WEBVTT
1
00:00:00.080 --> 00:00:08.820
Instead of thinking about nutritional compositing, you lay out, here's your image, here's the, the basic processes you expect, and then you refine as you go. With AI, you need to think about the result.
2
00:00:09.000 --> 00:00:21.160
This is Kristy Anzelmo, Chief Product Officer at Foundry, the company behind Nuke and Grip Tape. In this episode, how AI flips the way VFX artists work, how tools are being rebuilt for agents, and a whole bunch more.
3
00:00:21.340 --> 00:00:31.040
Like, anytime you're thinking about a product or a new feature, now you need to think about the human operator and the agentic operator of that, and how much you want to enable those different things.
4
00:00:31.100 --> 00:00:48.860
To see the team all the time building these crazy interfaces, I totally believe we're getting to this era of personalized software. From VP Land. At AI on the lot. [upbeat music] This is inside the AI studio.
5
00:00:51.640 --> 00:01:01.340
Kristy, tell me, for people not familiar with Foundry, what is Foundry? What do we do? Yes. [laughs] Yeah. So Foundry, we're makers of creative software for visual effects and animation.
6
00:01:01.840 --> 00:01:11.640
The company's been around for 30 years as of this year. Oh, nice. Crazy. [laughs] It's the birthday. We're best known for Nuke, and the Nuke compositing tool set.
7
00:01:11.860 --> 00:01:20.200
And we do have a portfolio of products that reach from pre-production with Flix, which is a storyboarding tool, like really a storyboarding management tool.
8
00:01:20.340 --> 00:01:29.120
Nuke Studio, which is used for in-camera VFX virtual production, and of course Nuke Katana and Mari, compositing, lighting, and look dev. And well, very exciting.
9
00:01:29.180 --> 00:01:38.200
We've had Grip Tape join the Foundry just three months ago now. Tell me about Grip Tape, 'cause we're at an AI conference, so like that's gonna be [laughs] the most- Yeah, definitely. This is a big, a big topic...
10
00:01:38.220 --> 00:01:51.720
piece of curiosity. Yeah. So, uh, Grip Tape. Yeah, so Grip Tape, it's a node-based orchestration platform. But what's unique about it is that it is a, well, truly transparent, scalable, extensible system.
11
00:01:51.840 --> 00:01:54.920
So it's Python- Mm-hmm... um, open source, Apache 2.
12
00:01:55.500 --> 00:02:08.300
I think it's also a bit unique about it is it can be used to make, you know, image generation workflows and do things that are quite accessible to artists, but also orchestrate pipeline processes and really pretty much anything you can drive with Python.
13
00:02:09.020 --> 00:02:29.710
So it's really a kind of broad foundation when you start to think about how you could use AI for media production, you know, not just for creating images, but also how you glue the pipeline together and how you build different kinds of code and automations and, yeah, that's why it stood out to us as being this platform that you could build the, the whole workflow on, that's a combination of tools, not just generative AI.
14
00:02:29.840 --> 00:02:40.590
Well, how are you thinking about Grip Tape, generative AI, and then also kind of more of Nuke and your traditional compositing pipelines? So yeah. So I mean, obviously AI is advancing fast, fast.
15
00:02:40.600 --> 00:02:49.240
I'm sure everybody you talk to, like the use of it, how it's used, the rules around using it, like that's changing daily, weekly, all the time. [laughs] Yeah. [laughs] Which is exciting, right?
16
00:02:49.620 --> 00:02:55.220
And I think what's nice about having a platform like Grip Tape, it's really up to the show, right?
17
00:02:55.320 --> 00:03:08.769
And like, you know, our customers can choose what models they wanna use, when they wanna use them, who has access to what, and manage all of that so they can choose the tools they wanna use for the job, because it varies so much around personal preference- Mm-hmm...
18
00:03:08.940 --> 00:03:17.760
and production guidelines and, and all of those things. So it's, it's really Grip Tape makes that platform. You can work locally, you can talk to the cloud, you can bring that all together.
19
00:03:18.000 --> 00:03:31.560
Well then, you know, with Nuke, obviously a compositing tool set, comp is really, I think a important stage to refine what you get from AI, but also to help, you know, when you're designing and directing what you want to happen.
20
00:03:32.220 --> 00:03:43.460
I think it shows that power of, you know, when you combine like generative AI and DCCs like Nuke, and you go back and forth between them to, you know, design the look, you know, build a camera, map it out in 3D space.
21
00:03:43.640 --> 00:03:53.860
You don't need a full like animation rig set up for everything. But just guiding with 3D with that camera, you can create some amazing control and set up really incredible shots.
22
00:03:54.109 --> 00:04:03.250
And yeah, it's really just the power of blending together. And it, well, you could probably see like Grip Tape coming into Foundry, we'll integrate that into tools. Mm-hmm.
23
00:04:03.300 --> 00:04:16.760
So when you're working in Nuke, compositors can have access to this expanded AI toolkit that you get with gen AI from where they work, but also using it as a way to orchestrate different kinds of workflows between applications.
24
00:04:16.880 --> 00:04:19.839
So like lighting and comp coming together and that kind of thing.
25
00:04:19.940 --> 00:04:31.490
What have you found are kind of some of the gaps in AI, gen AI, even with a tool like Grip Tape for a compositor or someone who's used to Nuke, where like they have- Mm... very detailed control over- Yeah...
26
00:04:31.490 --> 00:04:36.480
every part of the image. Definitely. The control, right? Yeah. [laughs] Controls. I think that's the word, right? Yeah. Is control.
27
00:04:37.040 --> 00:04:47.620
And we found too, like, you know, in Nuke we have a tool set called Copycat, which you can use to train your own models based on data in your shot. It, it uses overfitting to make a really bespoke model.
28
00:04:47.960 --> 00:04:54.320
So you can make like a plugin basically. And that's been out since 2021. We released it. It's crazy. How did it go?
29
00:04:54.730 --> 00:05:02.570
[laughs] Um, [laughs] but anyway, it's used, it's used a lot in production to, especially if you have a task that's repetitive, you can automate it. Mm-hmm.
30
00:05:02.930 --> 00:05:13.360
So we had quite a prominent use case I see a year ago for Dune 2, where they did the Fremen blue eyes. Mm-hmm. Like using this model they trained on data from the first film to do that automatically.
31
00:05:13.460 --> 00:05:18.140
And so like that tool set, even with that kind of workflow for artists, it's a big leap.
32
00:05:18.440 --> 00:05:28.760
Instead of thinking about, you know, traditional compositing, you lay out, here's your image, here's the, the basic processes you expect, and then you refine as you go. With AI, you need to think about the result, right?
33
00:05:28.780 --> 00:05:36.260
It's like, what's the vision that I want? And then what's the information I need to give to guide the model to make it? And I think with generative AI, it's still true, right?
34
00:05:36.500 --> 00:05:46.820
It's in a prompt versus a, a data set sometimes. But it can be a data set, I guess if you're making LoRAs and that kind of thing. But, um, yeah, it, it reverses the thinking a little bit.
35
00:05:47.160 --> 00:05:55.540
And so I think for people that have only done like traditional VFX, it's, it's a different approach, but I think we're seeing people get through that hurdle.
36
00:05:55.740 --> 00:05:59.580
But there's a challenge, like nobody wants to prompt everything. Oh. [laughs] You know?
37
00:05:59.620 --> 00:06:08.710
Well, I mean, do you think it's also a matter of just also as the models keep getting better, they just have more control and they could be integrated, like, um, even Google Omni- Mm...
38
00:06:08.740 --> 00:06:14.760
which had like very good modification features and some testing. So like do you just feel like you just get the pipeline and the tool set?
39
00:06:14.900 --> 00:06:22.350
Or, or, like how are you thinking about building out a tool set where it's flexible enough that- Mm... with the models changing all the time? You can kind of just swap it, swap in and out
40
00:06:23.290 --> 00:06:33.470
the models with your pipeline The models, definitely. I mean, that's, I think, one of the strengths of Grotape. Like, it's really quick to build a new model integration, and on with a coding agent, obviously.
41
00:06:33.530 --> 00:06:55.710
It's, you know, like minutes to get a new model in, and we say all the time, the team just vibe coding nodes, and the nodes are quite smart, so when you have those different parameters for control or how you might drive a prompt, I mean, I think there will be a line where we get to much more, like, interactive guiding of models, where you might need deeper integrations to bring the UI, like creative UI elements, you know, into another package.
42
00:06:56.410 --> 00:07:04.450
But from the way it currently stands, like the parts you can drive in an API, that can just dynamically be exposed. Yeah. You kinda mentioned vibe coding.
43
00:07:04.630 --> 00:07:16.710
I'm curious on like a user end, or also just even in future Nuke build-outs, like how are you thinking of users building out their own solutions, and then also opening or, like, letting agents control, like, building out a pipeline?
44
00:07:16.950 --> 00:07:21.550
Yeah, right. I mean, this is, this is where we're go- going, [laughs] isn't it? Like, now you need to think about...
45
00:07:21.590 --> 00:07:34.630
Obviously, my product management is my background, but it's like anytime you're thinking about a product or a new feature, now you need to think about the human operator and the agentic operator of that, and how much you want to enable those different, different things.
46
00:07:34.810 --> 00:07:41.800
I mean, it's a change, right? I think it's exciting, though, in terms of what you can build. And certainly even in Grotape, you know, you can make custom widgets.
47
00:07:41.950 --> 00:07:52.180
You see the team all the time building these crazy interfaces, and I can... I totally believe we're getting to this era of personalized software. Mm-hmm. Like, we can all develop apps and things just- Mm-hmm...
48
00:07:52.190 --> 00:08:01.700
I'm sure everyone has like some kind of dashboard they built with a [laughs]... an agent for managing something in their life. And, um, we're open to all of that really. You know, it's...
49
00:08:01.730 --> 00:08:10.890
There are great things about the interface in Nuke for managing scale and, you know, how the node graph scales and how you operate in it, which I think is its strength. I think we'll see more things coming into that.
50
00:08:11.350 --> 00:08:19.350
But also, even from Grotape, you can run a Nuke process, right, in a different UI, and that's okay. Mm. Okay. So- So I think it's gonna get to everything, right? [laughs] Okay.
51
00:08:19.550 --> 00:08:28.959
And so you're saying, like, it's also two-way between the two. Yeah, I think, well, it will be, like- For-... soon. The... You know, we'll expose, you know, models that run via Grotape- Mm-hmm...
52
00:08:28.970 --> 00:08:36.530
from Nuke, but also run Nuke processes from Grotape, which in many ways you can already do with Nuke's Python API and the way we manage, like, render processes.
53
00:08:36.640 --> 00:08:45.369
It's just one little kind of bit to make it slightly easier, but it's already possible really. Yeah. Cool. Christie, thanks so much. Appreciate it. Oh, thank you so much, Joey. Yeah. It's great to talk to you. Likewise.
1
00:00:00.080 --> 00:00:08.820
Instead of thinking about nutritional compositing, you lay out, here's your image, here's the, the basic processes you expect, and then you refine as you go. With AI, you need to think about the result.
2
00:00:09.000 --> 00:00:21.160
This is Kristy Anzelmo, Chief Product Officer at Foundry, the company behind Nuke and Grip Tape. In this episode, how AI flips the way VFX artists work, how tools are being rebuilt for agents, and a whole bunch more.
3
00:00:21.340 --> 00:00:31.040
Like, anytime you're thinking about a product or a new feature, now you need to think about the human operator and the agentic operator of that, and how much you want to enable those different things.
4
00:00:31.100 --> 00:00:48.860
To see the team all the time building these crazy interfaces, I totally believe we're getting to this era of personalized software. From VP Land. At AI on the lot. [upbeat music] This is inside the AI studio.
5
00:00:51.640 --> 00:01:01.340
Kristy, tell me, for people not familiar with Foundry, what is Foundry? What do we do? Yes. [laughs] Yeah. So Foundry, we're makers of creative software for visual effects and animation.
6
00:01:01.840 --> 00:01:11.640
The company's been around for 30 years as of this year. Oh, nice. Crazy. [laughs] It's the birthday. We're best known for Nuke, and the Nuke compositing tool set.
7
00:01:11.860 --> 00:01:20.200
And we do have a portfolio of products that reach from pre-production with Flix, which is a storyboarding tool, like really a storyboarding management tool.
8
00:01:20.340 --> 00:01:29.120
Nuke Studio, which is used for in-camera VFX virtual production, and of course Nuke Katana and Mari, compositing, lighting, and look dev. And well, very exciting.
9
00:01:29.180 --> 00:01:38.200
We've had Grip Tape join the Foundry just three months ago now. Tell me about Grip Tape, 'cause we're at an AI conference, so like that's gonna be [laughs] the most- Yeah, definitely. This is a big, a big topic...
10
00:01:38.220 --> 00:01:51.720
piece of curiosity. Yeah. So, uh, Grip Tape. Yeah, so Grip Tape, it's a node-based orchestration platform. But what's unique about it is that it is a, well, truly transparent, scalable, extensible system.
11
00:01:51.840 --> 00:01:54.920
So it's Python- Mm-hmm... um, open source, Apache 2.
12
00:01:55.500 --> 00:02:08.300
I think it's also a bit unique about it is it can be used to make, you know, image generation workflows and do things that are quite accessible to artists, but also orchestrate pipeline processes and really pretty much anything you can drive with Python.
13
00:02:09.020 --> 00:02:29.710
So it's really a kind of broad foundation when you start to think about how you could use AI for media production, you know, not just for creating images, but also how you glue the pipeline together and how you build different kinds of code and automations and, yeah, that's why it stood out to us as being this platform that you could build the, the whole workflow on, that's a combination of tools, not just generative AI.
14
00:02:29.840 --> 00:02:40.590
Well, how are you thinking about Grip Tape, generative AI, and then also kind of more of Nuke and your traditional compositing pipelines? So yeah. So I mean, obviously AI is advancing fast, fast.
15
00:02:40.600 --> 00:02:49.240
I'm sure everybody you talk to, like the use of it, how it's used, the rules around using it, like that's changing daily, weekly, all the time. [laughs] Yeah. [laughs] Which is exciting, right?
16
00:02:49.620 --> 00:02:55.220
And I think what's nice about having a platform like Grip Tape, it's really up to the show, right?
17
00:02:55.320 --> 00:03:08.769
And like, you know, our customers can choose what models they wanna use, when they wanna use them, who has access to what, and manage all of that so they can choose the tools they wanna use for the job, because it varies so much around personal preference- Mm-hmm...
18
00:03:08.940 --> 00:03:17.760
and production guidelines and, and all of those things. So it's, it's really Grip Tape makes that platform. You can work locally, you can talk to the cloud, you can bring that all together.
19
00:03:18.000 --> 00:03:31.560
Well then, you know, with Nuke, obviously a compositing tool set, comp is really, I think a important stage to refine what you get from AI, but also to help, you know, when you're designing and directing what you want to happen.
20
00:03:32.220 --> 00:03:43.460
I think it shows that power of, you know, when you combine like generative AI and DCCs like Nuke, and you go back and forth between them to, you know, design the look, you know, build a camera, map it out in 3D space.
21
00:03:43.640 --> 00:03:53.860
You don't need a full like animation rig set up for everything. But just guiding with 3D with that camera, you can create some amazing control and set up really incredible shots.
22
00:03:54.109 --> 00:04:03.250
And yeah, it's really just the power of blending together. And it, well, you could probably see like Grip Tape coming into Foundry, we'll integrate that into tools. Mm-hmm.
23
00:04:03.300 --> 00:04:16.760
So when you're working in Nuke, compositors can have access to this expanded AI toolkit that you get with gen AI from where they work, but also using it as a way to orchestrate different kinds of workflows between applications.
24
00:04:16.880 --> 00:04:19.839
So like lighting and comp coming together and that kind of thing.
25
00:04:19.940 --> 00:04:31.490
What have you found are kind of some of the gaps in AI, gen AI, even with a tool like Grip Tape for a compositor or someone who's used to Nuke, where like they have- Mm... very detailed control over- Yeah...
26
00:04:31.490 --> 00:04:36.480
every part of the image. Definitely. The control, right? Yeah. [laughs] Controls. I think that's the word, right? Yeah. Is control.
27
00:04:37.040 --> 00:04:47.620
And we found too, like, you know, in Nuke we have a tool set called Copycat, which you can use to train your own models based on data in your shot. It, it uses overfitting to make a really bespoke model.
28
00:04:47.960 --> 00:04:54.320
So you can make like a plugin basically. And that's been out since 2021. We released it. It's crazy. How did it go?
29
00:04:54.730 --> 00:05:02.570
[laughs] Um, [laughs] but anyway, it's used, it's used a lot in production to, especially if you have a task that's repetitive, you can automate it. Mm-hmm.
30
00:05:02.930 --> 00:05:13.360
So we had quite a prominent use case I see a year ago for Dune 2, where they did the Fremen blue eyes. Mm-hmm. Like using this model they trained on data from the first film to do that automatically.
31
00:05:13.460 --> 00:05:18.140
And so like that tool set, even with that kind of workflow for artists, it's a big leap.
32
00:05:18.440 --> 00:05:28.760
Instead of thinking about, you know, traditional compositing, you lay out, here's your image, here's the, the basic processes you expect, and then you refine as you go. With AI, you need to think about the result, right?
33
00:05:28.780 --> 00:05:36.260
It's like, what's the vision that I want? And then what's the information I need to give to guide the model to make it? And I think with generative AI, it's still true, right?
34
00:05:36.500 --> 00:05:46.820
It's in a prompt versus a, a data set sometimes. But it can be a data set, I guess if you're making LoRAs and that kind of thing. But, um, yeah, it, it reverses the thinking a little bit.
35
00:05:47.160 --> 00:05:55.540
And so I think for people that have only done like traditional VFX, it's, it's a different approach, but I think we're seeing people get through that hurdle.
36
00:05:55.740 --> 00:05:59.580
But there's a challenge, like nobody wants to prompt everything. Oh. [laughs] You know?
37
00:05:59.620 --> 00:06:08.710
Well, I mean, do you think it's also a matter of just also as the models keep getting better, they just have more control and they could be integrated, like, um, even Google Omni- Mm...
38
00:06:08.740 --> 00:06:14.760
which had like very good modification features and some testing. So like do you just feel like you just get the pipeline and the tool set?
39
00:06:14.900 --> 00:06:22.350
Or, or, like how are you thinking about building out a tool set where it's flexible enough that- Mm... with the models changing all the time? You can kind of just swap it, swap in and out
40
00:06:23.290 --> 00:06:33.470
the models with your pipeline The models, definitely. I mean, that's, I think, one of the strengths of Grotape. Like, it's really quick to build a new model integration, and on with a coding agent, obviously.
41
00:06:33.530 --> 00:06:55.710
It's, you know, like minutes to get a new model in, and we say all the time, the team just vibe coding nodes, and the nodes are quite smart, so when you have those different parameters for control or how you might drive a prompt, I mean, I think there will be a line where we get to much more, like, interactive guiding of models, where you might need deeper integrations to bring the UI, like creative UI elements, you know, into another package.
42
00:06:56.410 --> 00:07:04.450
But from the way it currently stands, like the parts you can drive in an API, that can just dynamically be exposed. Yeah. You kinda mentioned vibe coding.
43
00:07:04.630 --> 00:07:16.710
I'm curious on like a user end, or also just even in future Nuke build-outs, like how are you thinking of users building out their own solutions, and then also opening or, like, letting agents control, like, building out a pipeline?
44
00:07:16.950 --> 00:07:21.550
Yeah, right. I mean, this is, this is where we're go- going, [laughs] isn't it? Like, now you need to think about...
45
00:07:21.590 --> 00:07:34.630
Obviously, my product management is my background, but it's like anytime you're thinking about a product or a new feature, now you need to think about the human operator and the agentic operator of that, and how much you want to enable those different, different things.
46
00:07:34.810 --> 00:07:41.800
I mean, it's a change, right? I think it's exciting, though, in terms of what you can build. And certainly even in Grotape, you know, you can make custom widgets.
47
00:07:41.950 --> 00:07:52.180
You see the team all the time building these crazy interfaces, and I can... I totally believe we're getting to this era of personalized software. Mm-hmm. Like, we can all develop apps and things just- Mm-hmm...
48
00:07:52.190 --> 00:08:01.700
I'm sure everyone has like some kind of dashboard they built with a [laughs]... an agent for managing something in their life. And, um, we're open to all of that really. You know, it's...
49
00:08:01.730 --> 00:08:10.890
There are great things about the interface in Nuke for managing scale and, you know, how the node graph scales and how you operate in it, which I think is its strength. I think we'll see more things coming into that.
50
00:08:11.350 --> 00:08:19.350
But also, even from Grotape, you can run a Nuke process, right, in a different UI, and that's okay. Mm. Okay. So- So I think it's gonna get to everything, right? [laughs] Okay.
51
00:08:19.550 --> 00:08:28.959
And so you're saying, like, it's also two-way between the two. Yeah, I think, well, it will be, like- For-... soon. The... You know, we'll expose, you know, models that run via Grotape- Mm-hmm...
52
00:08:28.970 --> 00:08:36.530
from Nuke, but also run Nuke processes from Grotape, which in many ways you can already do with Nuke's Python API and the way we manage, like, render processes.
53
00:08:36.640 --> 00:08:45.369
It's just one little kind of bit to make it slightly easier, but it's already possible really. Yeah. Cool. Christie, thanks so much. Appreciate it. Oh, thank you so much, Joey. Yeah. It's great to talk to you. Likewise.
