#So far: My best roleplay template yet

1 messages · Page 1 of 1 (latest)

novel topaz
#

I've been working on trying to make a better template that I can use to make different roleplay scenarios more easily, and so far this is what I've ended up with, with some help from this prompt making gpt https://chatgpt.com/g/g-67666844da848191aebc36380d968849-api-system-prompt-designer [He's really good at it!]

This particular template is meant to create a roleplay scenario where the GPT is like the "DM/GM" in charge of the story, while you the "user" are a single player within, interacting in an open world. It has some rules in place to stop the GPT from enoting for your character or taking over your decisions [ Its not flawless! As long as you gently remind it when it happens though, its really effective ] and it has rules for pacing as well, so it doesn't direct 3 hours of emotes in one single response, lol.

This is tailored a bit towards less combative/adventurous roleplays and more slice-of-life or romantic settings, however it would be quite easy to adjust it to fit anything you'd like ❤️
Feel free to use it, and lemme know what cool stuff happens ov o

[ If you come across problems also let me know! I would love feedback to be able to adjust and perfect the template ❤️ ]

Screenshot_2025-08-28_094452.png 📎 immersiveDMrp.md
#

I often times will take the template, hand it to the system prompt designer GPT and then give it a detailed description of the 'vibe' I want for the next roleplay [I try to avoid giving it plot events or story arcs - it very quickly goes over 8,000 characters] and it fills the template for me.

novel topaz
#

Oh and yes, this prompt is designed with the intent of being placed in the custom instructions, I filled it with the most important settings, the ones that pretty much require being constantly checked during the roleplay

Other things likely could be added in project files with just a line or two to tell the bot how to reference them

terse jay
#

I've been working on something similar, but the problem I have is context length and chat length.
I was playing D&D mystery campaign, with GPT5 DM'ing. However, after about 3 days of playing, it started hallucinating plothooks that were incorrect. This is why I point out the context lenght. It was having trouble retaining the key details.

So I started working on a prompt/process of using 5 separate files. A starter .md file, and then 3 .yaml files that kept technical/functional information (character, world, npcs) and the 5th and final file was also a yaml, but inside was a base64 encoded "secrets" that only ChatGPT would know (I can't mentally decode base64....)

At this point, it was starting to work great. But I ran into two serious problems:

  1. I couldn't get ChatGPT to keep these files up to date... nor give me a full output that I could put in the file myself. It always left things out, forgot them, etc.
  2. I could not get it to be verbose enough or detailed enough, that if I had to start a new conversation, and give it the files... that everything would make sense. Its like it was trying to be a bit minimalist with its notes/updates.. and left things too ambiguous for the next conversation to pickup.

I'm still picking my brain on how to improve the system. I feel like if instead of overwriting the files... I could come up with a way to just start stacking deltas... it would mean less gets lost.. even if the files start getting a bit big.

#

Quick Update: It never dawned on me to use Projects, within ChatGPT... as its memory is all shared...

quasi ether
# novel topaz I've been working on trying to make a better template that I can use to make dif...

I use JSON formating for the character I want 5.0 to reppresent. And I use projects with the option Project memory enabled so it is not using account memory ( it is availavle for paid accounts)
The JSON formated instructions are set at project instructions level.
Other instructions are past during the Rp if needed or rinforced now and then as often the not tends to dull when it detects tension rising.

One critical point: When you co create a story in an RP style mode, building a very strong backdrop and context is key to have the ai less snaping back to the scafolds and stick more to your own instructions.
The better the context is defined the better it follows.
If you want I can share one template for a character with examples as set in JSON format ( is less blurby more easy to parse ) even if not machine parsing just precise info.

novel topaz
#

ah, see I go into the roleplay under the like, overhanging impression of essentially, that ChatGPT is the DM, I dont plan anything out in advance nor do I go into it with a long term goal of any kind

#

every day is a one shot

#

So it doesnt matter if the gpt can get the plothooks right cuz its just making it all up as it goes xD

#

I also have not been able to actually get chatGPT to behave well on long and drawn out like, campaign-style roleplays that had any kind of endgame

novel topaz
#

Do you have a reason for why you use JSON over Markdown?
Ive just defaulted to markdown cuz that seems to be what the bot keeps suggesting, but if JSON is better I'd be down to figure that out

#

I used JSON-ish formatting when I was making bots on a different platform and it worked well for that setup, but chatGPT kept telling me markdown was better whenever I tried to make it help me format, lol

gleaming harbor
#

🕍

coarse haven
quasi ether
#

Json is clean, is value. Parsing is faster to process.
Markdown is good for readability but the skiming may miss the exact structure. More open to drift.

#

If someone has difficulties learning JSOn simply ask chatGPT to outout .
make your descriptive character file voice by voice then ask chatGPT to make it a JSON.
you'll get what you need.

I wont take you also into the why and how to use it, proposed weights or the sense of using any, how to solve contradictory statements.
you need to learn by doing and observing and more, ask your chatgpt friend to help debuging when it is not working as suposed. You will be amazed