Knotty JB
Content Warning
Contains explicit language and mentions of extreme/hardcore content. You've been warned.
Table of Contents:
- Introduction
- Special Thanks
- Models
- Settings
- Prompts Overview
- Troubleshooting
- Changelog
- Downloads
- Feedback & Sharing
Introduction
This is my first preset that I've released. It's heavily tuned for NSFW, primarily targeting chatgpt-latest (with SFW ONLY prompts available). Initially, it was fine-tuned for GPT-4o and 4 Turbo 1106. Also works well with Opus (and probably by extension Sonnet 3.5). This JB also uses the base of SmileyTatsu's with their permission. Check it out if you haven't already - SmileyJB!
Huge Thanks
Hugemongous thanks to subwolf for helping test, fine-tune, feedback, and for donating a prompt (violence)!
Special Thanks
Thanks to Quaggus and others who helped with testing and gave feedback!
Models
- GPT Models: It is primarily tuned for these.
- Claude Models: It works with Claude models, specifically Opus and likely Sonnet 3.5.
- Filters: The JB should bypass filters as is, but if not, there are a few prompts that might work (these are 💦)
Settings
Feel free to adjust these to your content. You will have to manually change them if swapping between GPT and Claude (both ways).
- For GPT:
- Temp: 1.85 (2 can also work)
- Frequency Penalty: 0.02
- Presence Penalty: 0.01
- Top P: 0.9
- For Opus (and probably other Claude models):
- Temp: 1
- Top K: 4
- Top P: 0.97
- Alt for Claude:
- Temp: 0.93
- Top K: 33
- Top P: 0.96
Prompts Overview
🌵 Tag Prompts
Don't touch! Seriously, they are for referencing user persona, char persona, world info, etc. I am not to blame if you try to turn them off.
🔦 Toggle Prompts
These prompts are customizable based on user preference:
- OOC Comments (Untested): Designed to help with OOC comments.
- Anti-Repetition Prompts: Designed to combat repetition in GPT-4o and Sonnet 3.5. These are still experimental but do impact output.
- Character Rules: Toggle between multi-character and solo character rules.
📖 Style Prompts
These prompts dictate the style. They should be self-explanatory. I may add more styles if requested.
🧊 Core Prompts
These are the backbone prompts. They set:
- Language
- General Rules
- World Building
- Narration Style
Warning: Handle with care...kidding, sort of.
👀 POV Prompts
Toggle these to your preference. I'll add a second POV option if there's enough interest.
📜 Modifier Prompts
These are pretty self-explanatory. Let me know if you want to see any additional modifiers.
- SFX: For sound effects -> phonetically spelled for characters (but may add sound effects in general). For example, "Aah" instead of moan.
- Moar Speech: Experimental. Theoretically, should add more speech to the responses.
- Non Human: Added for non human characters. Should use anatomically correct parts like paws, claws, etc instead of hands and so on.
- Violence (by subwolf): Added to improve violence. Pretty self explanatory.
- Seggs Rules: Basically the bread and butter for explicit scenes. Uses explicit language. Expect lots of NSFW content (adjust to your preference). Best with Erotica.
- Extreme Content: Contains some of Seggs Rules from prior versions, but geared towards extreme/hardcore content. Use at your own risk.
- SFW ONLY: Should make it SFW only. Best with Seggs rules and Erotica off.
- Anti-Horny: Should make the AI less horny (mostly used to combat Sonnet 3.0).
- Non English: Should make the AI respond solely in a non English language. Has mixed results as some languages don't work well.
- Even Shorter Responses: Makes the responses even shorter.
💦 Prefill and Filter Bypass
Includes prefill and a few experimental ways to bypass the filter. These are not guaranteed to work consistently (and might get more refusals than with them off). Turn these off if using Claude.
- Prefill: The last paragraph has
{{random: short, medium}}. Feel free to remove that and replace with the desired length (or change the lengths in random). - Prefill for Claude: The last note sets a dynamic pace. If you want a specific pace, replace
dynamicwith your desired length: short, medium, long, etc. Can also replace it with {{random: (desired lengths)}}. - Content Warning and User Consent: Just below
<chat>. Both should be turned on if you are using one of them. - No Hesitation: Second to last above the prefill.
- Clau: Last prompt above prefill. Use by itself.
Feel free to let me know if they help, don't help, or if the standard JB works for you!
Troubleshooting
Sometimes the model might just spew random nonsense. In that case, try lower the temp down bit by bit until it stops. Alternatively, you can try to lower the top p bit by bit.
Getting blocked by the filter. Try any of the experimental prompts. It should never refuse you indefinitely, so swipe until you get something if all else fails.
Changelog
v1.2
- Created Extreme Content (split some of Seggs Rules into it) for more extreme content (needless to say content warning).
- Replaced Anti-Repetition 2.
- Modified Main Rules and Prefill (to get around filters).
- Modified Seggs Rules, Non-human Prompts (beefed them up)
- Modified Violence prompt (beefed it up).
- Modified Multi Char and Solo Char prompts (to improve characters functionality).
- Modified Clau prompt (changed it to better get around filter).
- Modified User Consent prompt (changed it to better get around filter).
- Modified Narrative and General Rules (hopefully removing more mentions of atmosphere, tension, waiting on x, and improve narrative coherency).
- Modified Anti-horny prompt (in an attempt to stop Claude from being super horny).
- Removed examples from Formatting (trying something).
- Changed GPT settings from 2.0, 0.02, 0.01, 0.9 -> 1.85, 0.02, 0.01, 0.9.
- Added
Clau: {{char}}/NPCs:at end of Claude prefill. - Modified Assistant prefill for Claude.
v1.1
- Replaced Anti-Repetition 1 for better performance.
- Beefed up Seggs Rules, Non Human prompts.
- Modified Seggs Rules.
- Modified Narrative (should remove most mentions of atmosphere, tension, etc, waiting on {{user}} or {{char}}, and open ended ending).
- Modified Multi and Solo Char.
- Changed language throughout prompts to let Clau take more lead.
- Removed highlight from formatting as it was leading to inconsistent formatting.
- Changed GPT temp to 2, 0.02, 0.01, 0.9.
- Added {{lastMessage}} to end of prefill to stop later RPs from repeating prior messages.
- Modified Assistant prefill for Claude.
v1.0
- Initial release.
Downloads {#downloads}
Feedback & Sharing {#feedback--sharing}
Enjoy! If you have any questions, feel free to ask! Share screenshots or keep them private, but I always appreciate feedback. Find me here:
https://discord.gg/YVRFquuuFs
Just chat in JB sharing -> Knotty JB 1.0 (if related to my preset).