fix(ditto): Rollback to v1.9.31 + Add Vitality Noise Injection (v1.9.41)

libs/Ditto/core/atomic_components/audio2motion.py

@@ -148,7 +148,14 @@ class Audio2Motion:
     
     def _update_kp_cond(self, res_kp_seq, idx):
         if self.fix_kp_cond == 0:  # 不重置
-            self.kp_cond = res_kp_seq[:, idx-1]
+            # [Vitality Fix v1.9.41]
+            # Problem: Feeding back smoothed output causes variance decay (EMA effect), leading to stasis.
+            # Solution: Inject micro-noise to keep the 'cascading generation' alive.
+            last_pose = res_kp_seq[:, idx-1]
+            noise_scale = 0.002 # Subtle tremor
+            noise = np.random.normal(0, noise_scale, last_pose.shape).astype(np.float32)
+            self.kp_cond = last_pose + noise
+            
         elif self.fix_kp_cond > 0:
             if self.clip_idx % self.fix_kp_cond == 0:  # 重置
                 self.kp_cond = self.s_kp_cond.copy()  # 重置所有

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 = "1.9.31"
+version = "1.9.41"
 license = {file = "LICENSE"}
 readme = "README.md"
 authors = [