nodes/stduhpf-ComfyUI--WanImageToVideoTiled
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Vace tiled vae

Browse Source
This commit is contained in:
Stéphane du Hamel
2025-08-03 19:12:02 +02:00
parent 361d261eac
commit 2d0d7da7b4
2 changed files with 92 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
__init__.py
+2
View File
@@ -5,6 +5,7 @@ NODE_CLASS_MAPPINGS = {
"WanFunControlToVideoTiledVAE": WanFunControlToVideoTiledVAE,
"WanFirstLastFrameToVideoTiledVAE": WanFirstLastFrameToVideoTiledVAE,
"WanFunInpaintToVideoTiledVAE": WanFunInpaintToVideoTiledVAE,
"WanVaceToVideoTiledVAE": WanVaceToVideoTiledVAE,
}
NODE_DISPLAY_NAME_MAPPINGS = {
@@ -12,4 +13,5 @@ NODE_DISPLAY_NAME_MAPPINGS = {
"WanFunControlToVideoTiledVAE": "WanFunControlToVideo (Tiled VAE encode)",
"WanFirstLastFrameToVideoTiledVAE": "WanFirstLastFrameToVideo (Tiled VAE encode)",
"WanFunInpaintToVideoTiledVAE": "WanFunInpaintToVideo (Tiled VAE encode)",
"WanVaceToVideoTiledVAE": "WanVaceToVideo (Tiled VAE encode)"
}
nodes_wan.py
+90
View File
@@ -215,3 +215,93 @@ class WanFunInpaintToVideoTiledVAE:
def encode(self, positive, negative, vae, width, height, length, batch_size, tile_size, overlap, temporal_size=64, temporal_overlap=8, start_image=None, end_image=None, clip_vision_output=None):
flfv = WanFirstLastFrameToVideoTiledVAE()
return flfv.encode(positive, negative, vae, width, height, length, batch_size,tile_size, overlap,temporal_size=temporal_size,temporal_overlap=temporal_overlap, start_image=start_image, end_image=end_image, clip_vision_start_image=clip_vision_output)
class WanVaceToVideoTiledVAE:
@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}),
"strength": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 1000.0, "step": 0.01}),
"tile_size": ("INT", {"default": 512, "min": 64, "max": 4096, "step": 64}),
"overlap": ("INT", {"default": 64, "min": 0, "max": 4096, "step": 32}),
"temporal_size": ("INT", {"default": 64, "min": 8, "max": 4096, "step": 4, "tooltip": "Amount of frames to encode at a time."}),
"temporal_overlap": ("INT", {"default": 8, "min": 4, "max": 4096, "step": 4, "tooltip": "Amount of frames to overlap."})
},
"optional": {"control_video": ("IMAGE", ),
"control_masks": ("MASK", ),
"reference_image": ("IMAGE", ),
}}
RETURN_TYPES = ("CONDITIONING", "CONDITIONING", "LATENT", "INT")
RETURN_NAMES = ("positive", "negative", "latent", "trim_latent")
FUNCTION = "encode"
CATEGORY = "conditioning/video_models"
EXPERIMENTAL = True
def encode(self, positive, negative, vae, width, height, length, batch_size, strength, tile_size, overlap, temporal_size=64, temporal_overlap=8, control_video=None, control_masks=None, reference_image=None):
latent_length = ((length - 1) // 4) + 1
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)
if control_video.shape[0] < length:
control_video = torch.nn.functional.pad(control_video, (0, 0, 0, 0, 0, 0, 0, length - control_video.shape[0]), value=0.5)
else:
control_video = torch.ones((length, height, width, 3)) * 0.5
if reference_image is not None:
reference_image = comfy.utils.common_upscale(reference_image[:1].movedim(-1, 1), width, height, "bilinear", "center").movedim(1, -1)
reference_image = vae.encode_tiled(reference_image[:, :, :, :3], tile_x=tile_size, tile_y=tile_size, overlap=overlap, tile_t=temporal_size, overlap_t=temporal_overlap)
reference_image = torch.cat([reference_image, comfy.latent_formats.Wan21().process_out(torch.zeros_like(reference_image))], dim=1)
if control_masks is None:
mask = torch.ones((length, height, width, 1))
else:
mask = control_masks
if mask.ndim == 3:
mask = mask.unsqueeze(1)
mask = comfy.utils.common_upscale(mask[:length], width, height, "bilinear", "center").movedim(1, -1)
if mask.shape[0] < length:
mask = torch.nn.functional.pad(mask, (0, 0, 0, 0, 0, 0, 0, length - mask.shape[0]), value=1.0)
control_video = control_video - 0.5
inactive = (control_video * (1 - mask)) + 0.5
reactive = (control_video * mask) + 0.5
inactive = vae.encode_tiled(inactive[:, :, :, :3], tile_x=tile_size, tile_y=tile_size, overlap=overlap, tile_t=temporal_size, overlap_t=temporal_overlap)
reactive = vae.encode_tiled(reactive[:, :, :, :3], tile_x=tile_size, tile_y=tile_size, overlap=overlap, tile_t=temporal_size, overlap_t=temporal_overlap)
control_video_latent = torch.cat((inactive, reactive), dim=1)
if reference_image is not None:
control_video_latent = torch.cat((reference_image, control_video_latent), dim=2)
vae_stride = 8
height_mask = height // vae_stride
width_mask = width // vae_stride
mask = mask.view(length, height_mask, vae_stride, width_mask, vae_stride)
mask = mask.permute(2, 4, 0, 1, 3)
mask = mask.reshape(vae_stride * vae_stride, length, height_mask, width_mask)
mask = torch.nn.functional.interpolate(mask.unsqueeze(0), size=(latent_length, height_mask, width_mask), mode='nearest-exact').squeeze(0)
trim_latent = 0
if reference_image is not None:
mask_pad = torch.zeros_like(mask[:, :reference_image.shape[2], :, :])
mask = torch.cat((mask_pad, mask), dim=1)
latent_length += reference_image.shape[2]
trim_latent = reference_image.shape[2]
mask = mask.unsqueeze(0)
positive = node_helpers.conditioning_set_values(positive, {"vace_frames": [control_video_latent], "vace_mask": [mask], "vace_strength": [strength]}, append=True)
negative = node_helpers.conditioning_set_values(negative, {"vace_frames": [control_video_latent], "vace_mask": [mask], "vace_strength": [strength]}, append=True)
latent = torch.zeros([batch_size, 16, latent_length, height // 8, width // 8], device=comfy.model_management.intermediate_device())
out_latent = {}
out_latent["samples"] = latent
return (positive, negative, out_latent, trim_latent)