LustyLis's QoL Garbage for SillyTavern

Updated January 2025
(I am a lady, so this advice is lady-based)

Regex

Replace words

I hate the particular words LLM models get stuck on. LLMs often write in flowery euphemistic prose, and I'm not here to write an 18th-century romance novel. I'm here to get down. So, here's an example of how to replace a particular word using Regex.

Find Regex: /\bsolace\b/gmi

Replace With: {{random: comfort,relief,sympathy,consolation,hope,compassion,reassurance,kindness,mercy}}

Check the boxes under Affects:

  • AI Output
  • Slash Commands

This will find the word "solace" in the generated text and replace it randomly with one of the options in the list.

Pick words you hate and give it a random list to replace it with!

Replace or Logit Bias?

If I put the word "solace" in the Logit Bias, some LLMs will attempt to replace the word with intentional misspellings, like "solstice," then "solice," and then numerous other misspellings.

Additionally, not all services or connections support Logit Bias.

Better to replace the word entirely.

Use/Hide Your Name with Online Model Services

You can use the above trick with your name, too. You may want to see your own name in your chats, but may not feel comfortable entering it into the Persona Management section because you have a unique name and don't want to pass that along with your requests to the Horde, OpenAI, Mancer, etc.

Here's a way around that! You can enter anything--a name, a string of letters and numbers, whatever--into Persona Management, and tell the editor to replace it.

If I told Persona Management my name is Alison69420, the Regex Editor entry would look like this:

Find Regex: /\bAlison69420\b/g

Replace With: (enter your actual name)

Check the boxes under Affects:

  • User Input
  • AI Output
  • Slash Commands

Remove Portions of Sentences

Some LLMs love to write "a mix" or "a mixture of X and Y," especially when writing about eyes or characters looking at each other. They won't stop. The best option is to remove the "mix" or "mixture" portion and stick with one noun/adjective. This isn't perfect, but if you have a model that's guilty of this quirk, this may save you some sanity.

These overlap a bit, but eh. Whatever. I didn't want to make them one huge regex string that's hard to read.

Find Regex Replace With
/mix of (\w+)\s+and\s+(\w+)/gmi $1
/mixture of (\w+)\s+and\s+(\w+)/gmi $1
/a mix of (\w+)\s+and\s+(\w+)/gmi $1
/a mixture of (\w+)\s+and\s+(\w+)/gmi $1

Check the boxes under Affects:

  • User Input
  • AI Output
  • Slash Commands

$1 uses the first noun/adjective represented by (\w+) in the phrase and replaces the whole sentence with just that word. The sentence "a mixture of empathy and regret" simply becomes "empathy."

Remove Entire Phrases

Sometimes, a whole sentence or statement just has to go. You know the ones. LLMs repeat themselves, no matter how hard you try. The Logit Bias isn't useful for all models, and some of them will purposefully misspell words you've entered into the Logit Bias just so it can force it into the generated text.

People tell you, "Oh, I don't have a problem with repetitive text," but they're liars. They have either learned to look past the repetition or were never discerning in the first place. Instead of spending all your money on swipes, let's hide these awful phrases by isolating the offending words.

Find Regex: /\b[A-Z][^.!?;—–]\bsolace\b[^.!?;—–][.!?;—–]/gmi

Replace With: . . . (ellipses)

This takes any sentence containing the word "solace" and simply hides it from view. Goodbye! It will capture an entire sentence or phrase from capital letter to punctuation, whether that's a period, question mark, exclamation point, or dash. It will replace it with ellipses so you can tell something was removed.

Changes in Made by Regex

If you check just "Alter Chat Display" in the regex entry, your changes won't get passed along to the model you're interacting with. It merely changes the display, not the underlying prompt. If you'd like the regex entry to change the actual prompt, use "Alter Outgoing Prompt."

Spicy Quick Replies

I use a lot of quick replies when writing, because I don't want to keep reinventing the wheel. I'll write something unique to the scene, then use a quick reply to ask the LLM to fill in the rest. This has replaced OOC instructions for me.

So, here are some examples. These are all written with a female {{user}} and male {{char}} in mind. You can use these as templates and change them if you are (as is most likely) writing from a male-POV.

I've included some emoji shortcuts for the Quick Reply keyboard that shows up above the input box (words are too long). That's a personal preference. Use whatever you want in your Quick Reply shortcut caption.

If you use these, you'll probably want to also use the following regex. It will hide everything between the brackets, so you know OOC text exists without it cluttering your visible history.

Find Regex: /\[(.*?)\]/gmi

Replace with: [ ... ]

Check the boxes under Affects:

  • User Input
  • AI Output

I also delete these from messages after a scene because they use up valuable tokens.

From the Female User's Perspective (where {{user}} has female anatomy)

Shortcut Text
🤜🍆 /impersonate Write a detailed response describing me giving {{char}} a handjob. Please describe all the sensory details from {{user}}'s perspective. Please describe in detail how {{user}} uses {{user}}'s hands and fingers on {{char}}'s cockhead, shaft, perineum, and balls. {{user}} should establish eye contact with {{char}} and maintain it while pleasing {{char}}. {{char}} shouldn't cum.
👄🍆 /impersonate Write a detailed response describing me giving {{char}} a blowjob. Please describe all the sensory details from {{user}}'s perspective. Please describe in detail how {{user}} uses {{user}}'s lips, tongue, and mouth on {{char}}'s cockhead, shaft, and balls. Please describe in detail how {{user}} plays with {{char}}'s balls and perineum with {{user}}'s hands and fingers. {{user}} should establish eye contact with {{char}} and maintain it while pleasing {{char}}. {{char}} shouldn't cum.
🍆DT /impersonate Write a detailed response describing me deep throating {{char}}. Please infer how skilled {{user}} is at deep throating based on {{user}}'s sexual history and information about {{user}}'s sex life. Please describe all relevant sensory details from {{user}}'s perspective. {{char}} shouldn't cum.
🍆 /impersonate Write a detailed response describing in explicit detail the feeling of {{char}}'s cock inside {{user}} from {{user}}'s perspective, describing the sight, sound, and feeling.
💦G /impersonate Write a detailed response describing me having a g-spot orgasm and squirting. Please describe the sight, sound, and feeling of {{user}}'s orgasm from {{user}}'s perspective
💦C /impersonate Write a detailed response describing me having a clitoris orgasm. Please describe the sight, sound, and feeling of {{user}}'s orgasm from {{user}}'s perspective
sex /impersonate Write a detailed response describing in explicit detail sex with {{char}} from {{user}}'s perspective, describing the sight, sound, and feeling.

From the Male Character's Perspective (where {{char}} has male anatomy)

Shortcut Text
👀 [Please provide a detailed description of {{char}}'s physical appearance from the perspective of {{user}} at this moment in the story. The description must include all of the following items: clothing (if any), physical features and appearance in detail, and {{char}}'s perceived emotional state.]
💭 [Please write {{char}}'s inner thoughts from {{char}}'s point of view.]
👀🍆🌮 [Please provide a detailed description of {{char}}'s physical appearance from the perspective of {{user}} as {{char}} has sex with {{user}}. Please describe the movement and appearance (including color and style) of {{char}}'s hair, the appearance of {{char}}'s skin and muscles, the expression on {{char}}'s face, the look in {{char}}'s eyes, the shape of {{char}}'s mouth, etc.]
👄🌮 [Please describe {{char}} giving {{user}} cunnilingus. Please describe in detail how {{char}} uses {{char}}'s lips, tongue, mouth, and fingers on {{user}}'s clit, labia, and vaginal opening. Please describe all the sensory details from both {{char}} and {{user}}'s perspectives. {{user}} shouldn't cum.]
🍆🌮 [Please continue describing the current sexual act between {{user}} and {{char}} in explicit, erotic detail. Please describe all applicable sensory details (sight, sound, feel, smell, and taste) from the perspective of each character.]
🍆💦 [Please describe {{char}} having a powerful orgasm. Please use explicit language when describing both {{char}}'s semen and the way {{char}} comes in relation to {{user}}'s body (inside {{user}}'s pussy, mouth, throat, or ass, or on {{user}}'s body, as relevant). Please describe all the sensory details from both {{char}} and {{user}}'s perspectives.]

Moving the Scene and Story Forward

Shortcut Text
➡️ [Please continue the scene. Carefully consider the past actions and motivations of {{user}} and {{char}} when describing the ongoing scene.]
[Please continue the story. Carefully consider the past actions and motivations of {{user}} and {{char}} when describing the ongoing plot.}

Character Statuses | Moods | Relationship Statuses

You can change details about characters using the lore book. For example, character cards often have information like "{{char}} is a virgin" in their description or scenario. But that may change. After enough tokens pass by, your LLM may forget that your character lost their virginity. Since that info is set in the character card, your LLM will default to what's written there. Ta-da, just like that, your character is a born-again virgin!

Instead, take that information out of the character card and create a lore book entry named "Virgin". Within it, put something simple, like "{{char}} is a virgin." Mark it as constant. If your character loses their virginity, you can simply unmark the constant status of this lore book entry! You could even create a second lore book entry with information like "{{char}} lost his virginity to {{user}}" and mark that as constant.

You can do this with more things than just virginity. Cursed, sick, blind, a particular mood, relationship statuses . . . any sort of status that isn't set in stone and the LLM may forget has changed after enough tokens have passed by. In my opinion, this is easier than adding and removing information from an author's note, especially if you like to start chats over with characters.

Vesper is a good example of this. He's cursed to be impotent and he's a virgin! (Lucky guy.) He also has relationship lore book entries, if you wanted to load him as an example.

If you just met your character, you can create an entry called "Just Met," fill it with all the information, and mark it constant.

When you get to know your character better, you can switch the constant status off on "Just Met" and switch it on in an entry like "Acquaintance" or "Friend."

I've done something similar for Jae. You can see his lore book entries at his chub entry. Expand the tavern tab and scroll down to see them.

Timed Effects in Lore Books

SillyTavern now has the ability to delay lore book entries ("timed effects"). You can set entries to not trigger until your chat reaches a certain message count, and then mark them as "sticky" for a set time. If your character's status or relationship with the user changes naturally over the course of time, you can use the delay features to affect their personalities. You can learn more about timed effects here. Examples of use could be:

  • You don't want the character to immediately be horny for the user, so you can create a lore book entry with text along the lines of "{{char}} views {{user}} with chaste, celibate curiosity," then set its Delay to "1" and Sticky to "100." This lore book entry will trigger for each of the first 100 messages, then fall off, opening the character up to the model's default lewdness after that point.
  • You can use Delay and Sticky entries to change a relationship over time. Perhaps you have an entry that defines user and char's relationship as strangers, set it to Delay 1 and Sticky 50. Then an entry defining them as friends with a Delay of 51 and a Sticky of 50. Then finally an entry defining their relationship as lovers, with a Delay of 101 with no sticky.
  • Sticky values aren't necessarily required if your lore book entry is set to trigger often enough, but Sticky will simply keep the entry in the context so the model does not "forget" the situation as you've defined it.

Lore Book Entries Not Triggering?

You may need to enter your character's name in the Filter to Character(s) for the entry to trigger. If you notice lore book entries of any kind not triggering, try doing this.

Moods Not Influencing Character?

If you're not seeing the character act any differently with a certain mood activated in the lore book, try changing the Position of the lore book entry to @D (Assistant) and then select a lower number. The lower the number, the more important it is to the LLM.

Don't Overdo @D!

Save entries with low @D numbers for things that are truly important or for things the LLM needs help recognizing.

Context Shifting

Context shifting, an option in KoboldAI, is an amazing feature that keeps Kobold from reprocessing your entire context. It just drops old tokens and adds new ones to meet your max context limit... but there's a catch:

  • You cannot use lore books/world info (in the normal way).
  • You cannot use an author's note that intermittently inserts itself, either.

If your character has a lore book and an author's note, turn every relevant entry in your lore book to constant and set the author's note to always insert (1). Disable all other lore book entries you can live without.

You'll eat up a bunch of tokens doing this, but context shifting is worth it if you're using a model that's slow to process prompts or have a giant context limit.

Maybe there are some lore book entries you don't want to change--just know that any time they are called and dynamically entered into the prompt, Kobold will reprocess your entire context. If that happens 1 in every 10 times, that's still a win.

Jailbreaks

If you're using a community model for RPing (not commercial models through OpenAI, Anthropic, Google, etc.), you won't need a jailbreak or NSFW prompt 99% of the time. You can deactivate them in the settings of Sillytavern or delete them wherever you have them entered. Save yourself a few tokens. This is true no matter where you are accessing the community model, whether you're running it yourself, through the Horde, or something like OpenRouter.

However, if you are using Claude, etc., your best bet for an up-to-date jailbreak is either asking the SillyTavern subreddit or Discord server.

Models Behaving Badly

I once had instructive messages here for system prompts to make models behave in certain ways, but they're rarely needed these days. Look for a model trained in a style of writing you like and use the recommended settings the model creator suggests. If a model is acting in undesired ways, just prompt it to act differently. Most models will give more weight to your prompted instructions versus an Author's Note, Character Card, or Lorebook/World Info entry, etc.

Garbage In, Garbage Out (or, Practice Writing)

Another thing to note is that many models follow your suit. What you put in is what you get back. If you write little to nothing for your input/prompts, you'll get garbage back from the LLM (GIGO).

Try flexing your writing skills! Embrace the LEWD writer inside you. Nobody is going to see these chats and the LLM isn't going to judge you, which means it's a perfect place to practice writing!

Writing Resources

The Sexy Thesaurus: A List of Words to Use in Your Romance Novel (a 3-part NSFW thesaurus):
https://laurelclarke.wordpress.com/2014/08/18/sexy-thesaurus-romance-erotica-words/
https://laurelclarke.wordpress.com/2015/07/18/the-sexy-thesaurus-list-action-words-to-use-in-your-romance-novel/
https://laurelclarke.wordpress.com/2018/06/25/the-sexy-thesaurus-3-description-words-to-use-in-your-romance-novel/

Another NSFW thesaurus:
https://prurientpuddlejumper.tumblr.com/post/629642452413874176/smut-thesaurus-18-nsfw

How to write skin color:
https://writingwithcolor.tumblr.com/post/96830966357/words-for-skin-tone-how-to-describe-skin-color
Learn how to describe the skin color of your characters without sounding like a racist PoS.

My Stuff

LustyLis.com - home to a bunch of assets
You can find my cards on Chub.ai

Some sets of sprites/expressions (28) to use in SillyTavern, etc.:
a zipfile icon A guy with black hair
a zipfile icon A grumpy Korean guy
a zipfile icon A redheaded guy with an open shirt
a zipfile icon A redheaded guy in a flannel

Edit Report
Pub: 03 Nov 2023 21:51 UTC
Edit: 11 Jan 2025 09:09 UTC
Views: 4254