ComfyUI version 0.3.56

Use the WanFunInpaintToVideo node without the clip_vision_output.

comfy/model_base.py

@@ -1110,9 +1110,10 @@ class WAN21(BaseModel):
             shape_image[1] = extra_channels
             image = torch.zeros(shape_image, dtype=noise.dtype, layout=noise.layout, device=noise.device)
         else:
+            latent_dim = self.latent_format.latent_channels
             image = utils.common_upscale(image.to(device), noise.shape[-1], noise.shape[-2], "bilinear", "center")
-            for i in range(0, image.shape[1], 16):
-                image[:, i: i + 16] = self.process_latent_in(image[:, i: i + 16])
+            for i in range(0, image.shape[1], latent_dim):
+                image[:, i: i + latent_dim] = self.process_latent_in(image[:, i: i + latent_dim])
             image = utils.resize_to_batch_size(image, noise.shape[0])
 
         if extra_channels != image.shape[1] + 4:
@@ -1245,18 +1246,14 @@ class WAN22_S2V(WAN21):
             out['reference_motion'] = reference_motion.shape
         return out
 
-class WAN22(BaseModel):
+class WAN22(WAN21):
     def __init__(self, model_config, model_type=ModelType.FLOW, image_to_video=False, device=None):
-        super().__init__(model_config, model_type, device=device, unet_model=comfy.ldm.wan.model.WanModel)
+        super(WAN21, self).__init__(model_config, model_type, device=device, unet_model=comfy.ldm.wan.model.WanModel)
         self.image_to_video = image_to_video
 
     def extra_conds(self, **kwargs):
         out = super().extra_conds(**kwargs)
-        cross_attn = kwargs.get("cross_attn", None)
-        if cross_attn is not None:
-            out['c_crossattn'] = comfy.conds.CONDRegular(cross_attn)
-
-        denoise_mask = kwargs.get("concat_mask", kwargs.get("denoise_mask", None))
+        denoise_mask = kwargs.get("denoise_mask", None)
         if denoise_mask is not None:
             out["denoise_mask"] = comfy.conds.CONDRegular(denoise_mask)
         return out

comfy_api/latest/_input_impl/video_types.py

@@ -8,6 +8,7 @@ import av
 import io
 import json
 import numpy as np
+import math
 import torch
 from comfy_api.latest._util import VideoContainer, VideoCodec, VideoComponents
 
@@ -282,8 +283,6 @@ class VideoFromComponents(VideoInput):
             if self.__components.audio:
                 audio_sample_rate = int(self.__components.audio['sample_rate'])
                 audio_stream = output.add_stream('aac', rate=audio_sample_rate)
-                audio_stream.sample_rate = audio_sample_rate
-                audio_stream.format = 'fltp'
 
             # Encode video
             for i, frame in enumerate(self.__components.images):
@@ -298,27 +297,12 @@ class VideoFromComponents(VideoInput):
             output.mux(packet)
 
             if audio_stream and self.__components.audio:
-                # Encode audio
-                samples_per_frame = int(audio_sample_rate / frame_rate)
-                num_frames = self.__components.audio['waveform'].shape[2] // samples_per_frame
-                for i in range(num_frames):
-                    start = i * samples_per_frame
-                    end = start + samples_per_frame
-                    # TODO(Feature) - Add support for stereo audio
-                    chunk = (
-                        self.__components.audio["waveform"][0, 0, start:end]
-                        .unsqueeze(0)
-                        .contiguous()
-                        .numpy()
-                    )
-                    audio_frame = av.AudioFrame.from_ndarray(chunk, format='fltp', layout='mono')
-                    audio_frame.sample_rate = audio_sample_rate
-                    audio_frame.pts = i * samples_per_frame
-                    for packet in audio_stream.encode(audio_frame):
-                        output.mux(packet)
-
-                # Flush audio
-                for packet in audio_stream.encode(None):
-                    output.mux(packet)
-
+                waveform = self.__components.audio['waveform']
+                waveform = waveform[:, :, :math.ceil((audio_sample_rate / frame_rate) * self.__components.images.shape[0])]
+                frame = av.AudioFrame.from_ndarray(waveform.movedim(2, 1).reshape(1, -1).float().numpy(), format='flt', layout='mono' if waveform.shape[1] == 1 else 'stereo')
+                frame.sample_rate = audio_sample_rate
+                frame.pts = 0
+                output.mux(audio_stream.encode(frame))
 
+                # Flush encoder
+                output.mux(audio_stream.encode(None))

comfy_extras/nodes_latent.py

@@ -1,6 +1,7 @@
 import comfy.utils
 import comfy_extras.nodes_post_processing
 import torch
+import nodes
 
 
 def reshape_latent_to(target_shape, latent, repeat_batch=True):
@@ -137,6 +138,41 @@ class LatentConcat:
         samples_out["samples"] = torch.cat(c, dim=dim)
         return (samples_out,)
 
+class LatentCut:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": {"samples": ("LATENT",),
+                             "dim": (["x", "y", "t"], ),
+                             "index": ("INT", {"default": 0, "min": -nodes.MAX_RESOLUTION, "max": nodes.MAX_RESOLUTION, "step": 1}),
+                             "amount": ("INT", {"default": 1, "min": 1, "max": nodes.MAX_RESOLUTION, "step": 1})}}
+
+    RETURN_TYPES = ("LATENT",)
+    FUNCTION = "op"
+
+    CATEGORY = "latent/advanced"
+
+    def op(self, samples, dim, index, amount):
+        samples_out = samples.copy()
+
+        s1 = samples["samples"]
+
+        if "x" in dim:
+            dim = s1.ndim - 1
+        elif "y" in dim:
+            dim = s1.ndim - 2
+        elif "t" in dim:
+            dim = s1.ndim - 3
+
+        if index >= 0:
+            index = min(index, s1.shape[dim] - 1)
+            amount = min(s1.shape[dim] - index, amount)
+        else:
+            index = max(index, -s1.shape[dim])
+            amount = min(-index, amount)
+
+        samples_out["samples"] = torch.narrow(s1, dim, index, amount)
+        return (samples_out,)
+
 class LatentBatch:
     @classmethod
     def INPUT_TYPES(s):
@@ -312,6 +348,7 @@ NODE_CLASS_MAPPINGS = {
     "LatentMultiply": LatentMultiply,
     "LatentInterpolate": LatentInterpolate,
     "LatentConcat": LatentConcat,
+    "LatentCut": LatentCut,
     "LatentBatch": LatentBatch,
     "LatentBatchSeedBehavior": LatentBatchSeedBehavior,
     "LatentApplyOperation": LatentApplyOperation,

comfy_extras/nodes_model_patch.py

@@ -108,7 +108,7 @@ class DiffSynthCnetPatch:
         img = kwargs.get("img")
         block_index = kwargs.get("block_index")
         spacial_compression = self.vae.spacial_compression_encode()
-        if self.encoded_image is None or self.encoded_image_size != (x.shape[-1] * spacial_compression, x.shape[-2] * spacial_compression):
+        if self.encoded_image is None or self.encoded_image_size != (x.shape[-2] * spacial_compression, x.shape[-1] * spacial_compression):
             image_scaled = comfy.utils.common_upscale(self.image.movedim(-1, 1), x.shape[-1] * spacial_compression, x.shape[-2] * spacial_compression, "area", "center")
             loaded_models = comfy.model_management.loaded_models(only_currently_used=True)
             self.encoded_image = self.model_patch.model.process_input_latent_image(self.encode_latent_cond(image_scaled.movedim(1, -1)))

comfy_extras/nodes_wan.py

@@ -139,16 +139,21 @@ class Wan22FunControlToVideo(io.ComfyNode):
 
     @classmethod
     def execute(cls, positive, negative, vae, width, height, length, batch_size, ref_image=None, start_image=None, control_video=None) -> io.NodeOutput:
-        latent = torch.zeros([batch_size, 16, ((length - 1) // 4) + 1, height // 8, width // 8], device=comfy.model_management.intermediate_device())
-        concat_latent = torch.zeros([batch_size, 16, ((length - 1) // 4) + 1, height // 8, width // 8], device=comfy.model_management.intermediate_device())
-        concat_latent = comfy.latent_formats.Wan21().process_out(concat_latent)
+        spacial_scale = vae.spacial_compression_encode()
+        latent_channels = vae.latent_channels
+        latent = torch.zeros([batch_size, latent_channels, ((length - 1) // 4) + 1, height // spacial_scale, width // spacial_scale], device=comfy.model_management.intermediate_device())
+        concat_latent = torch.zeros([batch_size, latent_channels, ((length - 1) // 4) + 1, height // spacial_scale, width // spacial_scale], device=comfy.model_management.intermediate_device())
+        if latent_channels == 48:
+            concat_latent = comfy.latent_formats.Wan22().process_out(concat_latent)
+        else:
+            concat_latent = comfy.latent_formats.Wan21().process_out(concat_latent)
         concat_latent = concat_latent.repeat(1, 2, 1, 1, 1)
         mask = torch.ones((1, 1, latent.shape[2] * 4, latent.shape[-2], latent.shape[-1]))
 
         if start_image is not None:
             start_image = comfy.utils.common_upscale(start_image[:length].movedim(-1, 1), width, height, "bilinear", "center").movedim(1, -1)
             concat_latent_image = vae.encode(start_image[:, :, :, :3])
-            concat_latent[:,16:,:concat_latent_image.shape[2]] = concat_latent_image[:,:,:concat_latent.shape[2]]
+            concat_latent[:,latent_channels:,:concat_latent_image.shape[2]] = concat_latent_image[:,:,:concat_latent.shape[2]]
             mask[:, :, :start_image.shape[0] + 3] = 0.0
 
         ref_latent = None
@@ -159,11 +164,11 @@ class Wan22FunControlToVideo(io.ComfyNode):
         if control_video is not None:
             control_video = comfy.utils.common_upscale(control_video[:length].movedim(-1, 1), width, height, "bilinear", "center").movedim(1, -1)
             concat_latent_image = vae.encode(control_video[:, :, :, :3])
-            concat_latent[:,:16,:concat_latent_image.shape[2]] = concat_latent_image[:,:,:concat_latent.shape[2]]
+            concat_latent[:,:latent_channels,:concat_latent_image.shape[2]] = concat_latent_image[:,:,:concat_latent.shape[2]]
 
         mask = mask.view(1, mask.shape[2] // 4, 4, mask.shape[3], mask.shape[4]).transpose(1, 2)
-        positive = node_helpers.conditioning_set_values(positive, {"concat_latent_image": concat_latent, "concat_mask": mask, "concat_mask_index": 16})
-        negative = node_helpers.conditioning_set_values(negative, {"concat_latent_image": concat_latent, "concat_mask": mask, "concat_mask_index": 16})
+        positive = node_helpers.conditioning_set_values(positive, {"concat_latent_image": concat_latent, "concat_mask": mask, "concat_mask_index": latent_channels})
+        negative = node_helpers.conditioning_set_values(negative, {"concat_latent_image": concat_latent, "concat_mask": mask, "concat_mask_index": latent_channels})
 
         if ref_latent is not None:
             positive = node_helpers.conditioning_set_values(positive, {"reference_latents": [ref_latent]}, append=True)
@@ -201,7 +206,8 @@ class WanFirstLastFrameToVideo(io.ComfyNode):
 
     @classmethod
     def execute(cls, positive, negative, vae, width, height, length, batch_size, start_image=None, end_image=None, clip_vision_start_image=None, clip_vision_end_image=None) -> io.NodeOutput:
-        latent = torch.zeros([batch_size, 16, ((length - 1) // 4) + 1, height // 8, width // 8], device=comfy.model_management.intermediate_device())
+        spacial_scale = vae.spacial_compression_encode()
+        latent = torch.zeros([batch_size, vae.latent_channels, ((length - 1) // 4) + 1, height // spacial_scale, width // spacial_scale], device=comfy.model_management.intermediate_device())
         if start_image is not None:
             start_image = comfy.utils.common_upscale(start_image[:length].movedim(-1, 1), width, height, "bilinear", "center").movedim(1, -1)
         if end_image is not None:
@@ -877,6 +883,68 @@ def get_audio_embed_bucket_fps(audio_embed, fps=16, batch_frames=81, m=0, video_
     return batch_audio_eb, min_batch_num
 
 
+def wan_sound_to_video(positive, negative, vae, width, height, length, batch_size, frame_offset=0, ref_image=None, audio_encoder_output=None, control_video=None, ref_motion=None, ref_motion_latent=None):
+    latent_t = ((length - 1) // 4) + 1
+    if audio_encoder_output is not None:
+        feat = torch.cat(audio_encoder_output["encoded_audio_all_layers"])
+        video_rate = 30
+        fps = 16
+        feat = linear_interpolation(feat, input_fps=50, output_fps=video_rate)
+        batch_frames = latent_t * 4
+        audio_embed_bucket, num_repeat = get_audio_embed_bucket_fps(feat, fps=fps, batch_frames=batch_frames, m=0, video_rate=video_rate)
+        audio_embed_bucket = audio_embed_bucket.unsqueeze(0)
+        if len(audio_embed_bucket.shape) == 3:
+            audio_embed_bucket = audio_embed_bucket.permute(0, 2, 1)
+        elif len(audio_embed_bucket.shape) == 4:
+            audio_embed_bucket = audio_embed_bucket.permute(0, 2, 3, 1)
+
+        audio_embed_bucket = audio_embed_bucket[:, :, :, frame_offset:frame_offset + batch_frames]
+        if audio_embed_bucket.shape[3] > 0:
+            positive = node_helpers.conditioning_set_values(positive, {"audio_embed": audio_embed_bucket})
+            negative = node_helpers.conditioning_set_values(negative, {"audio_embed": audio_embed_bucket * 0.0})
+            frame_offset += batch_frames
+
+    if ref_image is not None:
+        ref_image = comfy.utils.common_upscale(ref_image[:1].movedim(-1, 1), width, height, "bilinear", "center").movedim(1, -1)
+        ref_latent = vae.encode(ref_image[:, :, :, :3])
+        positive = node_helpers.conditioning_set_values(positive, {"reference_latents": [ref_latent]}, append=True)
+        negative = node_helpers.conditioning_set_values(negative, {"reference_latents": [ref_latent]}, append=True)
+
+    if ref_motion is not None:
+        if ref_motion.shape[0] > 73:
+            ref_motion = ref_motion[-73:]
+
+        ref_motion = comfy.utils.common_upscale(ref_motion.movedim(-1, 1), width, height, "bilinear", "center").movedim(1, -1)
+
+        if ref_motion.shape[0] < 73:
+            r = torch.ones([73, height, width, 3]) * 0.5
+            r[-ref_motion.shape[0]:] = ref_motion
+            ref_motion = r
+
+        ref_motion_latent = vae.encode(ref_motion[:, :, :, :3])
+
+    if ref_motion_latent is not None:
+        ref_motion_latent = ref_motion_latent[:, :, -19:]
+        positive = node_helpers.conditioning_set_values(positive, {"reference_motion": ref_motion_latent})
+        negative = node_helpers.conditioning_set_values(negative, {"reference_motion": ref_motion_latent})
+
+    latent = torch.zeros([batch_size, 16, latent_t, height // 8, width // 8], device=comfy.model_management.intermediate_device())
+
+    control_video_out = comfy.latent_formats.Wan21().process_out(torch.zeros_like(latent))
+    if control_video is not None:
+        control_video = comfy.utils.common_upscale(control_video[:length].movedim(-1, 1), width, height, "bilinear", "center").movedim(1, -1)
+        control_video = vae.encode(control_video[:, :, :, :3])
+        control_video_out[:, :, :control_video.shape[2]] = control_video
+
+    # TODO: check if zero is better than none if none provided
+    positive = node_helpers.conditioning_set_values(positive, {"control_video": control_video_out})
+    negative = node_helpers.conditioning_set_values(negative, {"control_video": control_video_out})
+
+    out_latent = {}
+    out_latent["samples"] = latent
+    return positive, negative, out_latent, frame_offset
+
+
 class WanSoundImageToVideo(io.ComfyNode):
     @classmethod
     def define_schema(cls):
@@ -906,57 +974,44 @@ class WanSoundImageToVideo(io.ComfyNode):
 
     @classmethod
     def execute(cls, positive, negative, vae, width, height, length, batch_size, ref_image=None, audio_encoder_output=None, control_video=None, ref_motion=None) -> io.NodeOutput:
-        latent_t = ((length - 1) // 4) + 1
-        if audio_encoder_output is not None:
-            feat = torch.cat(audio_encoder_output["encoded_audio_all_layers"])
-            video_rate = 30
-            fps = 16
-            feat = linear_interpolation(feat, input_fps=50, output_fps=video_rate)
-            audio_embed_bucket, num_repeat = get_audio_embed_bucket_fps(feat, fps=fps, batch_frames=latent_t * 4, m=0, video_rate=video_rate)
-            audio_embed_bucket = audio_embed_bucket.unsqueeze(0)
-            if len(audio_embed_bucket.shape) == 3:
-                audio_embed_bucket = audio_embed_bucket.permute(0, 2, 1)
-            elif len(audio_embed_bucket.shape) == 4:
-                audio_embed_bucket = audio_embed_bucket.permute(0, 2, 3, 1)
+        positive, negative, out_latent, frame_offset = wan_sound_to_video(positive, negative, vae, width, height, length, batch_size, ref_image=ref_image, audio_encoder_output=audio_encoder_output,
+                                                                          control_video=control_video, ref_motion=ref_motion)
+        return io.NodeOutput(positive, negative, out_latent)
 
-            positive = node_helpers.conditioning_set_values(positive, {"audio_embed": audio_embed_bucket})
-            negative = node_helpers.conditioning_set_values(negative, {"audio_embed": audio_embed_bucket * 0.0})
 
-        if ref_image is not None:
-            ref_image = comfy.utils.common_upscale(ref_image[:1].movedim(-1, 1), width, height, "bilinear", "center").movedim(1, -1)
-            ref_latent = vae.encode(ref_image[:, :, :, :3])
-            positive = node_helpers.conditioning_set_values(positive, {"reference_latents": [ref_latent]}, append=True)
-            negative = node_helpers.conditioning_set_values(negative, {"reference_latents": [ref_latent]}, append=True)
+class WanSoundImageToVideoExtend(io.ComfyNode):
+    @classmethod
+    def define_schema(cls):
+        return io.Schema(
+            node_id="WanSoundImageToVideoExtend",
+            category="conditioning/video_models",
+            inputs=[
+                io.Conditioning.Input("positive"),
+                io.Conditioning.Input("negative"),
+                io.Vae.Input("vae"),
+                io.Int.Input("length", default=77, min=1, max=nodes.MAX_RESOLUTION, step=4),
+                io.Latent.Input("video_latent"),
+                io.AudioEncoderOutput.Input("audio_encoder_output", optional=True),
+                io.Image.Input("ref_image", optional=True),
+                io.Image.Input("control_video", optional=True),
+            ],
+            outputs=[
+                io.Conditioning.Output(display_name="positive"),
+                io.Conditioning.Output(display_name="negative"),
+                io.Latent.Output(display_name="latent"),
+            ],
+            is_experimental=True,
+        )
 
-        if ref_motion is not None:
-            if ref_motion.shape[0] > 73:
-                ref_motion = ref_motion[-73:]
-
-            ref_motion = comfy.utils.common_upscale(ref_motion.movedim(-1, 1), width, height, "bilinear", "center").movedim(1, -1)
-
-            if ref_motion.shape[0] < 73:
-                r = torch.ones([73, height, width, 3]) * 0.5
-                r[-ref_motion.shape[0]:] = ref_motion
-                ref_motion = r
-
-            ref_motion = vae.encode(ref_motion[:, :, :, :3])
-            positive = node_helpers.conditioning_set_values(positive, {"reference_motion": ref_motion})
-            negative = node_helpers.conditioning_set_values(negative, {"reference_motion": ref_motion})
-
-        latent = torch.zeros([batch_size, 16, latent_t, height // 8, width // 8], device=comfy.model_management.intermediate_device())
-
-        control_video_out = comfy.latent_formats.Wan21().process_out(torch.zeros_like(latent))
-        if control_video is not None:
-            control_video = comfy.utils.common_upscale(control_video[:length].movedim(-1, 1), width, height, "bilinear", "center").movedim(1, -1)
-            control_video = vae.encode(control_video[:, :, :, :3])
-            control_video_out[:, :, :control_video.shape[2]] = control_video
-
-        # TODO: check if zero is better than none if none provided
-        positive = node_helpers.conditioning_set_values(positive, {"control_video": control_video_out})
-        negative = node_helpers.conditioning_set_values(negative, {"control_video": control_video_out})
-
-        out_latent = {}
-        out_latent["samples"] = latent
+    @classmethod
+    def execute(cls, positive, negative, vae, length, video_latent, ref_image=None, audio_encoder_output=None, control_video=None) -> io.NodeOutput:
+        video_latent = video_latent["samples"]
+        width = video_latent.shape[-1] * 8
+        height = video_latent.shape[-2] * 8
+        batch_size = video_latent.shape[0]
+        frame_offset = video_latent.shape[-3] * 4
+        positive, negative, out_latent, frame_offset = wan_sound_to_video(positive, negative, vae, width, height, length, batch_size, frame_offset=frame_offset, ref_image=ref_image, audio_encoder_output=audio_encoder_output,
+                                                                          control_video=control_video, ref_motion=None, ref_motion_latent=video_latent)
         return io.NodeOutput(positive, negative, out_latent)
 
 
@@ -1019,6 +1074,7 @@ class WanExtension(ComfyExtension):
             WanCameraImageToVideo,
             WanPhantomSubjectToVideo,
             WanSoundImageToVideo,
+            WanSoundImageToVideoExtend,
             Wan22ImageToVideoLatent,
         ]
 

comfyui_version.py

@@ -1,3 +1,3 @@
 # This file is automatically generated by the build process when version is
 # updated in pyproject.toml.
-__version__ = "0.3.53"
+__version__ = "0.3.56"

main.py

@@ -112,6 +112,7 @@ import gc
 
 
 if os.name == "nt":
+    os.environ['MIMALLOC_PURGE_DELAY'] = '0'
     logging.getLogger("xformers").addFilter(lambda record: 'A matching Triton is not available' not in record.getMessage())
 
 if __name__ == "__main__":

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "ComfyUI"
-version = "0.3.53"
+version = "0.3.56"
 readme = "README.md"
 license = { file = "LICENSE" }
 requires-python = ">=3.9"

requirements.txt

@@ -1,5 +1,5 @@
 comfyui-frontend-package==1.25.11
-comfyui-workflow-templates==0.1.68
+comfyui-workflow-templates==0.1.70
 comfyui-embedded-docs==0.2.6
 torch
 torchsde