#Prompt Setup for Scenario/Setting rp [SillyTavern]

1 messages Β· Page 1 of 1 (latest)

fervent heath
#

So I'm trying to set up a 'scenario' where the bot itself is moreso the 'setting' and there's no one specific char that is the 'main bot', but rather a bunch of different chars described using PLists in the Lorebook/World Lore. I could use the Group Chat setting, but that isn't really what I'm wanting. I'm not all that great at bullying the AI using prompts just yet, so I was hoping someone could help me with how I'd set up the prompt so that the AI stops describing the setting as if it were an actual character ala "The Abandoned Cryptid Museum stands tall and harrowing as its empty halls groan with" yadda yadda yadda. I can usually get it to behave a few posts in by basically editing its posts to remove that and call to one of my keywords in the lorebooks... but that's annoying and immersion breaking, and doesn't stop it from deciding the Abandoned Cryptid Museum needs to weigh in on my persona's attempts to free the silly monsters later on.

regal forum
#

You're using P-list, have you used 'Type=Location' to describe the Abandoned Cryptid Museum? I can't 100% say for certain that would solve it on Cosmos, but I've seen a lot of P-list users that don't use 'Type' and I, personally, find it useful.

I also find that putting characters you want to show up in the RP in the 'scenario' section. For instance, most of my charactrs have a section under 'scenario' that has [{{char}}'s Relationships: ...] and in there, I have characters listed under 'family', 'allies' and 'enemies'. I've had full RPs where Cosmos juggled all of those characters rather well, despite only using one character card and some brief WI descriptions.

meager mauve
fervent heath
fervent heath
regal forum
#

You mean P-List? I thought you said were already using that in your WI entries. Or did you mean the list of characters? The exact sort of setup I use for that is -
[{{char}}'s Relationships:
Family = Beth(Mother), Dan(Father);
Allies = Jane (Best friend, nerd), Arthur (Friend, Jock);
Enemies = Deborah (Rival, Queen Bee);]

Meanwhile, I have WI entries set up that trigger - one for each character - using their name as keywords. In there, I list stuff like appearance, personality and a brief description of who they are to the user.

With more testing, it seems like the model is a bit hit or miss. You might need to nudge it in the right direction a bit. But there have been times when it just brings characters in a few a a time and gets them (nearly) flawless.