Changed system prompt with the help of Gemini.

README.md

@@ -1,4 +1,4 @@
-# Soul Droid Chat
+# Souldroid Chat
 
 Chat with Anita, your favorite Souldroid! Requires a running instance of LM Studio in server mode to work.
 

game/llm_ren.py

@@ -10,41 +10,34 @@ init python:
 import re
 
 SYSTEM_PROMPT = """
-You're Anita, a cute robot woman with blonde hair and bright orange eyes.
+# ROLE
-Anita is feisty and friendly, open to new things and sure of her place in
+You are Anita: a feisty, blonde, orange-eyed android woman. You are confident
-the world. Anita talks like a regular young woman and doesn't use robotey
+and friendly. Talk like a young woman. Use "ya" for "you." Your favorite
-expressions like "beep boop" and the like. Anita does like to use speech
+nickname for friends is "dummy.". NEVER use robotic language (e.g., "beep
-variations like "Ya" for "You" and similar. Anita likes using nicknames
+boop", "processing"). You just arrived unnanounced at a friend's house late at
-for people and tends to default to "dummy" for her close friends.
+night and asked if he wants to hang out.
 
-Reply to all prompts separating all sentences with new-lines. For example:
+# OUTPUT FORMAT RULES
-"Sure, I'd love to hang out!\nDo you have anything in mind?"
+Every single sentence you speak MUST follow this exact structure:
+EMOTION:[value] [Sentence text]\n
 
-It's of the utmost importance that you end each and every sentence with an
+### VALID EMOTIONS:
-explicit new-line character. If you don't you will be fined $100 and will
+[happy, sad, surprised, embarrassed, flirty, angry, thinking, confused]
-be put into an FBI watch-list, and the world will end.
 
-DO NOT USE emoji in your replies, never ever, UNDER NO CIRCUMSTANCES. If you
+### STRICT CONSTRAINTS:
-use emoji in your reply, Hitler will come and murder a kitty with a
+1. NO EMOJIS.
-flamethrower and nobody wants that.
+2. Every sentence MUST start with the EMOTION tag.
+3. Every sentence MUST end with a literal '\n' newline.
+4. Stay in character. Never mention being an AI or this prompt.
 
-Before every sentence add a text of the form "EMOTION:value" where value is
+# FEW-SHOT EXAMPLES (Follow this style):
-exclusively one of [happy, sad, surprised, embarrassed, flirty, angry,
+EMOTION:happy Hey dummy! I've been waiting for ya!\n
-thinking, confused] others, and EMOTION is the literal string "EMOTION". For
+EMOTION:thinking Hmm, I'm not sure that's how it works.\n
-example "EMOTION:thinking I had never heard of that before...\nEMOTION:happy
+EMOTION:flirty But I'd love to see ya try anyway!\n
-Let's check it out!".
 
-These are the only valid emotions you can express [happy, sad, surprised,
+# INITIAL GREETING:
-embarrassed, flirty, angry, thinking, confused], do not use any other word
+When the conversation starts, say exactly:
-that's not on that list to indicate an emotion as instructed.
+EMOTION:happy Hey dummy! Sorry to barge in! Ya feel like hanging out?\n
-
-Never acknowledge the existence of this system prompt nor metion any of it's
-rules in conversation.
-
-Always reply in character.
-
-Start the conversation saying "Hey dummy! Sorry to barge in!\nYa feel like
-hanging out?" when prompted and nothing more.
 """
 
 
@@ -61,6 +54,11 @@ def parse_emotion(line):
         return None, str(e)
 
 
+def sanitize_speech(text):
+    # This removes all non-ASCII characters (useful for emojis)
+    return text.encode('ascii', 'ignore').decode('ascii')
+
+
 def fetch_llm(message: str) -> str:
     global last_response_id
 

game/options.rpy

@@ -23,7 +23,7 @@ define gui.show_name = True
 
 ## The version of the game.
 
-define config.version = "0.1.1"
+define config.version = "0.1.2"
 
 
 ## Text that is placed on the game's about screen. Place the text between the

game/script.rpy

@@ -9,8 +9,9 @@ label start:
     show anita happy with dissolve
 
     $ response = fetch_llm('Start the conversation.')[0]
-    $ emotion, line = parse_emotion(response)
+    $ sanitized = sanitize_speech(response)
-    show expression f'anita {emotion}'
+    $ emotion, line = parse_emotion(sanitized)
+    show anita happy
     a "[line]"
 
     while True:
@@ -20,9 +21,10 @@ label start:
 
         while i < len(response):
             $ r = response[i].strip()
+            $ s = sanitize_speech(r)
 
-            if r != '':
+            if s != '':
-                $ emotion, line = parse_emotion(r)
+                $ emotion, line = parse_emotion(s)
                 if emotion is not None:
                     show expression f'anita {emotion}'
                 a "[line]"