Welcome to Chatbot AI General #140, the thread for discussing and improving AI pony chatbots.▶ MLP Botshttps://mlpchag.neocities.orgSpreadsheet (CAI bots + Old repository):https://docs.google.com/spreadsheets/d/1J7BeqJVDS51cXF8Pgm2YZaFq-Z6ykSJTSillyTavern website extension: https://github.com/MLPChag/SillyTavern-Chag-SearchBunch of Character Cards (Descriptions/No Greetings):https://docs.google.com/spreadsheets/d/1Y6LNOCqAZAWIX-OBEv55HjlzcEdpUh-XbKCpbpL6v5kCAI bots converted to Tavern: https://files.catbox.moe/ckurq1.zipExpression packs: https://rentry.org/ChagExpressions!!!GALLERY!!!: https://drive.google.com/drive/u/2/folders/1Ao-h5HFGMPllSrzSBKM_BvGSiU9f0c2U▶ How do I start?1) Select a Frontend2) Select an AI model3) Select Jailbreak4) Select bots5) Lovemaking with AI mares!Starting in this hobby can be confusing and difficult. If it’s your first time and you’re lost,▶ ASK THE THREAD! ◀Novice-to-advanced guide: https://rentry.org/onrmshttps://rentry.org/sukino-findings▶ SillyTavern (preferred frontend)https://github.com/SillyTavern/SillyTavernOn Android: https://rentry.org/TermuxSTApp that voices pony responses in ST: https://drive.google.com/drive/folders/16Ss26VBmgzcSuTGzhaHqRuyVRceTf-YB▶ More frontends:Risu: https://risuai.xyzAgnai: https://agnai.chat▶ Localshttps://rentry.org/lunarmodelexperiments>>>/g/lmg/Mistral Nemo base model fine-tuned on fimfics: https://huggingface.co/Ada321/Nemo_Pony_2/tree/main▶ JailbreaksMLP JB: https://rentry.org/znon7vxeMore JB and guides: https://rentry.org/jb-listingHypebots for Tavern: https://rentry.org/pn3hb▶ BotmakingEditors: https://agnai.chat/editorGuides: https://rentry.org/meta_botmaking_listAdvanced: https://rentry.org/AdvancedCardWritingTricks▶ MiscellaneousScrapped/Collected rentries: rentry.org/mrhdChag Arts: https://drive.google.com/drive/u/2/folders/1yfajin_hB5rW_jy6WyFjRaRfDA3EUGiJhttps://rentry.org/ChagMiscellaneoushttps://rentry.org/ChagArchivePrevious thread: >>42503427**News:**- Deepseek V3.1 released: https://x.com/deepseek_ai/status/1958417062008918312- OpenAI releases GPT-5 models, all preceding models slated for deprecation https://openai.com/index/introducing-gpt-5- Z.ai releases GLM-4.5 https://z.ai/blog/glm-4.5- Claude Opus 4.1 released: https://www.anthropic.com/news/claude-opus-4-1
Anchor for bots, lorebooks, scenarios.Previous >>42503430
Anchor for technical stuffs (Proxies, Presets, Updates, Models etc.)
Anchor for asking for bots, lorebooks, scenarios etc.
Arts corner:GPT-pony >>42503500, >>42503535, >>42506964 Cute gatekeepers >>42504011 Miku pony >>42506856 Screenshots corner:Party >>42504272 Would you rather >>42504302 Brainy >>42504860 Midnight Sparkle >>42504918
Screenshots corner 2:Grilling >>42505226 Sassy Chorbo >>42505565 Exclusive Mare Club >>42505593, >>42505694 Is Twi ok? >>42505771 Who can it be now >>42506041 Fertility Priestess Twilight >>42506117 Ancient Ponk >>42507163 Cute Rainbowshine >>42507178
Screenshots corner 3:Le fancy piss >>42507384 Autistic >>42507696 Miku makes pancaku >>42508566 Recycling >>42510282 Gemini >>42510331 Luna plays WoW >>42510733, >>42510956 Magicjob >>42511048 Playing with OC pony >>42512845
Screenshots corner 4:Guests >>42513220 Thanks Celly AI >>42513698 Gamma-Band Aetheric Resonance Cascade >>42514687
VN Anchor:Last releases: https://equestrai-ponyponyparadise.neocities.org/ Anons' mods (Dashie mod and Nightmare AU): https://equestrai-ponyponyparadise.neocities.org/mods Custom characters: https://equestrai-ponyponyparadise.neocities.org/characters VN logs:Oo >>42504094 StrawMare >>42505106 Complicated magic >>42505706 INCONCEIVABLE >>42509850 Consort >>42509861 Rainbow Crash >>42510593 Scootaloo tries to fly >>42511715, >>42511735
These days, you can get legit good AI for free and without needing any demanding hardware on your end.Install SillyTavern:github.com/SillyTavern/SillyTavernTutorial and video:aicharactercards.com/guides/sillytavern-installation-instructions/Then choose a model. I'd recommend starting with Gemini, but you can try Deepseek, K2 etc. for free.rentry.org/sukino-findings#if-you-want-to-use-an-online-ai
>>42515285Gemini 2.5 Pro with that preset vs deepseek with my own (slightly modified ash-chatseek).Both are slop, but one uses a lot more words. IMO while deepseek is clueless, it's at least sharper than gemini and keeps up with the card's instructions. And I used gemini heavily in the past.
>>42515341picrel
Lovemaking with AI mares.
>>42515333Checked.Fingers crossed but Gemini 2.5 pro seems to be really stable right now.
>>42515366Echoing this sentiment. I switched back to Deepseek when Gemini was having all its problems, but the general quality of the writing is a notable step down from Gemini. Gemini pro has been working for me so far, and is the best model I'm aware of for ponies.
>>42515351
>>42515416I prefer some other models for chatting or stories, but it's very good for the VN IMO.
Reposting since it could also interest other anons.>>42515132>>42515149https://github.com/notstat/SillyTavern-SwipeModelRoulette
>>42515416Please share a good log with gemini, I have long-lasting gemini fatigue but I maybe there's a preset which works
>>42515431btw if anyone wants this extension to work with regular messages instead of just swipes in line 18 replace MESSAGE_SWIPED with MESSAGE_SENT
Has anyone tried this script? https://github.com/qvink/SillyTavern-MessageSummarizeI just started using it and testing it but wanted to know if any anons have used it
>>42515527i think thats the one i use, im not really sure if its helping or not but it hasnt really given me much problems
>>42515437I'd read it
>>42515330I don't know if anyone else is having these issues, but every time I head on up to Canterlot for a second visit to Celestia, the background doesn't load. Alone or with Twilight.Additionally, the exit dialogue for visiting Celestia alone is the same for if I visited with Twilight, resulting in a phantom student. Spooky.I tested this on a fresh install and on my version(With the myriad of edits).
>>42515527it's probably the best currently availablebut the summarization is per message or per X number of messages, which isn't very effectiveAnd the summary requests are sent directly without the preset which will break on some models, there's no quality control unless you manually check the summaries"no, Anon. I cannot summarize you fucking the horse pussy"i'm working on a fork of the extension, trying to add the following functionality:auto-detect the end of a 'scene' (logical break point in the story) and summarize the entire scene.auto-hide the messages within that scene so they're removed from context, just the summary remaininghave a second LLM check the summary for quality controllater i'm hoping to integrate with automatically parsing out details that should go into a temporary lorebook for the specific chat, but that's very ambitious and i'm not sure how feasible with the current limitations on extensions
>>42516117That sounds very promising, anon. I am a sucker for long stories so I'm trying every tool available. Would gladly help testing it when ready
>>42516423there's another problem i'm speculating about how to solve. AI referencing things without a lorebook entry being active.example - start of a new chatprompt: anon wanders ponyvillelorebooks active: ponyvilleresponse: he wanders ponyville, passing Twilight who is doing x,y,z problem: this portrayal of Twilight is completely out of characater for our version. her lorebook wasn't active when the AI made it's response, because the user hadn't mentioned herspeculated solution:have the AI response go to a local proxy, parse it for any lorebooks it would trigger, and if they're not currently active, regenerate the response with those for context, before passing it back to SillyTavern
>>42515764
Well that was a bit of an overreaction
>>42516804Deserved tbdesu.
>>42516104Will take a look. It does that on the second week/second visit for you?
>>42516804100% deserved
>Tried to browse MLP cards in chub.ai after a long time>First thing I saw is Crossed x MLPAre there any alternative sites other than chub.ai for cards?
>>42517149mlpchag.neocities.org>>42515304
>>42516804She's annoyed hers is small.
>>42517179Harmonizer has some headcanons inside, like Cadance not being an alicorn and Shining being a Duke instead. If you like the changes, it can be worth using, or you can modify it a bit to fit your own headcanon. Depending on the model you use, and if you don’t have headcanons, you can also simply choose not to use Lorebooks for FiM. Most big models already know a lot of G4, so depending on what’s inside, a Lorebook can just bloat your context with things the model already knows.
>>42517199Yeah, sorry. Just realized the entries are headcanon.
>>42515330A fake calendar system for the VN would be neat, something like Rune Factory or Harvest Moon. With special days like Winter Wrap-Up, Running of the Leaves, Nightmare Night, etc., or characters' anniversaries.
>>42516980Unfortunately yes. Interestingly, the black screen thing doesn't happen for the Nightmare AU.
I've just realised I must have been using something that's neither Agnai nor Silly in the early days. I distinctly remember making bots that are not on my Agnai account, and some chats that are not there too. Did Pygmalion have its own interface or something?
>>42517273Maybe RisuAI? Or maybe mixed memory with old CharacterAI bots? >Did Pygmalion have its own interface or something? There was a Pygmalion booru before Chub, but it was only for uploading bots. They weren't linked to a model, and you couldn't chat directly with them.It's dead.
>>42517273original tavern maybe?
>>42516438I thought the portrayal of Twilight by the LLMs is usually one sides. Like she can be a huge nerd or lack any social skills but I supposed if you use a lorebook it would help.
>>42517273Maybe KoboldAI connected to Google Colab?
>>42517214It's something I do want to implement, maybe after adding a proper Canterlot map.
>>42517628I've been collecting so many background images in preparation for a canterlot map.
Does anyone know for sure if the 3.7 sonnet on opus6 was scuffed or not? I couldn’t really figure out if it was when I used it.
>>42517826Define scuffed.
>>42516104>>42517267Both will be fixed for the next release, thank you anon. I was aiming for today, but there's another thing I just noticed, so I may release it tomorrow at worst.
>>42517954
>>42517971>Celestia blushingcute
Another day without Gemini shitting itself.
>>42518225I dunno, every VN response is telling me my rate limit's exceeded no matter how long I wait or what API key I use. I get a reply through sometimes but it's annoying as shit.
>>42518243It's working fine for me right now on the free tier with Pro, but I just saw this new limit. Could it be that your context is really heavy?
Status: Banished back to Earth from Equestria by Celly. Still possessed by ancient sorceress mare. Completed hostile takeover of rich doomsday cult.WTB nuclear fission reactor to build portal back home.
>>42518337You made a card for the possessing spirit?
>>42518364Yeah, she used to just be in my lorebook, but I knew I wanted to take the story to this reverse isekai direction with her as the main character.
>>42518314>Could it be that your context is really heavy?Couldn't be, I haven't messed with context. I redownloaded the game to see if it helps with it being "fresh", dunno if that means anything in this case. Maybe it is just picrel.
>>42518386>I haven't messed with context What I meant is that if you played a lot, it could be big and maybe cause issues.No need to download the game again, just for testing, you can start a new game and see if the conversation with Twilight works without problems. Or launch a Quick Start.If it does, then it's indeed that your context is too big. Try making a summary of your current save if that's the case.
I thought it was a fun scenehttps://files.catbox.moe/bodyij.pnghttps://files.catbox.moe/4rit8j.PNGhttps://files.catbox.moe/3ulgud.PNGhttps://files.catbox.moe/pcbgwj.PNGhttps://files.catbox.moe/cf3pb9.PNGhttps://files.catbox.moe/yvijse.PNG
>>42518447Cute!
>>42517149>Crossed x MLPWhat?
Hey, anyone else tried the opus6 chatgpt-4o-latest? I don't know why but for me it is ignoring my first message and then responding to it on its second reply, then ignoring the subsequent message and responding to it on the third reply and so on. For example:<intro>Me:"blablabla"Bot:Something that still makes sense for the character and scene, yet ignoring my input.Me:"Something else"Bot:Response to "blablabla"Me:"Give me the square root of 9"Bot:Response to "Something else"Me:"..."Bot:"3."And so forth. This has never happened to me before in any other proxy.
>>42515307Been playing grounded recently, and some old art came to mind so I wanted to write a micro survivalist pony. https://chub.ai/characters/CakeSparkle/peanut-brittle-9d42f2c74999
>>42518579yeah, this is a common problem I have had with using the higher end bots like opus. it really doesnt help either ive been using open router and using those ones are fucking expensive...
>>42518598Thanks for the bot, anon! Not really that important, but while checking her I saw: >her mouth trembles, but she cannot seem to speak. It should be "Her" with an uppercase.
>>42518579I don't remember having this problem before with it, but while testing the models, I was pretty convinced that chatgpt-4o-latest was either not really the latest or a pretty scuffed one. Sonnet models are legit, and I believe GPT-4.1 is legit too.
>>42518337I want to force her to play hotseat 4X games with me.
>>42518616my mistake, I wanted a friend to proof read but I got impatient
>>42518598here is the full art by the way
>>42518667No problem. It was minor, and it's always good to have hand-written greetings. I'll try her later.
These three have some funny interactions.
>>42518703You found enough sprites for them?
>>42518806I scoured the internet for them, spliced where I could and recoloured a bunch. Managed to get enough to make character cards for Night Guard and Nightmare Moon Guards. I also did Unicorn Royal Guard and the Pegasus Royal Guard. MIGHT do Crystal Guard. Either way, I'm waiting till VN updates so I can update the Nightmare AU, or I finish the lot. Whichever comes first.
>>42518598Cute. Thanks.
>>42518629It happens to me with all the models except the "opus 6" one. Mind sharing what preset you used?
>>42518923Modified: https://files.catbox.moe/m8u6qe.json With instructions for RP. Remove the NSFW part for SFW.Not mine, but an anon also posted his version based on it: >>42502680 It hasn’t happened to me in a while, but I remember having your problem with some bloated presets with certain sources in the past.
does anyone know if gemini 2.5 pro is still shit or they fixed it?
>>42518949It's been good for me since yesterday, but apparently an anon >>42518243 had problems with it, maybe because of >>42518314
>>42518940That may be the issue then. I have large world info with entries that have different insertion depths, big cards, and my preset is quite big too.
I hope you're all proud of yourselves.
>>42519047Has she seen the jar?
>>42519047I'm proud of my precious LLM for writing Twi, Spike, and their dynamic better than the show ended up doing.
>>42519047I regret nothing.
>>42518956Anon here, had no idea recapping the story could do... whatever it did. Pretty nifty!
>>42518956well I just tried it and no, it doesnt work for me.made a request and returned a message that the limit quota was reached, changed the model from pro to flash using the same key and it worked without a problem.
>>42519230Strange...
>>42519660>"There is nothing wrong with carrot cakes."
>>42519047>"Four-channel?">"Four-chan," she corrects automatically,Unintentional(?) bonus humor.
>>42519713Almost certainly unintentional.
>>42519660I like carrot cakes...
>>42519716>>42519713The fuck you mean, anon? All LLM humour is unintentional. They don't have free will or conscious thought. Stop humanising models.
>>42519967Unintentional in terms of 4chan-4channel jokes not being in the training data. It's not humanization outside of the language used to describe it, and that's just because it's easier to write it that way.
>>42519967>The fuck you mean, anon? All LLM humour is unintentional. They don't have free will or conscious thought. Stop humanising models.You're putting way more thought into this than the original poster did. No one is claiming that an LLM has conscious thought or free will. There is still a difference between writing it produces that is written to be humorous and writing that is not.
>>42518598Added, thanks anon.https://mlpchag.neocities.org/view?card=CakeSparkle/Peanut%20Brittle.png
>>42520021>No one is claiming that an LLM has conscious thought or free willCheck the Internet lately?
>>42520199correction then: no one HERE at this current time is claiming that an LLM has conscious thought or free will
>>42520203I am.
>>42515330anon, the filly pinkie pie character download is pointing to filthy rich instead
>>42520289Fuck. Must have fucked up in one of my latest updates, because I remember downloading the right one from the website for testing. I'll update it as soon as I can, but here's the link for now: https://files.catbox.moe/e6zr6j.zip
>>42520289Corrected.https://equestrai-ponyponyparadise.neocities.org/characters
>>42520526Phrasing.
>>42520570She deserved it.
>>42520570Hnnng
>>42519827
Just for info, what was anons' experience with free Gemini Pro today? It was smooth on my end, I got some >the model is overloaded messages, but they weren't that frequent.
>>42520931It's been working fine for me, I haven't seen any overloaded model messages, just the occasional internal server error. Those errors may or may not disappear after a couple more attempts, but if they don't I just switch to the next API key (because I have a metric fuckload of API keys) and it'll start working again.
>>42520931>I got some the model is overloadedmessages, but they weren't that frequent.Exact same for me. Was surprised at how short it lasted after the shit on the 28th.
>>42520931Exact same for me. A shame they reduced it to 50, but it's easy to make keys.
>>42520931Its been working fine for me the last few days. Didn't want to say anything, because I felt that might jinx it, but it seems like its back to normal for now.
>>42515416It's alright, but it's notably shit for cooming.
>>42521334What's better?
>>42521351For cooming specifically, I find the Sonnets to be better for the most part. Also 4.5, but it's not like you can use that now.
I love this story so much.I haven't spammed logs of every kino moment cause I doubt people care, but building a portal to Equestria with the help of a tulpa and a kooky cult in California is some of the most fun I've had with AI in ages. Would recommend.
