Update dependencies, add Demo Video

README.md

@@ -3,8 +3,11 @@
 The goal of this project was to recreate Neuro-Sama, but only running on local models on consumer hardware.
 The original version was also created in only 7 days, so it is not exactly very sophisticated.
 
+DEMO VIDEO: [Youtube VOD](https://youtu.be/bywM1gyAEM4?si=706o2oialxbPEUBD&t=1073)
+
 ![Screenshot of demo stream](./images/stream.png)
 
+
 ## Features
 - Realtime STT for natural voice input
 - Realtime TTS for natural voice output

constants.py

@@ -33,14 +33,15 @@
 
 MULTIMODAL_MODEL = "openbmb/MiniCPM-Llama3-V-2_5-int4"
 
-MULTIMODAL_CONTEXT_SIZE = 8192
+MULTIMODAL_CONTEXT_SIZE = 1000 #8192 # Trying out 1000 tokens to limit short term memory
 
 # This is the multimodal strategy (when to use multimodal/text only llm) that the program will start with.
 # Runtime changes will not be saved here.
 # Valid values are: "always", "never"
-MULTIMODAL_STRATEGY = "always"
+MULTIMODAL_STRATEGY = "never"
 
 # This is the monitor index that screenshots will be taken. THIS IS NOT THE MONITOR NUMBER IN DISPLAY SETTINGS
+# Monitor 0 is a "virtual" monitor contains all monitor screens.
 PRIMARY_MONITOR = 0
 
 # LLM SPECIFIC SECTION: Below are constants that are specific to the LLM you are using

modules/vtubeStudio.py

@@ -206,7 +206,9 @@ def set_movement_status(self, status):
             self.outer.enabled = status
             self.outer.signals.sio_queue.put(('movement_status', status))
             if status:
-                self.outer.queue.clear()
+                #Clear queue
+                while not self.outer.queue.empty():
+                    self.outer.queue.get()
 
         def get_movement_status(self):
             return self.outer.enabled