Merge 9ee905bc47 into acbf08cd60

comfy/model_base.py

@@ -1248,22 +1248,23 @@ class WAN21_HuMo(WAN21):
         if audio_embed is not None:
             out['audio_embed'] = comfy.conds.CONDRegular(audio_embed)
 
-        if "c_concat" not in out:  # 1.7B model
             reference_latents = kwargs.get("reference_latents", None)
-            if reference_latents is not None:
+
+            if "c_concat" not in out and reference_latents is not None and reference_latents[0].shape[1] == 16:  # 1.7B model
                 out['reference_latent'] = comfy.conds.CONDRegular(self.process_latent_in(reference_latents[-1]))
-        else:
-            noise_shape = list(noise.shape)
-            noise_shape[1] += 4
-            concat_latent = torch.zeros(noise_shape, device=noise.device, dtype=noise.dtype)
-            zero_vae_values_first = torch.tensor([0.8660, -0.4326, -0.0017, -0.4884, -0.5283, 0.9207, -0.9896, 0.4433, -0.5543, -0.0113, 0.5753, -0.6000, -0.8346, -0.3497, -0.1926, -0.6938]).view(1, 16, 1, 1, 1)
-            zero_vae_values_second = torch.tensor([1.0869, -1.2370, 0.0206, -0.4357, -0.6411, 2.0307, -1.5972, 1.2659, -0.8595, -0.4654, 0.9638, -1.6330, -1.4310, -0.1098, -0.3856, -1.4583]).view(1, 16, 1, 1, 1)
-            zero_vae_values = torch.tensor([0.8642, -1.8583, 0.1577, 0.1350, -0.3641, 2.5863, -1.9670, 1.6065, -1.0475, -0.8678, 1.1734, -1.8138, -1.5933, -0.7721, -0.3289, -1.3745]).view(1, 16, 1, 1, 1)
-            concat_latent[:, 4:] = zero_vae_values
-            concat_latent[:, 4:, :1] = zero_vae_values_first
-            concat_latent[:, 4:, 1:2] = zero_vae_values_second
-            out['c_concat'] = comfy.conds.CONDNoiseShape(concat_latent)
-            reference_latents = kwargs.get("reference_latents", None)
+            else:
+                concat_latent_image = kwargs.get("concat_latent_image", None)
+                if concat_latent_image is None:
+                    noise_shape = list(noise.shape)
+                    noise_shape[1] += 4
+                    concat_latent = torch.zeros(noise_shape, device=noise.device, dtype=noise.dtype)
+                    zero_vae_values_first = torch.tensor([0.8660, -0.4326, -0.0017, -0.4884, -0.5283, 0.9207, -0.9896, 0.4433, -0.5543, -0.0113, 0.5753, -0.6000, -0.8346, -0.3497, -0.1926, -0.6938]).view(1, 16, 1, 1, 1)
+                    zero_vae_values_second = torch.tensor([1.0869, -1.2370, 0.0206, -0.4357, -0.6411, 2.0307, -1.5972, 1.2659, -0.8595, -0.4654, 0.9638, -1.6330, -1.4310, -0.1098, -0.3856, -1.4583]).view(1, 16, 1, 1, 1)
+                    zero_vae_values = torch.tensor([0.8642, -1.8583, 0.1577, 0.1350, -0.3641, 2.5863, -1.9670, 1.6065, -1.0475, -0.8678, 1.1734, -1.8138, -1.5933, -0.7721, -0.3289, -1.3745]).view(1, 16, 1, 1, 1)
+                    concat_latent[:, 4:] = zero_vae_values
+                    concat_latent[:, 4:, :1] = zero_vae_values_first
+                    concat_latent[:, 4:, 1:2] = zero_vae_values_second
+                    out['c_concat'] = comfy.conds.CONDNoiseShape(concat_latent)
             if reference_latents is not None:
                 ref_latent = self.process_latent_in(reference_latents[-1])
                 ref_latent_shape = list(ref_latent.shape)

comfy_api_nodes/nodes_kling.py

@@ -807,6 +807,7 @@ class OmniProTextToVideoNode(IO.ComfyNode):
                 ),
                 IO.Combo.Input("aspect_ratio", options=["16:9", "9:16", "1:1"]),
                 IO.Combo.Input("duration", options=[5, 10]),
+                IO.Combo.Input("resolution", options=["1080p", "720p"], optional=True),
             ],
             outputs=[
                 IO.Video.Output(),
@@ -826,6 +827,7 @@ class OmniProTextToVideoNode(IO.ComfyNode):
         prompt: str,
         aspect_ratio: str,
         duration: int,
+        resolution: str = "1080p",
     ) -> IO.NodeOutput:
         validate_string(prompt, min_length=1, max_length=2500)
         response = await sync_op(
@@ -837,6 +839,7 @@ class OmniProTextToVideoNode(IO.ComfyNode):
                 prompt=prompt,
                 aspect_ratio=aspect_ratio,
                 duration=str(duration),
+                mode="pro" if resolution == "1080p" else "std",
             ),
         )
         return await finish_omni_video_task(cls, response)
@@ -872,6 +875,7 @@ class OmniProFirstLastFrameNode(IO.ComfyNode):
                     optional=True,
                     tooltip="Up to 6 additional reference images.",
                 ),
+                IO.Combo.Input("resolution", options=["1080p", "720p"], optional=True),
             ],
             outputs=[
                 IO.Video.Output(),
@@ -893,6 +897,7 @@ class OmniProFirstLastFrameNode(IO.ComfyNode):
         first_frame: Input.Image,
         end_frame: Input.Image | None = None,
         reference_images: Input.Image | None = None,
+        resolution: str = "1080p",
     ) -> IO.NodeOutput:
         prompt = normalize_omni_prompt_references(prompt)
         validate_string(prompt, min_length=1, max_length=2500)
@@ -936,6 +941,7 @@ class OmniProFirstLastFrameNode(IO.ComfyNode):
                 prompt=prompt,
                 duration=str(duration),
                 image_list=image_list,
+                mode="pro" if resolution == "1080p" else "std",
             ),
         )
         return await finish_omni_video_task(cls, response)
@@ -964,6 +970,7 @@ class OmniProImageToVideoNode(IO.ComfyNode):
                     "reference_images",
                     tooltip="Up to 7 reference images.",
                 ),
+                IO.Combo.Input("resolution", options=["1080p", "720p"], optional=True),
             ],
             outputs=[
                 IO.Video.Output(),
@@ -984,6 +991,7 @@ class OmniProImageToVideoNode(IO.ComfyNode):
         aspect_ratio: str,
         duration: int,
         reference_images: Input.Image,
+        resolution: str = "1080p",
     ) -> IO.NodeOutput:
         prompt = normalize_omni_prompt_references(prompt)
         validate_string(prompt, min_length=1, max_length=2500)
@@ -1005,6 +1013,7 @@ class OmniProImageToVideoNode(IO.ComfyNode):
                 aspect_ratio=aspect_ratio,
                 duration=str(duration),
                 image_list=image_list,
+                mode="pro" if resolution == "1080p" else "std",
             ),
         )
         return await finish_omni_video_task(cls, response)
@@ -1036,6 +1045,7 @@ class OmniProVideoToVideoNode(IO.ComfyNode):
                     tooltip="Up to 4 additional reference images.",
                     optional=True,
                 ),
+                IO.Combo.Input("resolution", options=["1080p", "720p"], optional=True),
             ],
             outputs=[
                 IO.Video.Output(),
@@ -1058,6 +1068,7 @@ class OmniProVideoToVideoNode(IO.ComfyNode):
         reference_video: Input.Video,
         keep_original_sound: bool,
         reference_images: Input.Image | None = None,
+        resolution: str = "1080p",
     ) -> IO.NodeOutput:
         prompt = normalize_omni_prompt_references(prompt)
         validate_string(prompt, min_length=1, max_length=2500)
@@ -1090,6 +1101,7 @@ class OmniProVideoToVideoNode(IO.ComfyNode):
                 duration=str(duration),
                 image_list=image_list if image_list else None,
                 video_list=video_list,
+                mode="pro" if resolution == "1080p" else "std",
             ),
         )
         return await finish_omni_video_task(cls, response)
@@ -1119,6 +1131,7 @@ class OmniProEditVideoNode(IO.ComfyNode):
                     tooltip="Up to 4 additional reference images.",
                     optional=True,
                 ),
+                IO.Combo.Input("resolution", options=["1080p", "720p"], optional=True),
             ],
             outputs=[
                 IO.Video.Output(),
@@ -1139,6 +1152,7 @@ class OmniProEditVideoNode(IO.ComfyNode):
         video: Input.Video,
         keep_original_sound: bool,
         reference_images: Input.Image | None = None,
+        resolution: str = "1080p",
     ) -> IO.NodeOutput:
         prompt = normalize_omni_prompt_references(prompt)
         validate_string(prompt, min_length=1, max_length=2500)
@@ -1171,6 +1185,7 @@ class OmniProEditVideoNode(IO.ComfyNode):
                 duration=None,
                 image_list=image_list if image_list else None,
                 video_list=video_list,
+                mode="pro" if resolution == "1080p" else "std",
             ),
         )
         return await finish_omni_video_task(cls, response)