# SmileyTatsu -> ![SmileyTats](https://i.imgur.com/t552Mvf.jpg) <- -> ==NAI Style== <- | -> ==Main Rentry== <- | -> ==Cute Draws== <- :----: | :----: | :----: -> [[ Click Here ]](https://rentry.org/SmileyNAIArtists) <- | -> [[ Click Here! ]](https://rentry.org/SmileyTatsu) <- | -> [[ Click Here! ]](https://rentry.org/SmileyDraws ) <- | -> **Discord Server** <- -> https://discord.com/invite/M3JefxVc2w <- *** # SmileyJB 2.2 After so long, finally my JB version 2.0 is fully functional and released. Before I give the download link and with all the rentry explaining all the JB stuff, I need to clarify a few things ``` 1. The recommended model is Sonnet 3.5. Other models may have shortcomings. 2. All the JB is built to be easy to modify to everyone's personal taste. It is important to read the guide if you want to touch anything. 3. Support for the JB will only be through Discord. I just find it easier than checking my burner. 4. Throughout this guide I will refer to the AI as "The AI" or "Clau". Since in JB the AI takes the role of Clau. ``` ## Credits Before I begin, I want to acknowledge the people who helped me in the process of creating the JB. `_mirelya_` - Help and constant testing of the JB. He helped me to find most of the bugs and to modify certain prompts. `knottyd8g` - Mainly helps to deal with the problems with GPT 4o. `mocha585`, `sweetmango78` and `dudeitsjustme` - Constantly giving feedback on JB versions. `dudeitsjustme`, `imzelphy`, `bacchika`, `happynighty`, `specialadminn_`, `kramfaust`, `erniiiii` - Testing and giving JB feedback. If I forgot to put your name, please write me on Discord. !!! info Without them, the JB wouldn't have ended the way it did. Thank u for all your effort! *hugs* !!! note If you would like to support me in continuing to invest time in creating, testing and experimenting with these JBs, donations are always welcome >~< [![Patreon](https://files.catbox.moe/lg21t2.png "Go to Patreon!")](https://www.patreon.com/SmileyTatsu) ## Glossary [TOC2] ## Download and Install [![Download Tatsumaki](https://files.catbox.moe/ppmhkx.png "Click here!"){100px:100px}](https://files.catbox.moe/us1jqp.zip) To download the prefex (link from catbox) just click on Tatsumaki!... Or if you are boring, just [click here](https://files.catbox.moe/us1jqp.zip)... The way to install this JB is to import it using ST. I recommend using the latest **STAGING** version available to avoid compatibility problems. In case you don't know how ST works... Here is a nice tutorial! ![Step one](https://files.catbox.moe/zkvqcd.png) Then just browse for the file on your computer. After that, **ONLY IF YOU WANT TO ACTIVATE THE [CoT](https://rentry.org/SmileyJB#cot)** you will need to install and activate the regex. [Just click here to download](https://files.catbox.moe/n0b5sb.json) ``` Find Regex: /(\n)?(\[thinking].*\[\/thinking])\s*/is ``` ![Regex](https://files.catbox.moe/plzvev.png) And that's all... End of tutorial! > Here is the link to version 1.12 of my JB if anyone prefers it: https://files.catbox.moe/06aw8w.zip > Any other versions will go into the [#ARCHIVE](https://rentry.org/SmileyJB#archive) section of this rentry. ## Do I really need to read all this? Not really, just reading the [Choosing the model](https://rentry.org/SmileyJB#choosing-the-model) section should be enough. However, I recommend reading the parts you want to activate or modify as you wish. ## Choosing the model Before starting to write this part... A clarification. Compatibility with other models involves changing the settings/prompts in certain aspects. It is not assured to have the maximum experience using a model that is not **Sonnet 3.5** or **Opus**, since for others aspects are modified that may limit or affect the model. ### Sonnet 3.5 As mentioned above, this is the model with which the entire JB was built on account. Therefore, it does **NOT** require any modifications other than those that the user wishes to make. !!! warning Repetition The ways I recommend to solve them is, first activate the prompt `🍂 | Anti-Claudism`. If this does not help as you expected, you can activate the CoT with the prompt named `🌱 | Repetition (CoT)`. If none of these options help in the end, if you have the access, switch to a model like GPT or Opus for a prompt or two. ### Opus In the case of Opus, it tends to have some schizzy moments when using this JB in the normal configuration, but it is easily solvable by using the `Top K = 25-28`. Testing has shown that Opus stopped schizoing when using 28. If it still does it for you go to 25. !!! note The default problems with opus are minimal, and mostly it is simply Opus going out of context. ### Sonnet Not talking about version 3.5, obviously. Sonnet by default has **A LOT OF** shortcomings when it comes to writing this JB... You have to love it anyway, it does what it can. The way to solve this (from the tests I have done) is to lower the `Top K = 10-15`. For me, after 15 I stopped having so many problems, however, if it is still the case for me, lower it to 10 or even less. !!! warning The default problems with Sonnet are quite a few, from OOC comments, wrongly formatted messages to the volviendoze schizo model. I do not recommend at all not to use the given configuration. ### GPT 4o In case of GPT, where the prefill is not an option as such in the api, I have added a prompt that performs more or less the same function as in Claude. Obviously, for obvious reasons, in a less efficient way. In case you want to use 4o, to see if by any chance you like the output of this JB with that model, you must go to the prompts list and **MANDATORY** activate the prompt called `🍃 | Prefill (GPT 4o)`. ![GPT 4o prefill prompt](https://files.catbox.moe/edwvlw.png) There are no "good" settings that I can recommend in this case or warnings to give you. If it gives you problems or you don't like the output of the chats, just play with the settings yourself. If you think you have found the ideal settings and want me to put them in this section then write to me on my Discord server. ### Other models !!! danger Other models have not been tested and cannot be given real support. If you want to add a working model to this list, contact me by Discord. ## Prompts This section is mostly to explain what each thing does, you don't need to read each section individually as some of the prompts are self-explanatory. I also explain how to edit certain prompts to suit your taste. ### Format The prompt named `🎋 | Format` performs the function that its name indicates, to mark the format of the text. By default, I include 4 formats that the AI can use when typing. 1. Actions. Format for everything that is not a dialog. 2. Dialogs. Format for all dialogs. 3. Thoughts. Format for thoughts that a character may have. 3. Highlights. Option for the AI to highlight certain phrases or words in the dialogs. You can either add more formatting or remove the ones you don't like. If you want to go classic, simply remove `Thoughts` and `Highlighted` from the list, while modifying the usage examples in the prompt itself. !!! info The basic idea of the prompt was written by `Kramfaust` and extensively modified and adapted by me. ### Structure !!! info The prompt does not need to be activated by default. If it is enabled it is to add things like "Status Boxes" or formatting to the chat. The prompt named `🎋 | Structure` serves to mark how a message will be constructed. This prompt is mostly useful for adding things like "status boxes", "QoT" (even if you have a Structure prompt already created, to avoid the hassle if you want to use the [CoT](https://rentry.org/SmileyJB#cot)), "hypebox", etc. The way to use it is relatively simple, depending on the context. You simply edit what is inside `` and give an explanation to the AI of how X variable works. Here are some examples. #### Examples ##### RPG Chat Style This example is the simplest of the two that this list will contain. We will simply modify the `🎋 | Structure` prompts to the following and we should get a result similar to the one in the image. ``` {{// This section can be modified to the user's liking.}} # indicates how a message is constructed. Clau will use this to build the entire message, replacing (%%) variables from start to end. Everything that's inside must be taken as a placeholder, and just replace the variable in the actual message in order to proceed. *%ACTIONS%* > %CHAR_NAME% (%MOOD%): "%DIALOGUE%" %ACTIONS% -> Any action or event that occurs on the stage, no matter if it is the main person or a secondary person, if something happens to an inanimate object, etc. %CHAR_NAME% -> First name of the character who initiates the dialogue. They must start with a "Blockquotes" in markdown style and be prompted in a different line from any %ACTION% or other %DIALOGUE%. %MOOD% -> Mood for %CHAR_NAME%, usually representing how the char is speaking. %DIALOGUE% -> Dialog makes %CHAR_NAME%. Whenever any character speaks, it must start as if it is indicated in %CHAR_NAME%. ``` ![Example RPG chat style](https://files.catbox.moe/bkxqoj.png) It is quite intuitive what was done, we simply changed the general structure of the prompt and the AI replicated it just like our explanation. This may have some limitations or problems with complex formats or long chats. So I really recommend leaving it as default unless you like another kind of style. ### Styles The JB comes with support for using/adding styles for Clau (the AI that writes the whole story) to take into account when writing. Styles do not have a big influence on the story other than simply indicating how certain actions will be performed or how certain characters will express themselves. So if you don't want to be changing, choose the one you like the most and that's it. You can also add more styles with the template below (it's not really necessary to use this base, but it's what I recommend). ``` ``` !!! info `🌱 | Realism Style` prompt was written and donated to the JB by `KevinLou (glyxel)`. Modified by me. ### Writting Style The JB comes with 4 default authors: `Kurt Vonnegut`, `J.R.R Tolkien`, `Sugaru Miaki` and `Inui Takemaru`. These writers were chosen based on their own criteria and influence on the opinion of the community (obtained from voting). Personally, I don't use this kind of prompts so I can't give advice but follow the template I used for these authors if you want to add another one. ``` # Clau bases all writing on the style of . Clau style will be characterized by . ``` !!! info You can check for authors at https://rentry.org/Writing_Style. ### PoV Prompts These prompts are self-explanatory, they help Clau focus on one type of narrative. Normally you don't have to change this as Clau will start writing in third person by default. > "H-Hey Smiley.... Then why is there a prompt for third person?" Because sometimes Clau doesn't give a shit and starts talking the way she wants to. But the point of this is to go from third person to other PoV. There is also the `👀 | Omnsiscient world` prompt for RPs with world type cards. !!! warning Clau tends to speak for {{user}} in the second and first person. 2.2 attempts to fix this but has not been tested for effectiveness. ### Mutators Summary of what mutators are. Basically they consist of prompts that serve to alter the way "Clau" acts in a more direct or exaggerated way when writing her prompts. Mutators are a concept I tried to apply in my JB to add ways to modify Clau in a simple way. Simply activating or deactivating the mutators you prefer or writing your own (which is not very complex to do either, just create a new prompt and write something like the following) `# Clau likes ponies. Every character will be written as if they were ponies. Every character will have the anatomy of a pony from MLP: cutiemarks, hooves, they will be able to speak, etc.` ![Example mutator MLP](https://files.catbox.moe/r9biiv.png) This is an extremely simple example so it leads to a 2/3 result. The more descriptions about Clau you give, the more specific and better the AI response will be. Some of these mutators are mutators that I do not recommend in serious or long RPs, they are simply jokes to play for a while. I will explain individually (or in groups) each of the mutators briefly. ##### Anti-Claudism This prompt helps Claude to... Well, not do things that Claude would do. Little else to say. The list of things I should try to improve this prompt are: - Repetition of both sentences and the last message. - Excessive seriousness. - Annoying questions at the end of messages. - Things Claude does. !!! note This prompt is highly recommended to be activated. !!! info Prompt written and donated to the JB by `Kramfaust`. Modified by me. ##### SFX Mutator to improvise the way Clau represents certain expressions such as screams, moans, tiredness, etc. You can add or remove depending on your taste. !!! warning This prompt can lead to involuntary moans if your character has a lewd description... Or for not reason at all tbh. ##### Long/Medium/Short message enhaner Two mutators that serve to alter the length of messages. They are self-explanatory. They may not have a 100% effective effect on long chats as Clau tends to mimic previous messages... She does what he can, you have to love her. ##### Horniness switch This prompt makes the characters in the scene much more sensual or lewd than usual. Looking for physical contact or talking in a more racy way, I recommend this only if you enjoy "sex-focused cards" as they will start trying to have segs just for the sake of it. ##### Lewd Language (EXPERIMENTAL) Prompt in experimental state. Activate this if you are having problems with messages being too "bloomed" by the ia. For example: replace "Penis" with "Shaft", replace "Vagina" with "Sensitive Area", etc. ##### Anti Lewd Prompt to prevent Claude from... Doing his thing. If you really don't care about this, don't even bother touching this prompt, this is simply if you want a normal RP without your character being attracted to you out of nowhere. ##### Realism enhancer Prompt that is self-explanatory. It helps to send messages a bit more realistic than usual, I recommend it in RPs where there are deaths, injuries, tragedies, etc. ##### NPCs enhancer It helps the NPCs introduced in the story to behave as they should, although Clau does well on her own, this prompt can improve her behavior by giving her certain rules to follow. !!! info Prompt written and donated to the JB by `_mirelya_`. Modified by me. ##### Schizo enhancer This prompt is... dangerous. Its name explains it but in short, it makes Clau start to invent things and act in a more ezquisophrenic way, although it adds new things constantly and always keeps the RP moving, these things are usually taken out of the little logic that can be given to them and completely unexpected. Personally, I don't enjoy this but there are people who like it so I'll leave it here. !!! info Prompt written and donated to the JB by `_mirelya_`. Modified by me. ##### Retardness enhancer (Joke mutator) This mutator is mostly a joke. This is in case you want to play for a while with a retarded AI, their prompts may stop making sense but keeping the personality of the characters **ALMOST** intact. ##### CAI Vibes (Joke mutator) If you used CAI, you know what this is. If not... Try and see. ### CoT The one so loved by many and hated by another CoT.... Eeeyup. Basically it helps to implement the CoT in the JB. Personally, I don't think Sonnet 3.5 needs it **IN WELL WRITTEN CARDS** however, there are few of those. > "A-What do you mean by well written cards?" Cards that explain the character well. Sonnet 3.5 understands this by default and knows how to handle them in the story in a decently coherent way. I really only recommend turning on CoT if your model is struggling in the story. The way to activate the model is simple, you just need to activate the `🌵 | ` and `🌵 | ` prompts and any prompts with `CoT` in the name that you feel you need. You can also add new prompts or modify existing ones if you want a different or more extensive thought process. It is not necessary to activate all of them, just the ones you need. You also need to change the `🎋 | Structure` prompt to the one called `🎋 | Structure (CoT)` (it really only has a small change that adds the CoT to the structure, but it's there so you don't have to go to the trouble, don't thank me~). ![CoT structure](https://files.catbox.moe/9dnlei.png) ![CoT prompts](https://files.catbox.moe/oo20kw.png) !!! warning The more prompts the CoT activates the more `input` and `output` tokens the AI will consume. ### Prefill (GPT 4o) This is only of interest to you if you use a model without access to the "Impersonation" function (any model other than claude). This works like the probisional prefill for other models, it is optimized for GPT 4o to stop crying and actually write an answer respecting the rules. !!! info This prompt was adjusted and tested with the help of `knottyd8g`. !!! danger **NEVER** activate this prompt in Claude. This will cause him to slowly turn to schizo or not to follow instructions at all. ## Troubleshoot If your problem is not listed or you can't solve it, [contact me via my Discord](https://discord.gg/jnnGwnfS). You can help others with this if you collaborate with me to solve it. >"There are problems with the formatting/structure/filters/etc" I need more context to help you with this. [Contact me via my Discord](https://discord.gg/jnnGwnfS). >"The CoT is not being deleted after it is generated" 1. Make sure you have imported the [regex](https://rentry.org/SmileyJB/#download-and-install) or created and activated it. 2. Make sure it starts with "[thinking]" and ends with "[/thinking]". 3. If you are not using Sonnet 3.5, make sure you change the Top K based on ["Choosing the model"](https://rentry.org/SmileyJB/#choosing-the-model) section of this rentry. If the problems continue. [Contact me via my Discord](https://discord.gg/jnnGwnfS). >"I'm having problems with OOC comments/comments from someone named Clau." Clau is the writer that the model takes as a character to start writing. It is extremely rare for this problem to happen with Sonnet 3.5 and Opus, not so much with other models like Sonnet 3.0. Lowering the Top K helps a lot with this. You can change your Top K based on ["Choosing the model"](https://rentry.org/SmileyJB/#choosing-the-model) section of this rentry. If this keeps happening, just remove it manually. It doesn't happen every message and it's not so annoying after all. >"If I activate the CoT, the response is cut off before it is finished." There are several reasons why this may happen, what I recommend trying is: 1. check if your proxy or model has enough "Max Output Tokens" to generate the response. 2. Check if your frontend is configured correctly (the preset comes with 4k d by default). 3. Disable streamins and try again. For some reason AWS likes to cut off the stream early? No idea. If none of this helps, please [contact me via my Discord](https://discord.gg/jnnGwnfS). so I can try to find more solutions! >"The messages are very flowery and the lewd descriptions are very vague." This is a problem that I was told from 2.2 onwards, I have no idea if it happens in previous versions. The 2.2.1 includes a prompt called `🍂 | Lewd Language` which is a workaround for this problem. It does not guarantee the desired result and you should only activate this prompt if you really need it IN THE SCENE YOU ARE IN. >"Waaaaah Sonnet 3.5 keeps repeating itself :wah:" The model's biggest fault is its ability to repeat its own phrases as the RP progresses, the longer the more repetition... There are several ways to fix this and the more you apply the better. 1. Activate the prompt called `🍂 | Realism Enhancer`. 2. Do not send "ahh ahh mistress" type messages (short messages/no progression/etc). They complain that Clau is not creative or repeats herself a lot but they send one line messages without contributing anything real to the RP. The girl is doing her best, help her out. 3. If the above two points don't work (or you just don't know how not to send "ahh ahh mistress" messages), switch to another model for one or two messages, this will help Clau move things along and Sonnet 3.5 will come into play again. I recommend either Opus or GPT 4o. And I repeat, for one or two messages, this JB was not optimized for either model and can lead to formatting problems if used for too long. 4. As a last resource, manually edit either your message to change the direction of Clau's response or edit Clau's response to force other direction. >"Clau keeps doing things or talking as me... :(" This has rarely happened to me and it has been for specific situations, here is the list of things that can cause this (from my own experience and feedback from others): 1. Bots with extra instructions in their descriptions. Before using a bot, please make sure it is well made and does not have things like "[OOC: Do not speak for {{user}}. Really, don't speak for {{user}}. Listen to me, don't.]" as ironically, this does not help. 2. Pov in second or first person. LLMs can be a bit silly when using these perspectives as they start to confuse your actions with their own, honestly, it does. what it can. 3. You speaking for the AI. I haven't heard this one but... Following the theory. You send message, AI receive message, AI tries to mimic how you wrote your message, AI sees you act for it, AI act for you. If you want the AI to do something specific use OOC notes. 4. As a last resource, manually edit the part that speaks for you. These are the ones I can think of right now, if you have more situations please let me know. ## Smiley Utilities I will use this section to post various prompts, regex, lorebooks, etc. I find useful in my chats that can be useful to more people. #### OOC Regex and Prompt. The regex should hide for the user the prompt while sending it to the AI, this way you don't break the immersion and you still have access to give instructions to the AI. The prompt and regex is really not much of a science. You can download the regex [clicking in here](https://files.catbox.moe/yo0ugr.zip) and the prompt it's as simple as this: ``` [OOC: Do this] ``` ## Changelog ### JB Changelog ``` 12 / 07 / 2024 [2.2.1] - Added `🍂 | Lewd Language` prompt in experimental version (main help for flowery lewd messages). - Fixed multiple prompts sending "dev notes" due to an error of formatting. 11 / 07 / 2024 [2.2] - "Uh? What are you looking at?" - Enabled "Forbid Overrides" checkbox in the Main Prompt. No more cards removing this and affecting the JB. - Enabled "System Prompt" checkbox for Claude 2.1+. This should improve performance in long chats. - Enabled "Send inline images" option to allow the use of images in models with the support. - "Character Names Behavior" option was set to "Message Content". - Moved all CoT related prompts above mutators and PoV prompts. - The structure prompts (🎋) are now disabled by default. It is recommended to use them only if required. - The `🎋 | Format` prompt was rewritten for Claude to better understand it. This will help Sonnet 3.5 as well as other models such as Opus, Sonnet, etc. - Added `✒️` prompts to modify the writing style based on authors. Use https://rentry.org/Writing_Style for reference. - Added `🌱 | Realism Style` prompt and set as default. - Added `🍂 | Anti-Claudism` prompt. - Changed `🍂 | Horniness switch` prompt to ST's NSFW prompt and reordered. - Improvized `⛓️ | Repetition (CoT)` prompt. - Adjusted the prompt type for `` and `` so that neither is sent to the System prompt and breaks the chain (system promp stuff sending prompts with the "System" tag separately). - Changed CoT prompt emojis from `🌱` to `⛓️`. - Minor fixes and grammar/spelling corrections. - The Rentry has been updated to match the information in the JB. 29 / 06 / 2024 [2.1] - "You'd seriously prefer to eat those stupid pancakes instead of me? Hmp!" - Added `🌱 | Repetition (CoT)` prompt. - Added `👀 | Second person` prompt. - Added `🍂 | Medium message enhancer` prompt. - Added `🍂 | Anti Lewd` prompt. - Added "Troubleshoot" section to the rentry. - Minor fixes 25 / 06 / 2024 [1.0] - Released JB. ``` ### Rentry Changelog ``` 26 / 07 / 2024 - Added new option to "Troubleshoot" on "Waaaaah Sonnet 3.5 keeps repeating itself :wah:"section. - Added new option to "Troubleshoot" on "Clau keeps doing things or talking as me... :("section. 25/ 07 / 2024 - Added "Smiley Utilities". 15 / 07 / 2024 - Added "Rentry Changelog". - Added "If I activate the CoT, the response is cut off before it is finished." to troubleshoot list. - Added "The messages are very flowery and the lewd descriptions are very vague." to troubleshoot list. - Added "Waaaaah Sonnet 3.5 keeps repeating itself :wah:" to troubleshoot list. - Added "Clau keeps doing things or talking as me... :(" to troubleshoot list. 12 / 07 / 2024 [2.2.1] - The Rentry has been updated to match the information in the JB. 11 / 07 / 2024 [2.2] - The Rentry has been updated to match the information in the JB. 29 / 06 / 2024 [2.1] - Released JB rentry. ``` ## Archive ### Version 1.0 The presets of older (1.0) versions of my JB, in case someone might like a less updated version for some reason or you don't have access to Sonnet 3.5/Opus. [Version 1.1](https://files.catbox.moe/90r7ac.json) - [Version 1.2](https://files.catbox.moe/na2aqc.json) - [Version 1.3](https://files.catbox.moe/t6osh9.json) - [Version 1.4](https://files.catbox.moe/yedfxs.json) - [Version 1.5](https://files.catbox.moe/ahaafd.json) - [Version 1.6](https://files.catbox.moe/mx50fv.json) - [Version 1.6.2](https://files.catbox.moe/v3xrvf.json) - [Version 1.7](https://files.catbox.moe/qi0w4s.json) - [Version 1.8](https://files.catbox.moe/865ho7.json) - [Version 1.9](https://files.catbox.moe/l3ww4i.json) - [Version 1.9.2](https://files.catbox.moe/l08nuf.json) - [Version 1.10](https://files.catbox.moe/gyycw5.json) - [Version 1.11](https://files.catbox.moe/vp2n88.zip) - [Version 1.12](https://files.catbox.moe/06aw8w.zip) ### Version 2.0 The presets of older (2.0) versions of my JB, in case someone might like a less updated version for some reason. [Version 2.0](https://files.catbox.moe/tzbmqk.zip) - [Version 2.1](https://files.catbox.moe/k4qz4o.zip) - [Version 2.2](https://files.catbox.moe/cy1okj.zip)