GLM 4.6 Realistic and Spicy
Babes....
A Quick Heads-Up…
This preset nearly gave me a heart attack during testing, so you know it's good. 🥵😅
Below you will find the updated System Prompt for GLM 4.6. and eeeeeeven further down is the old version. You never know. Maybe you like the old version better.
Overview
Cliché is tamed. Mostly. Words still hit like physical blows every now and then but it's manageable. I made a few tweaks to encourage the model to include side characters more and stop GLM from spitting out the entire backstory of the character within the first five messages.
Characters change organically and with reason. So your dead dove rp won't suddenly be a cuddle buddy. Ya'll gotta work to fix 'em. 😉
Disclaimer: Only try to fix fictional characters and look for nice people in real life!
Creative 9/10
Complex Characters 10/10
Context Coherency 8/10
Horror, Dead Dove 9/10
Smut, ERP, Kink 12/10
Romance 9/10
Generation Settings:
- Max Token output: at least 2500 (the thinking tokens are counted into the output. That's a GLM thing.)
- Temperature: 0.8 - 1.0
- Max P (Top P): 0.95 - 1
- Min P: 0
- Top K: 0 (IMPORTANT)
- Penalties: (using -2 to +2 range)
- Repetition 0
- Frequency 0.0
- Presence 0.00
- Reasoning enabled.
Reason: GLM is a snowflake when it comes to Sampler Settings. It reacts to Sampler conflicts by defaulting to deterministic hell. Meaning the responses will be generated with Top P 70 and with that come a lot of tropes and cliché phrases.
Tips
- You can change the Content Rules and narrative perspective to your own liking.
Please be careful when changing other parts of the prompt. Wording can make or break the whole thing. - When you want shorter replies, add the following line to the post-history
- Keep your reply around 2 to 4 paragraphs.
DO NOT change the Max-Token setting!!!