Support lucy edit

nodes.py

@@ -1230,6 +1230,25 @@ class WanVideoAddExtraLatent:
         updated = dict(embeds)
         updated["extra_latents"] = extra_latents_list
         return (updated,)
+    
+class WanVideoAddLucyEditLatents:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": {
+                    "embeds": ("WANVIDIMAGE_EMBEDS",),
+                    "extra_latents": ("LATENT",),
+                }
+        }
+
+    RETURN_TYPES = ("WANVIDIMAGE_EMBEDS",)
+    RETURN_NAMES = ("image_embeds",)
+    FUNCTION = "add"
+    CATEGORY = "WanVideoWrapper"
+
+    def add(self, embeds, extra_latents):
+        updated = dict(embeds)
+        updated["extra_channel_latents"] = extra_latents["samples"]
+        return (updated,)
 
 class WanVideoMiniMaxRemoverEmbeds:
     @classmethod
@@ -2126,6 +2145,7 @@ NODE_CLASS_MAPPINGS = {
     "WanVideoRoPEFunction": WanVideoRoPEFunction,
     "WanVideoAddPusaNoise": WanVideoAddPusaNoise,
     "WanVideoAnimateEmbeds": WanVideoAnimateEmbeds,
+    "WanVideoAddLucyEditLatents": WanVideoAddLucyEditLatents,
     }
 
 NODE_DISPLAY_NAME_MAPPINGS = {
@@ -2163,4 +2183,5 @@ NODE_DISPLAY_NAME_MAPPINGS = {
     "WanVideoRoPEFunction": "WanVideo RoPE Function",
     "WanVideoAddPusaNoise": "WanVideo Add Pusa Noise",
     "WanVideoAnimateEmbeds": "WanVideo Animate Embeds",
+    "WanVideoAddLucyEditLatents": "WanVideo Add LucyEdit Latents",
 }

nodes_sampler.py

@@ -791,6 +791,11 @@ class WanVideoSampler:
                     noise_multipliers[idx] = noise_multiplier_list[i]
                 log.info(f"Using Pusa noise multipliers: {noise_multipliers}")
 
+        # lucy edit
+        extra_channel_latents = image_embeds.get("extra_channel_latents", None)
+        if extra_channel_latents is not None:
+            extra_channel_latents = extra_channel_latents[0].to(noise)
+
         latent = noise.to(device)
 
         #controlnet
@@ -1171,6 +1176,14 @@ class WanVideoSampler:
                             humo_audio_input_neg = None
                 else:
                     humo_audio_input = humo_audio_input_neg = None
+
+                if extra_channel_latents is not None:
+                    if context_window is not None:
+                        extra_channel_latents_input = extra_channel_latents[:, context_window].to(z)
+                    else:
+                        extra_channel_latents_input = extra_channel_latents.to(z)
+                    z = torch.cat([z, extra_channel_latents_input])
+
                 base_params = {
                     'x': [z], # latent
                     'y': [image_cond_input] if image_cond_input is not None else None, # image cond