From 66d4c5b90d7dc606e0c18a2b0387fecf14d7b020 Mon Sep 17 00:00:00 2001
From: kijai <40791699+kijai@users.noreply.github.com>
Date: Mon, 25 Aug 2025 12:50:41 +0300
Subject: [PATCH] Slice possible alpha channel away before encoding

---
 multitalk/nodes.py | 4 ++--
 nodes.py           | 2 ++
 2 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/multitalk/nodes.py b/multitalk/nodes.py
index cfd7428..0ecea7d 100644
--- a/multitalk/nodes.py
+++ b/multitalk/nodes.py
@@ -78,8 +78,8 @@ class MultiTalkWav2VecEmbeds:
             "normalize_loudness": ("BOOLEAN", {"default": True, "tooltip": "Normalize the audio loudness to -23 LUFS"}),
             "num_frames": ("INT", {"default": 81, "min": 1, "max": 10000, "step": 1, "tooltip": "The total frame count to generate."}),
             "fps": ("FLOAT", {"default": 25.0, "min": 1.0, "max": 60.0, "step": 0.1}),
-            "audio_scale": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 100.0, "step": 0.1, "tooltip": "Strength of the audio conditioning"}),
-            "audio_cfg_scale": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 100.0, "step": 0.1, "tooltip": "When not 1.0, an extra model pass without audio conditioning is done: slower inference but more motion is allowed"}),
+            "audio_scale": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 100.0, "step": 0.01, "tooltip": "Strength of the audio conditioning"}),
+            "audio_cfg_scale": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 100.0, "step": 0.01, "tooltip": "When not 1.0, an extra model pass without audio conditioning is done: slower inference but more motion is allowed"}),
             "multi_audio_type": (["para", "add"], {"default": "para", "tooltip": "'para' overlay speakers in parallel, 'add' concatenate sequentially"}),
         },
             "optional" : {
diff --git a/nodes.py b/nodes.py
index 8002fcc..79775f6 100644
--- a/nodes.py
+++ b/nodes.py
@@ -864,6 +864,7 @@ class WanVideoImageToVideoEncode:
 
         # Resize and rearrange the input image dimensions
         if start_image is not None:
+            start_image = start_image[..., :3]
             if start_image.shape[1] != H or start_image.shape[2] != W:
                 resized_start_image = common_upscale(start_image.movedim(-1, 1), W, H, "lanczos", "disabled").movedim(0, 1)
             else:
@@ -873,6 +874,7 @@ class WanVideoImageToVideoEncode:
                 resized_start_image = add_noise_to_reference_video(resized_start_image, ratio=noise_aug_strength)
         
         if end_image is not None:
+            end_image = end_image[..., :3]
             if end_image.shape[1] != H or end_image.shape[2] != W:
                 resized_end_image = common_upscale(end_image.movedim(-1, 1), W, H, "lanczos", "disabled").movedim(0, 1)
             else: