Merge pull request #95 from CelebrityPunks/fix/model-size-selection-ignored

Fix: selecting 0.6B model still downloads and uses 1.7B

backend/main.py

@@ -600,6 +600,8 @@ async def generate_speech(
         if not profile:
             raise HTTPException(status_code=404, detail="Profile not found")
         
+        # Generate audio
+
         # Resolve model size and load the correct model FIRST.
         # This must happen before create_voice_prompt_for_profile because that
         # function calls load_model_async(None), which falls back to self.model_size.