test

comfy/ldm/wan/model.py

@@ -440,7 +440,7 @@ class WanModel(torch.nn.Module):
 
         # embeddings
         self.patch_embedding = operations.Conv3d(
-            in_dim, dim, kernel_size=patch_size, stride=patch_size, device=operation_settings.get("device"), dtype=torch.float32)
+            20, dim, kernel_size=patch_size, stride=patch_size, device=operation_settings.get("device"), dtype=torch.float32)
         self.text_embedding = nn.Sequential(
             operations.Linear(text_dim, dim, device=operation_settings.get("device"), dtype=operation_settings.get("dtype")), nn.GELU(approximate='tanh'),
             operations.Linear(dim, dim, device=operation_settings.get("device"), dtype=operation_settings.get("dtype")))

comfy/model_base.py

@@ -1064,6 +1064,7 @@ class WAN21(BaseModel):
     def concat_cond(self, **kwargs):
         noise = kwargs.get("noise", None)
         extra_channels = self.diffusion_model.patch_embedding.weight.shape[1] - noise.shape[1]
+        print(f"extra channels: {extra_channels}, noise shape: {noise.shape}, patch embedding weight shape: {self.diffusion_model.patch_embedding.weight.shape}")
         if extra_channels == 0:
             return None
 
@@ -1080,6 +1081,7 @@ class WAN21(BaseModel):
                 image[:, i: i + 16] = self.process_latent_in(image[:, i: i + 16])
             image = utils.resize_to_batch_size(image, noise.shape[0])
 
+        print(f"image shape: {image.shape}")
         if not self.image_to_video or extra_channels == image.shape[1]:
             return image
 
@@ -1098,8 +1100,10 @@ class WAN21(BaseModel):
                 mask = torch.nn.functional.pad(mask, (0, 0, 0, 0, 0, noise.shape[-3] - mask.shape[-3]), mode='constant', value=0)
             if mask.shape[1] == 1:
                 mask = mask.repeat(1, 4, 1, 1, 1)
+            
+            print(f"Mask shape: {mask.shape}, noise shape: {noise.shape}")
             mask = utils.resize_to_batch_size(mask, noise.shape[0])
-
+        print(f"image shape: {image.shape}, mask shape: {mask.shape}")
         return torch.cat((mask, image), dim=1)
 
     def extra_conds(self, **kwargs):

comfy_extras/nodes_wan.py

@@ -657,7 +657,6 @@ class WanTrackToVideo:
                 zero_frames = torch.zeros(3, 81 - 1, height, width)
                 
                 start_image = start_image.permute(3,0,1,2)  # C, T, H, W
-                start_image = start_image * 2 - 1
                 res = torch.concat([
                         start_image.to(start_image.device),
                         zero_frames
@@ -668,7 +667,7 @@ class WanTrackToVideo:
                     res.permute(1,2,3,0)[:, :, :, :3]  # T, H, W, C
                 )[0]
                 y = torch.concat([msk, y])
-
+                print('?')
                 motion_patched = patch_motion(processed_tracks, y, temperature, (4, 16), topk)[None]
                 mask, video = motion_patched[:, 0:4], motion_patched[:, 4:]
                 # Add motion features to conditioning