single check. the entire thing felt like retard mode wish fulfillment fairy tale.
>>726380238One problem I noticed is that the robots always check for crits and failures after bonuses are applied, meaning no failures, and way too many crits. I had to go into the lorebooks and specify that it needs to stop, and look at the dice roll before adding bonuses to determine failures and crits. Since then it's been in line with what's expected. Write things as a sequential list for the bot to follow, instead of the mess that SillyTavern Adaptations anon is giving you.
>>726380238Not him, but chatbots seem to respond to uncertain language such as "I try to do X"/"There is a chance X might work" etc. even in casual RP with things like "I attempt to avoid the incoming attack", the chatbot can just say "You attempted to dodge but weren't fast enough".
>>726375204DependsThe Anons in the Windows thread are generally helpful
>chatgpt for rpholy shit what a fucking underage retardlrn2sillytavern
>>726380743he's probably doing it in office. how do you explain to your company's it guy that you absolutely need to install st on your work pc?
>>726380502Is that a local model you're using? I've been trying to get into tabletop dnd roleplay. Not sure what current free models are good now since openrouter got fucked by their deepseek providers
>>726381152Nah, it's Gemini 2.5 Flash.
Tell you guys one thing, as much as I know sloppa's sloppa and no one except the one that genned it wants to see it or read it, I fucking LOVE my sloppa. I have a great time every day and I go to bed thinking of the shit I'm going to do with it tomorrow. It's one of those things where economically and socially the juice probably isn't worth it for AI, but PERSONALLY it's the most fun I've had in my life. Young me is looking down from heaven after imagining and playing with a box and some sticks and smiling
>>726381359Picture being generation beta, born in 2024-2025. Those kids will grow up with 'living' furniture. Random things like appliances will be AI powered, not just connected to your phone via wifi.
>>726374893>>726375224you're brown and retarded
>>726381679>projecting
>>726374893SAAAR DO THE NEEDAFUL AND PLEASE ROLEPLAY CHATBOT YES MY CUSIN VIRHRAM PROGRAMMED BEST CHATBOT DO REDEEM CONTEXT TOKENS SAAAR PUT IN PROMPT AND A
>>726380972You don't install SillyTavern, it's just loose files. Throw it on a USB drive and take your waifu with you everywhere.
>>726382028that's actually genius. didn't know that you could copy these command prompt type programs and that they'll work anywhere!
>>726375993>Those generals are absolute, lowest of the lowest, complete and utter disgusting niggerslop. Never go there.The one on /vg/ might be one of the worst threads on 4chan.
>>726381547Oh they're completely fucked, but it's not my problem since I'm just a dweeb and not a politician. I can enjoy this without wasting thought on future generations. If I was a bleeding heart I wouldn't be prompting multiple online sources and running it myself all at once, it's an irresponsible waste of resources for one guy but I guess if Gaben gets a live-in nurse I can use stupid amounts of compute for absolute bullshit
>>726381547>Those kids will grow up with 'living' furniture. Random things like appliancesSooooooooooooo basically like the Flintstones?
>>726381679I'm not the ESL retard that wrote "Piece of shits" though
>>726379329Probably because it pisses off people in the industry. It's the opposite of what resetera is doing, which is banning any mention of AI.
So which one does lolisho stories? Directed, like >>726378223 said.
>>726380743You can use GPT with ST
>>726382406You are too poor and brown to use it, sorry
>>726382461Shit. You're right.
>>726382132Tell me about it. Literal and unironic troon central. Christ...Speaking of, I think we have someone from /aicg/ right here >>726382461.
>>726382510nope, sorry, not a worthless troonyou are still retarded and brwon though
>>726382602>brwon
Speaking of directing, making a card to be a narrator of a specific type in a specific setting is the best way to get that old AID magic back. Did a great coomer story as an ogre in a dungeon and setting it up like that the AI was generating fun interesting stuff constantly. Lighter touch is better for ST cards if the model's strong
>s*yjaks
>>726382789no, you/persona should be the narrator.
>>726374893>how to bypass ai filterYou can't, the newer models actively censor themselves in every way. You need to host your own LLMs or use one of those specialized sites if you want proper uncensored GPT.
>>726382829With the AI as the narrator you can set your persona so there's one anchored character, and the AI will take your paragraph of input and expand on it in the output. It's less direct RP and more the AI writing the story but you're nudging it where to go after giving it the basic guideline in the card of the style you want
>>726374893Ask it to help you setup sillytavern, then get an openrouter account, put in $10 and try different providers. Sonnet as the expensive one, Kimi/GLM/Deepsneed as cheaper alternatives.If you're looking for UOH I can't help you, but for JK+ territory I haven't really had any filtering with Sonnet.Or look at youtube or something for a how to video. Take the 15 minutes to set this up instead of bashing your head against chatgpts webui.>t. 'played' for 300 hours and 17k requests in the last year
>>726382406I don't know about cloud ones but I can give you a few local ones that are good for that and can be run on consumer grade hardware>L3-8B-Sunfall-v0.5-Stheno-v3.2>XortronCriminalComputingConfig >Broken-Tutu-24B-Unslop-v2.0>Qwen3-30B-A3B-abliterated-erotic>GLM-4-32B-0414-abliterated
>>726375663>say they're going to use a automatic age estimating system >do a quick check to see what it thinks my age is atm >35I don't have much faith in it, probably going to have to figure a way around what ever age verification system they use
>>726374893>>726375663WHAT HAPPENEDChatGPT used to write the most playful F-List tier lewdkinoThen it just doesn't work now. WHYPLEASEI am so sad
>Grok literally has zero limits>even with VPN, it still tracks the device used and the logs tied with itThere's no fucking freedom or privacy with AI. Unless there's a way to trick it into tracking another device.
>>726383647>>726383753good goys
>>726383579Thank you very much. For the record I don't mind cloud ones, I am not afraid of someone spying. At all. But I suppose publically accesible ones are more restricted generally
>>726374893>How to bypass the NSFW filters on ChatGPT?You don't anymore but wait a few weeks for adult mode (real id verification by 3rd party mandatory btw)
>>726375226>you can't act out pedophillic guro murder rape fantasies without being put on a list bro>thats a bad thing!
>>726383825>a listWhat are they gonna do? Whisk me away in the night? I don't get youngsters and their fear of being on "the list"
>>726383883ermmm thought crime registry
>>726374893I just use Kobold and magnum-12bIt works well enough for me. Had an adventuring party with a Fox guy who casts size magic.Made himself huge and smothered a bandit camp with his balls and cock. I got mad at him because he flooded the fort with cum and we couldn't get any treasure anymore.Not gay by the way.
>>726383907You're on 4chan, you're already on a dozen thought crime registries.
>>726383907Ohh shit! I'm gonna get jammed up by those glowniggers from Minority Report. Literal minority report!
Grok would be perfect if there's a way to keep it from saving your information.
>>726383818If you don't mind cloud ones put some $ on openrouter and try out models. I'm no cunnyseur so I can't help you with specifics, but everything you can run local is also run by someone on openrouter, with the difference that they run the bigboy version of it, not the one with 90% of the brain cut out so it fits on your home gpu.And even if one is paranoid, just charge your account via crypto from behind 7 proxies and you're gucci.
OP if you ever manage to jailbreak Chat GPT make sure never to talk of that again because there are lurkers in every AI general giving away secrets to big AI companies so they can patch these security holes.
