[pull] master from comfyanonymous:master

.github/workflows/windows_release_package.yml

@@ -7,19 +7,19 @@ on:
         description: 'cuda version'
         required: true
         type: string
-        default: "128"
+        default: "129"
 
       python_minor:
         description: 'python minor version'
         required: true
         type: string
-        default: "12"
+        default: "13"
 
       python_patch:
         description: 'python patch version'
         required: true
         type: string
-        default: "10"
+        default: "6"
 #  push:
 #    branches:
 #      - master
@@ -64,6 +64,8 @@ jobs:
             ./python.exe get-pip.py
             ./python.exe -s -m pip install ../cu${{ inputs.cu }}_python_deps/*
             sed -i '1i../ComfyUI' ./python3${{ inputs.python_minor }}._pth
+
+            rm ./Lib/site-packages/torch/lib/dnnl.lib #I don't think this is actually used and I need the space
             cd ..
 
             git clone --depth 1 https://github.com/comfyanonymous/taesd

comfy/ldm/wan/model.py

@@ -391,6 +391,7 @@ def __init__(self,
                  cross_attn_norm=True,
                  eps=1e-6,
                  flf_pos_embed_token_number=None,
+                 in_dim_ref_conv=None,
                  image_model=None,
                  device=None,
                  dtype=None,
@@ -484,6 +485,11 @@ def __init__(self,
         else:
             self.img_emb = None
 
+        if in_dim_ref_conv is not None:
+            self.ref_conv = operations.Conv2d(in_dim_ref_conv, dim, kernel_size=patch_size[1:], stride=patch_size[1:], device=operation_settings.get("device"), dtype=operation_settings.get("dtype"))
+        else:
+            self.ref_conv = None
+
     def forward_orig(
         self,
         x,
@@ -526,6 +532,13 @@ def forward_orig(
         e = e.reshape(t.shape[0], -1, e.shape[-1])
         e0 = self.time_projection(e).unflatten(2, (6, self.dim))
 
+        full_ref = None
+        if self.ref_conv is not None:
+            full_ref = kwargs.get("reference_latent", None)
+            if full_ref is not None:
+                full_ref = self.ref_conv(full_ref).flatten(2).transpose(1, 2)
+                x = torch.concat((full_ref, x), dim=1)
+
         # context
         context = self.text_embedding(context)
 
@@ -552,6 +565,9 @@ def block_wrap(args):
         # head
         x = self.head(x, e)
 
+        if full_ref is not None:
+            x = x[:, full_ref.shape[1]:]
+
         # unpatchify
         x = self.unpatchify(x, grid_sizes)
         return x
@@ -570,6 +586,9 @@ def forward(self, x, timestep, context, clip_fea=None, time_dim_concat=None, tra
             x = torch.cat([x, time_dim_concat], dim=2)
             t_len = ((x.shape[2] + (patch_size[0] // 2)) // patch_size[0])
 
+        if self.ref_conv is not None and "reference_latent" in kwargs:
+            t_len += 1
+
         img_ids = torch.zeros((t_len, h_len, w_len, 3), device=x.device, dtype=x.dtype)
         img_ids[:, :, :, 0] = img_ids[:, :, :, 0] + torch.linspace(0, t_len - 1, steps=t_len, device=x.device, dtype=x.dtype).reshape(-1, 1, 1)
         img_ids[:, :, :, 1] = img_ids[:, :, :, 1] + torch.linspace(0, h_len - 1, steps=h_len, device=x.device, dtype=x.dtype).reshape(1, -1, 1)

comfy/model_base.py

@@ -1124,7 +1124,11 @@ def concat_cond(self, **kwargs):
                 mask = mask.repeat(1, 4, 1, 1, 1)
             mask = utils.resize_to_batch_size(mask, noise.shape[0])
 
-        return torch.cat((mask, image), dim=1)
+        concat_mask_index = kwargs.get("concat_mask_index", 0)
+        if concat_mask_index != 0:
+            return torch.cat((image[:, :concat_mask_index], mask, image[:, concat_mask_index:]), dim=1)
+        else:
+            return torch.cat((mask, image), dim=1)
 
     def extra_conds(self, **kwargs):
         out = super().extra_conds(**kwargs)
@@ -1140,6 +1144,10 @@ def extra_conds(self, **kwargs):
         if time_dim_concat is not None:
             out['time_dim_concat'] = comfy.conds.CONDRegular(self.process_latent_in(time_dim_concat))
 
+        reference_latents = kwargs.get("reference_latents", None)
+        if reference_latents is not None:
+            out['reference_latent'] = comfy.conds.CONDRegular(self.process_latent_in(reference_latents[-1])[:, :, 0])
+
         return out
 
 

comfy/model_detection.py

@@ -373,6 +373,11 @@ def detect_unet_config(state_dict, key_prefix, metadata=None):
         flf_weight = state_dict.get('{}img_emb.emb_pos'.format(key_prefix))
         if flf_weight is not None:
             dit_config["flf_pos_embed_token_number"] = flf_weight.shape[1]
+
+        ref_conv_weight = state_dict.get('{}ref_conv.weight'.format(key_prefix))
+        if ref_conv_weight is not None:
+            dit_config["in_dim_ref_conv"] = ref_conv_weight.shape[1]
+
         return dit_config
 
     if '{}latent_in.weight'.format(key_prefix) in state_dict_keys:  # Hunyuan 3D

comfy_api/latest/_ui.py

@@ -12,7 +12,7 @@
 try:
     import torchaudio
     TORCH_AUDIO_AVAILABLE = True
-except ImportError:
+except:
     TORCH_AUDIO_AVAILABLE = False
 from PIL import Image as PILImage
 from PIL.PngImagePlugin import PngInfo

comfy_extras/nodes_wan.py

@@ -103,6 +103,63 @@ def encode(self, positive, negative, vae, width, height, length, batch_size, sta
         out_latent["samples"] = latent
         return (positive, negative, out_latent)
 
+class Wan22FunControlToVideo:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": {"positive": ("CONDITIONING", ),
+                             "negative": ("CONDITIONING", ),
+                             "vae": ("VAE", ),
+                             "width": ("INT", {"default": 832, "min": 16, "max": nodes.MAX_RESOLUTION, "step": 16}),
+                             "height": ("INT", {"default": 480, "min": 16, "max": nodes.MAX_RESOLUTION, "step": 16}),
+                             "length": ("INT", {"default": 81, "min": 1, "max": nodes.MAX_RESOLUTION, "step": 4}),
+                             "batch_size": ("INT", {"default": 1, "min": 1, "max": 4096}),
+                },
+                "optional": {"ref_image": ("IMAGE", ),
+                             "control_video": ("IMAGE", ),
+                             # "start_image": ("IMAGE", ),
+                }}
+
+    RETURN_TYPES = ("CONDITIONING", "CONDITIONING", "LATENT")
+    RETURN_NAMES = ("positive", "negative", "latent")
+    FUNCTION = "encode"
+
+    CATEGORY = "conditioning/video_models"
+
+    def encode(self, positive, negative, vae, width, height, length, batch_size, ref_image=None, start_image=None, control_video=None):
+        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)
+        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]]
+            mask[:, :, :start_image.shape[0] + 3] = 0.0
+
+        ref_latent = None
+        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])
+
+        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]]
+
+        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})
+
+        if ref_latent is not None:
+            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)
+
+        out_latent = {}
+        out_latent["samples"] = latent
+        return (positive, negative, out_latent)
+
 class WanFirstLastFrameToVideo:
     @classmethod
     def INPUT_TYPES(s):
@@ -733,6 +790,7 @@ def encode(self, vae, width, height, length, batch_size, start_image=None):
     "WanTrackToVideo": WanTrackToVideo,
     "WanImageToVideo": WanImageToVideo,
     "WanFunControlToVideo": WanFunControlToVideo,
+    "Wan22FunControlToVideo": Wan22FunControlToVideo,
     "WanFunInpaintToVideo": WanFunInpaintToVideo,
     "WanFirstLastFrameToVideo": WanFirstLastFrameToVideo,
     "WanVaceToVideo": WanVaceToVideo,