วีดีโอ

thanks alex

Thanks for taking the time to go into detail on the node widgets.

Cool! Thank you.

thanks for the info

You can also create a motion vector pass directly from the stmap. Ben McEwan has a great write up on his blog.

Well explained! Thanks for sharing your process in detail.

Hi Alex, Motion Blur node works much better when you connect vector generator to both inputs, one set to foreground and other to background, great videos by the way, all the best, cheers.

unfortunately I was not able to install it.I have experience in installing many cattery gizmo etc. but not this one:(

@alexvillabon how to install ViTMatte in nuke

How to install ViTMatte in nuke 15

plz installation tutorial video 😂😂😂😂😂😂😂😂😂

It looks like a great tool! Unfortunately it is no longer possible to download it from Github anymore. I'm getting this message "There aren’t any releases here" 😢 I hope it comes back!

thank you, great tool!! And I want to ask you Which tool is 10.39 ?

Where it is available??? Send me link

thank you! I tried tested it on M1 mac but I am getting errors right away....is this compatible with mac as well?

I downloaded the files, including the extra ones and put them in the vitmatte folder. When previewing the node, I get a really long error: Exception caught processing model: The following operation failed in the TorchScript interpreter. Traceback of TorchScript, serialized code (most recent call last): File .../__torch__.py, line 13, in forward image_and_trimap = {"image": image, "trimap": trimap} model = self.model _0 = ((model).forward(image_and_trimap, ))["phas"] ~~~~~~~~~~~~~~ <--- HERE return torch.contiguous(_0) File .../vitmatte.py, line 21, in forward images, H, W, = _0 backbone = self.backbone features = (backbone).forward(images, ) ~~~~~~~~~~~~~~~~~ <--- HERE decoder = self.decoder outputs = (decoder).forward(features, images, ) File .../vit.py, line 21, in forward _0 = __torch__.modeling.backbone.utils.get_abs_pos patch_embed = self.patch_embed x0 = (patch_embed).forward(x, ) ~~~~~~~~~~~~~~~~~~~~ <--- HERE pos_embed = self.pos_embed pretrain_use_cls_token = self.pretrain_use_cls_token File .../utils.py, line 10, in forward x: Tensor) -> Tensor: proj = self.proj x0 = (proj).forward(x, ) ~~~~~~~~~~~~~ <--- HERE return torch.permute(x0, [0, 2, 3, 1]) def get_abs_pos(abs_pos: Tensor, File .../conv.py, line 23, in forward weight = self.weight bias = self.bias _0 = (self)._conv_forward(input, weight, bias, ) ~~~~~~~~~~~~~~~~~~~ <--- HERE return _0 def _conv_forward(self: __torch__.torch.nn.modules.conv.Conv2d, File .../conv.py, line 29, in _conv_forward weight: Tensor, bias: Optional[Tensor]) -> Tensor: _1 = torch.conv2d(input, weight, bias, [16, 16], [0, 0], [1, 1]) ~~~~~~~~~~~~ <--- HERE return _1 Traceback of TorchScript, original code (most recent call last): File "nuke_vitmatte.py", line 72, in forward } return self.model(image_and_trimap)["phas"].contiguous() ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ <--- HERE File .../vitmatte.py, line 42, in forward images, H, W = self.preprocess_inputs(batched_inputs) features = self.backbone(images) ~~~~~~~~~~~~~ <--- HERE outputs = self.decoder(features, images) File .../vit.py, line 396, in forward def forward(self, x): x = self.patch_embed(x) ~~~~~~~~~~~~~~~~ <--- HERE if self.pos_embed is not None: x = x + get_abs_pos( File .../utils.py, line 205, in forward def forward(self, x): x = self.proj(x) ~~~~~~~~~ <--- HERE # B C H W -> B H W C x = x.permute(0, 2, 3, 1) File .../conv.py, line 463, in forward def forward(self, input: Tensor) -> Tensor: return self._conv_forward(input, self.weight, self.bias) ~~~~~~~~~~~~~~~~~~ <--- HERE File .../conv.py, line 459, in _conv_forward weight, bias, self.stride, _pair(0), self.dilation, self.groups) return F.conv2d(input, weight, bias, self.stride, ~~~~~~~~ <--- HERE self.padding, self.dilation, self.groups) RuntimeError: Input type (float) and bias type (struct c10::Half) should be the same Haven't seen anybody else with this problem, any idea what's causing it?

Edit 02 - Works ! I forgot to put a roto node before the inference node. Cheers. End Edit 02 Edit 01 - I downloaded the folder from git and put the external cat files in the same folder. Now the node is showing up in Nuke but it's not doing anything. End Edit 01 Hi Alex . The nodes not showing up in nukex 15 . The archive which is linked to a g drive only has a cat and a pt file. Unlike the other folders that have icons and gizmo files. Did you also download the archive from the external link ? Cheers, b

sorry, where i could find and put cat file? I read gethub, but not understand((

Works on CPU but on GPU i got this message. (Nuke14, Windows). RuntimeError: Input type (CUDAFloatType) and weight type (CUDAHalfType) should be the same Investigating what it could be...

I wish there was a way it could output roto shapes so artists can take over and make corrections.

topaz ai plugin fails to load in my nuke what do i do?

cool

this is insanely good. Thanks for sharing

I found the edge thickness to be not very useful tbh. Internally, it's just two FilerErodes - one expanding, one extracting the given matte, and mixing them so the larger one becomes the 0.5 the inference expects; after the inference, it gets masked by the larger one. So far, I always got better results by putting that to 0 and feeding a manually adjusted "double mask" with the "translucency range" set to 0.5

I think the download link is broken, like someone mentioned, it leads to the "There aren’t any releases here" page.

Am I the only one who cant figure out how to download it?) Link in description lead to "There aren’t any releases here"

i am already using to help keying green screens an it works very well..

Even with the boiling, that's not bad at all! I would imagine you could throw it through temporalmedian and train copycat off 20-30 of the best frames n get a pretty solid result

🤯Cool!

Not working for me, i have other cat tools added as well but this one is saying "no cat file is selected" any solution?

Hi thank you for the vid, followed the instruction but when adding the node I get an error 'no cat file is selected' any idea?

Thanks for showing that off again! I had a look at his GitHub and saw ViTMatte, which does have incredible output on hair. I saw Julian Kreussers tutorial on Advanced Removals on Fundrys Learning Channel, but instead of only using ModNet I keymixed it with ViTMatte; ViTMatte for fine detail/translucency like hair/motionblur, and ModNet for a fast solid body). Then trained a CopyCat with that (like you also showed in the ComfyUI-Normalmap Video), and the output was intense. Using his trick "stabilizing" it with the Denoise, I got a very decent matte for roughly an hour of work and 45min of training on a 100f shot. Amazing what we can get these days

Buenísimo. Gracias Alex.

I love these videos

I have been using RIFE through flowframes for a while now. So cool to see it integrated into Nuke. It really is the best frame interpolation tool out there.

looks good

Versioning, it’s so easy now, thanks Alex!

Just now came across your channel. Are you trying to figure out how to use comfy + ue5 + nuke? Not many are trying to innovate like that and i'd love to know if this is the case, because I'm doing the exact some thing.

thank you so much Alex! It helped me a lot. I used Ray render for mattepaint and I had exactly this problem with overscan. I had to go back to the skyline render because I didn’t have time to deal with overscan for the ray render. And now there is a solution.

For something like the runner it would work better if it could work temporally.

That LaMa result is very useable as a first pass. Much better than photoshop generative fill.

Hi do you just drag the demo image into comfyui and let the manager figure out the missing nodes ? Or did you download the model files separately. Thanks

where can I get the pmask node?

How much ram do you have and which GPU are you running?

Something i have been looking for months, thanks a lot for the insight

i see this in beauty retouches <3

Clever

very nice and informative! fantastic sharing my friend!

I hope this gets better. However if we get a clean matte for single frames, then we could then use the output to train copycat. Cheers

Could you please make a cat file "base" and more large model ? sorry for asking, I dont know any code or program :(