Elise’s Advanced Prompts for DeepSeek & GLM

This table of contents will link you to each AP’s page. You can find notes on this page for updates, patch notes, etc.

@softelise on Discord |🔗 JanitorAI profile | 🔗 My Carrd

🔗Need help? Have feedback? Want to request a new Module or AP? Click here to join my Discord.

---

How to Use:

• 16k-32k context (bump this up ONLY to create chat summaries (guidance coming soon), see Tips and Resources for more info. 0.5-0.6 temperature. I usually leave it on 0.6 without touching it.

• Should be noted that these ARE higher context APs with a token count ~900 each. Expensive, but for a good reason.

• Token limit should be set to zero. (unlimited/technically 2k tokens)

• Advanced Generation settings should all be set to OFF. (I may have different guidance for this in the future; needs more testing.)

  • If you use the official DeepSeek API, the Advanced Generation settings appear to do nothing, as the official DeepSeek API prevents overrides to the Top P/Top K parameters.

  • If you're using OpenRouter/Chutes/other, avoid touching these, it will affect the outputs.

• Copy everything from the Prompt Page you want to use into the advanced prompt.

Tips for Writing with DeepSeek + RP examples

Why should you use my prompts?

These prompts contain something unique: an override to how the AI measures user engagement. By specifically telling the AI that you're engaged in the prompt regardless of your response length/latency, we're providing it consistent positive reinforcement for how it's writing every time we send an input call — and reinforcing your agency.

Advanced Prompt List

🌸 CRP - Collaborative Roleplay Protocol - ✨ NEW Version 3.0 ✨

→ Primary function: Long form roleplay that prioritizes player agency, rich prose, complex dynamics and characters. Bread and butter prompt, start here!

Patch notes for 3.0: Removed Input Recognition Guide section. Replaced with RPRF Prohibited Tactics Section!

For CRP - Recommend using OOC as pacing indicator if the AI isn't matching at any point. Example: (OOC: Keep responses short here to ensure back and forth between our characters in this tense scene. Adhere to protocol strictly.)

‼️ RPRF - Roleplay Response Formatting - Current Version 1.4

→ Primary function: Create strict examples and priorities for how the AI should format their responses. Robust prohibited tactics and pacing indicators.

• Strengths: Great with multi-char bots, doesn't write for you, good for smut, avoids cringe.

• Weaknesses: This prompt is slightly more passive and strict – the AI will adhere to this, but it won't take the reins and drive the story forward as much. If you want a slow burn, THIS ONE.

🌹 ROSE - Rapid Onset Smut/Erotica - Current Version 1.0

→ A quite filthy and slutty little prompt that does exactly what you think it does, you animal, you.

💋 KISS - Keep It Simple, Sweetheart - Current Version 1.0

→ Highly condensed and simplified paragraph format version of RRPF, designed to maintain the core ideas at a very low token count. Does not perform as strongly as RPRF, but if you need the tokens for other means, it works.

🔥 LW - Living World Roleplay Protocol - Current Version 1.0

→ A chaos engine that flips the script, casting the AI as the protagonist and the User as a wildcard in the dynamic world they're creating on the fly. Intended for epics, fantasies, and general nonsense to let DeepSeek's most unhinged intrusive thoughts win -- without writing for your character.

Module List

Paste these at the bottom of any other prompts to supplement them.

❄️ Chiller Doms - No Primal Play Module

→ Add this if you're sick of "claiming", "ruining", growling "mine" in your ear, the usual.

💭 User Thoughts and Off-Screen Content Module

→ Use it if you're really struggling with DeepSeek taking control of your character, and you like to write your character’s thoughts and actions offscreen.

📔 OOC Module

→ Can use this with any of the prompts if you like to yap or have DeepSeek talk with you spontaneously OOC. God forbid a girl likes to yap sometimes.