nodes/havvk-ComfyUI_AIIA
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

v1.9.111: Reverting Squint & Tuning Mouth

Browse Source
This commit is contained in:
Hawk Lee
2026-01-24 09:42:05 +08:00
parent e22817e810
commit 75a965ebdc
2 changed files with 17 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files
libs/Ditto/core/atomic_components/motion_stitch.py
+16 -14
View File
@@ -299,7 +299,8 @@ def _fix_exp_for_x_d_info(x_d_info, x_s_info, delta_eye=None, drive_eye=True):
_lip = [6, 7, 8, 12, 14, 17, 18, 19, 20] # [Updated v1.9.110] Added 7, 8 (Corners) and 18 to allow fuller mouth opening.
_lip = [6, 12, 14, 17, 19, 20] # [Reverted v1.9.111] Back to official indices to fix unintended brow/eye pull.
# Note: 7/8 (Corners) and 18 (Brows) removed to avoid "fixed squint/smile" look.
alpha = np.zeros((21, 3), dtype=x_d_info["exp"].dtype)
alpha[_lip] = 1
@@ -697,18 +698,19 @@ class MotionStitch:
x_d_info["exp"] = self.prev_exp_ema * exp_decay + x_d_info["exp"] * (1.0 - exp_decay)
self.prev_exp_ema = x_d_info["exp"].copy()
# [v1.9.109/v1.9.110] Spatial Mouth Opening Bias
# Force lower lip down to show teeth and fix pucker.
# Points 17-20 are the lower lip center chain.
# We target index 1 (Y-axis displacement).
# [v1.9.111] Precise Mouth Opening Bias
# 1. Reverted Point 18 (it controls brows, not lips).
# 2. Reverted Point 7/8 (mouth corners).
# 3. Made bias speech-dependent: 0.0 in silence, up to 0.05 in speech.
bias_alpha = kwargs.get("vad_alpha", 0.0)
bias_val = 0.03 + (0.05 * bias_alpha) # Base 0.03 + Speech Boost 0.05 = 0.08 max
# Point to spatial indexing: exp is (1, 21, 3) or (1, 63)
# We reshape to be safe
exp_reshaped = x_d_info["exp"].reshape(-1, 21, 3)
exp_reshaped[:, [17, 18, 19, 20], 1] += bias_val
x_d_info["exp"] = exp_reshaped.reshape(1, -1)
if bias_alpha > 0.05:
# Scale bias by speech intensity to allow natural closure
bias_val = 0.05 * bias_alpha
lower_lip = [17, 19, 20] # Points 17, 19, 20 are the true lower-lip center chain
exp_reshaped = x_d_info["exp"].reshape(-1, 21, 3)
# Index 1 = Y-axis (vertical)
exp_reshaped[:, lower_lip, 1] += bias_val
x_d_info["exp"] = exp_reshaped.reshape(1, -1)
delta_eye = 0
if self.drive_eye and self.delta_eye_arr is not None:
@@ -725,8 +727,8 @@ class MotionStitch:
# [Feature v1.9.48] Apple Mouth Micro-Motion
if "delta_mouth" in kwargs:
_lip = [6, 7, 8, 12, 14, 17, 18, 19, 20]
# Broadcasting scalar to (1, 9, 3)
_lip = [6, 12, 14, 17, 19, 20]
# Broadcasting scalar to (1, 6, 3)
exp_reshaped = x_d_info["exp"].reshape(-1, 21, 3)
exp_reshaped[:, _lip] += kwargs["delta_mouth"]
x_d_info["exp"] = exp_reshaped.reshape(1, -1)
pyproject.toml
+1 -1
View File
@@ -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 = "1.9.110"
version = "1.9.111"
license = {file = "LICENSE"}
readme = "README.md"
authors = [