v1.9.192: Deactivated Volumetric Jaw Gain to fix mouth asymmetry. Maintained pitch stability optimizations.

libs/Ditto/core/atomic_components/audio2motion.py

@@ -387,7 +387,7 @@ class Audio2Motion:
             
             if self.clip_idx % 20 == 0:
                  mode_s = "SPEECH" if is_talking else "IDLE"
-                 print(f"[v1.9.190 {mode_s}] Pressure: {pressure*100:.0f}% (Delta={self.delta_p:+.2f} Target={self.target_bias_deg:+.1f})")
+                 print(f"[v1.9.192 {mode_s}] Pressure: {pressure*100:.0f}% (Delta={self.delta_p:+.2f} Target={self.target_bias_deg:+.1f})")
         
         # [v1.9.156] Virtual Last Frame for Startup Stabilization
         # If this is the VERY first chunk, we treat the source photo as the "prev frame"

libs/Ditto/core/atomic_components/motion_stitch.py

@@ -533,8 +533,8 @@ class MotionStitch:
         self.fix_exp_a2 = (1 - _a1) + _a1 * _a2
         self.fix_exp_a3 = _a2
         
-        # [Debug v1.9.190] Verify Code Sync
-        print(f"[AIIA Debug] MotionStitch Setup: v1.9.190. LATEST VERSION LOADED.")
+        # [Debug v1.9.192] Verify Code Sync
+        print(f"[AIIA Debug] MotionStitch Setup: v1.9.192. LATEST VERSION LOADED.")
 
 
         if self.drive_eye and self.delta_eye_arr is not None:
@@ -701,8 +701,9 @@ class MotionStitch:
         # Only boost if the AI is actually trying to open the mouth (y > 0)
         mask = y_intent > 0.001
         if np.any(mask):
-            # [v1.9.190] Volumetric Jaw Excitation (Boosted to 1.5)
-            gain = 1.0 + 1.5 * np.exp(-y_intent[mask] / 0.04)
+            # [Reverted v1.9.192] Removed Volumetric Jaw Excitation
+            # We revert to 1.0 gain to prevent potential mouth asymmetry (crooked mouth)
+            gain = 1.0 
             exp_reshaped[:, lower_lip, 1][mask] *= gain
 
         # 2. Volumetric Mouth Micro-Motion (Breathing + Corners 7,8)

pyproject.toml

@@ -1,7 +1,7 @@
 [project]
 name = "aiia"
 description = "The Ultimate AI Audio/Video toolkit for ComfyUI. Features an enhanced Ditto (with optimizations that outperform official demos and other SOTA talking head models in lip-sync accuracy and natural motion), EchoMimic V3 & FLOAT, VibeVoice & CosyVoice 3.0 (Zero-Shot Voice Cloning), Multi-Role Podcast Generation, and a powerful Media Browser."
-version = "1.9.191"
+version = "1.9.192"
 license = {file = "LICENSE"}
 readme = "README.md"
 authors = [