WEBVTT
1
00:00:00.120 --> 00:00:11.780
You can't deny the power that we now have with just a couple years of development. I can only imagine 10 years into this, the way VFX and traditional computer graphics have had two or three decades.
2
00:00:11.880 --> 00:00:28.100
Like, if AI gets a good solid decade in, where will we be? Welcome back to Denoised. Addy, how's it going? What's up, Joey? Long time no see. I know, in the virtual space. Okay, so since our last episode, we had
3
00:00:29.100 --> 00:00:35.940
a little, little, uh, comment, heat in the comments, and [laughs] a little, uh- Things got a little heated, didn't it? A little heated, yeah.
4
00:00:36.080 --> 00:00:50.080
So our last episode was about, well, the thumbnail was about how when I was at TIFF and I talked to people about hybrid AI, as soon as they heard the word AI, they assumed sloppy microdrama kinda stuff. Right.
5
00:00:50.180 --> 00:01:00.380
Uh, or just slop in general. The nuance of the episode was, as I explained that we were talking about shooting with real people and using AI to help production, then it was like, "Oh, okay, that sounds pretty cool."
6
00:01:00.820 --> 00:01:14.680
The comments were the exact [laughs] knee-jerk reactionary comments of, "Yeah, AI is slop." N- well, yeah, there is, there is, I think, a percentage of folks that'll just say e- you know, AI is slop.
7
00:01:14.720 --> 00:01:27.940
But then what I was shocked by was a lot of the commenters were M&A people, like, you know, VFX this, uh, studio that, like some of their screen names. And the comment is like, "Well, AI will never blank.
8
00:01:28.320 --> 00:01:40.040
AI will never..." Like, how... I don't understand how you could be so sure that it'll never do certain things, because [laughs] we're just, like, year two or year three into this thing.
9
00:01:40.880 --> 00:01:47.380
Yeah, there literally are things where it's like, yeah, I don't think it'll be that good at that. And then a week later... Even the past few weeks, the amount of progress- Yes...
10
00:01:48.060 --> 00:01:57.500
if, the things it can do with Blender and computer control has been, I would not have expected that or expected that- Exactly... this year Like, it surprises us, and we're skeptics, right?
11
00:01:57.600 --> 00:02:07.180
And if you follow the podcast journey from, like, a year ago to now, we have had [laughs] our doubts on certain things, and now we're, we're s- surprised ourselves, right? Like,
12
00:02:08.340 --> 00:02:26.260
but then again, I think the mindset is, is that we are on this journey into, um, watching this technology sort of unfold and empower us in making better creative choices, you know, unlocking financial opportunities that may not have existed with traditional VFX.
13
00:02:27.100 --> 00:02:37.240
And we're just in for the ride, right? Like, and we're covering it in our own lens. We're not championing for slop. We never have, [laughs] right? Like, we hate slop as much as...
14
00:02:37.300 --> 00:02:44.540
I mean, I'll laugh at it as a meme, but I, it's not consumable content to me. Yeah, it's not the, what I want my main entertainment to be. Yeah.
15
00:02:44.640 --> 00:02:50.120
And we're constantly, um, sort of, uh, keeping an open mind as this thing progresses.
16
00:02:50.220 --> 00:02:58.460
Of course, there are some things that happen in the AI space that you and I don't agree with, and we talk about it, especially on the ethics side, right?
17
00:02:58.480 --> 00:03:06.480
[laughs] Uh, like when it comes to, like, a Tilly Norwood and, and sort of the drama around that, like we're, that's distasteful, I think, in my opinion.
18
00:03:06.720 --> 00:03:14.200
But having said that, you can't deny the power that we now have with just, just a couple years of development.
19
00:03:14.440 --> 00:03:27.380
And I can only imagine 10 years into this, the way VFX and traditional computer graphics have had two or three decades. Like, if AI gets a good solid decade in, where will we be?
20
00:03:27.780 --> 00:03:38.500
Yeah, and the speed everything's going and where we're at just now, three to five years into when, like, we had crazy Will Smith eating spaghetti, and where we're at now. Yeah.
21
00:03:39.120 --> 00:03:48.660
Uh, and yeah, and it's always like, this is the worst it will ever be. Yeah. So again- Yeah... um, uh, uh, look, shout out to some of our regular listeners who sort of just commented on- There were-...
22
00:03:48.700 --> 00:03:54.420
on regular things Yeah, and supportive. Yeah. Uh, but- Or no, there were a lot of support comments in there- Yeah... from regular listeners, so we appreciate that. Yeah.
23
00:03:54.440 --> 00:04:01.520
But yeah, I found it amusing because it was like, oh, all the comments literally validated the exact premise of the episode, which was [laughs] people hear AI- Yeah...
24
00:04:01.580 --> 00:04:10.880
and just have a knee-jerk reaction And they immediately have a knee-jerk reaction. Right. You're only doing yourself a disservice if you're not curious about it, if you're not willing to test it out.
25
00:04:10.960 --> 00:04:20.560
I, I think if you just sort of face away from the entire thing and just pretend that this is gonna go away because it's wrong or whatnot, I don't think it is.
26
00:04:21.040 --> 00:04:32.060
I don't think it's gonna go away, regardless of how it was built. The question is, like, can we build, build it better, build it more ethically now that it's here and it's kinda proven itself? Yeah.
27
00:04:32.120 --> 00:04:41.000
And I, actually, we'll get into a lot of, I, this, I think, more when we talk about, uh, Katzenberg's essay, uh, in a minute. There was one comment, though, that I did find funny.
28
00:04:41.260 --> 00:04:50.060
"What does any of this have to do with creating art? Quote, 'Data, spreadsheets, outputs.' Do you even hear yourselves?" I think that's about Jeff. Oh. But [laughs] that was really- Okay, yeah.
29
00:04:50.200 --> 00:04:52.160
I, I think- Because [laughs]
30
00:04:52.220 --> 00:04:59.580
I thought it was funny though because I was thinking like, let me tell you, there's not a single film on Earth that exists that a spreadsheet was not involved- And I will correct that person-... to make it possible...
31
00:04:59.600 --> 00:05:08.420
because early animated movies were made with spreadsheets. Yes. Like animation tables are- Yeah. Yes, absolutely... r- they're still spreadsheets under the hood. So spreadsheets- Yeah... can make art, dude.
32
00:05:08.520 --> 00:05:22.700
[laughs] Or dedead. Deater, deader deadler. Yeah. Okay. Katzenberg. Yeah, my, my old, uh, boss's boss's boss's boss. [laughs] You work directly for Jeffrey Katzenberg? I wish.
33
00:05:22.780 --> 00:05:28.280
Uh, actually, um, yeah, I mean, I, I, I have a lot of personal anecdotes of Jeffrey Katzenberg.
34
00:05:28.340 --> 00:05:36.240
I worked at DreamWorks Animation during sort of the golden years in the early 2000s, uh, when Jeffrey was very involved in running the company. He,
35
00:05:37.220 --> 00:05:47.820
you would drive in, you would see his car already on the driveway, you know, as you're driving in, and then as you're driving out of DreamWorks going home, his car is still there. That guy was a workaholic.
36
00:05:47.860 --> 00:05:54.320
He was there day in and day out. Even on some weekends, you would see him. Uh, he had a team of assistants, and, um,
37
00:05:55.780 --> 00:06:03.300
I was in my 20s then, and I, um, I had a couple of, like, meetings and dinners with him here and there, and, um-
38
00:06:04.624 --> 00:06:20.504
As much as people like to bag on his sort of set of choices and, like, I guess you could say, quote-unquote, "ruthless style of leadership," I found his work ethic and sort of his sense of focus really inspiring.
39
00:06:20.624 --> 00:06:25.864
So he's actually one of my heroes [chuckles] from my early days of my career. Yeah. That's awesome.
40
00:06:25.944 --> 00:06:37.824
I mean, yeah, he did a lot of car- good stuff like, uh, so I mean, Jeffrey Katzenberg, uh, part- initially in Disney during their '90s revivement of, like, the golden age of Disney. He brought Disney back to life. Yeah.
41
00:06:37.924 --> 00:06:53.904
Like, Disney was going through crazy malaise, uh, during the '80s, and then in the '90s, Katzenberg was behind The Lion King, Aladdin, you know, um, all of, like, the hits that you associate the classic Disney era with, uh, with the 2D animated look.
42
00:06:54.264 --> 00:07:00.324
Mm-hmm. That's primarily- Lion King, Aladdin... yeah, his brainchild. Side thing. Did you listen to the Acquired podcast? No. Was he on there?
43
00:07:01.224 --> 00:07:12.504
No, but they did a two-part Disney history, Disney episodes that were, uh, excellent. Um, they- I am a big fan of the Acquired podcast, [chuckles] so I have to go back- Yeah... and catch that. Yeah.
44
00:07:12.604 --> 00:07:20.844
Uh, they're the most recent ones. They were, yeah, in total, it's like eight hours, nine hours, and I s- [chuckles] I still wanted more. [chuckles] Tell me more.
45
00:07:20.844 --> 00:07:29.004
[chuckles] But yeah, part two is on the modern age, the post, uh, Dis- Walt Disney era, and that's where it gets into a few more about Katzenberg. Yeah. That's like Michael Eisner, Bob Iger.
46
00:07:29.024 --> 00:07:30.864
Eisner, Katzenberg, and, um, Frank Wells.
47
00:07:31.124 --> 00:07:42.164
Fascinating sort of view of leadership and how the, how a massive organization like that evolved and kept up with its times rather than sort of just dither away and die, kinda like an IBM.
48
00:07:42.184 --> 00:07:54.344
It was on the verge of dying multiple times. [chuckles] Multiple times. Yeah. Okay. Katzenberg, Disney, then left Disney, started DreamWorks, then left, then started Quibi. [chuckles] RIP Quibi.
49
00:07:54.384 --> 00:08:04.984
Uh, he actually started a couple of other startups before, uh, Quibi. Um, I had the pleasure of working on a platform called Pitch, P-T-C-H.
50
00:08:05.704 --> 00:08:15.104
This was in 20- Back when it was cool to, uh, omit, omit the, uh, vowels from the name. Yeah. [chuckles] Yeah. This, this, this was like, um- It's like the-... pre-TikTok, pre-Vine. Yeah.
51
00:08:15.184 --> 00:08:29.064
Pitch was the first image and video social media platform, and it was designed really well, and, uh, Jeffrey was so- just so... He was, like, 10 years ahead of his time. This was like 2011, 2012, something like that.
52
00:08:29.724 --> 00:08:42.844
And as you know, TikTok and everything really hit big, uh, in the, in the late 2010s, right? And so I think the limitation was really on the network and the phone itself rather than the, the platform. That makes sense.
53
00:08:42.844 --> 00:08:54.624
Because our phones weren't able to really download video and play back video well, and then, you know, we're around 3G or 4G then, which was not conducive. Yeah. The early iPhones couldn't shoot video. Yeah.
54
00:08:54.664 --> 00:09:07.624
I mean, this, this pl- It was- This- So, uh, Pitch actually had, um, iPhone or mobile editing on its platform then. Like, you could shoot a video and edit it in your phone, which was pretty novel at the time.
55
00:09:08.404 --> 00:09:18.944
That's cool. Yeah. Other startups, K- uh, Quibi. Then I didn't realize he's been in the Bay Area for, like, 10 years. Um, and so he ha- he's had a venture fund, and he's been investing in various startups.
56
00:09:19.264 --> 00:09:24.724
And then this week, he posted a pretty long essay on X called The World Is Changing: AI for Creativity.
57
00:09:25.144 --> 00:09:35.704
The essay didn't really announce anything, but it was sort of, like, his observations and where he's seen AI and how it's being used in creativity. Now, three years ago, he did say something to the effect of
58
00:09:36.724 --> 00:09:49.244
that he predicted three years ago, in three years, AI would either replace or be able to do 90% of what, uh, it would take to do a high, uh, high-quality animated film.
59
00:09:49.884 --> 00:10:01.124
We're not quite there yet, but I think we're pretty close to that. Just so our haters are aware, [chuckles] we just said we're not quite there yet. We're getting close. Look, well, A, I think we talked about this.
60
00:10:01.184 --> 00:10:11.924
Animation's the e- Animation's the easiest one that could be- It's a easier target, for sure... assisted. Yeah. Assisted first. Yeah. Yeah. Because you don't have the uncanny valley issue. Uh, it works well with styles.
61
00:10:11.964 --> 00:10:20.984
We know animation works pretty well- Yeah... with AI assist. Animation is almost there. So to quote, "I asked one of the leading AI models creators a question that had been challenging me for months.
62
00:10:21.064 --> 00:10:27.004
What is the difference between reasoning and creating? Its answer changed how I think about almost everything happening in the industry."
63
00:10:27.224 --> 00:10:35.364
So he's looking at it like right now, AI models are really good at reasoning, solving things, predicting, not good at creativity, not good at creating. Yeah.
64
00:10:35.624 --> 00:10:44.044
We talk about creating is the act of creating something net new from your old experiences, and this is what we're so good at, right? Mm-hmm.
65
00:10:44.084 --> 00:10:53.464
Like, if you think about Tarantino or whoever, I just saw Once Upon a Time in Hollywood [chuckles] very recently, so, like, I'm thinking, uh, I'm in my Tarantino head.
66
00:10:53.584 --> 00:11:02.604
Um, like, he comes up with those ideas from all of the thousands of movies that he's watched. He's, like, a massive cinema fanatic, right?
67
00:11:02.664 --> 00:11:14.564
Like, so that database is in his head, and he is pulling threads from this movie, that spaghetti Western, and that, you know, w- sci-fi or whatnot. But then the thing that he's creating is net new,
68
00:11:15.464 --> 00:11:30.004
and this is what AI models just can't do yet, is just effectively create net new things. But I don't think they have to, because that's why we're here. You know, we can essentially have AI to help us with that.
69
00:11:30.504 --> 00:11:39.964
Yeah, and that's where we get it. The, the, the human, the person is still the creative driver using these tools. Quote, "Creating is fundamentally generative. It produces something that didn't exist before.
70
00:11:40.304 --> 00:11:52.264
And crucially, there's no single right answer waiting to be found." Exactly. Like, when you are, uh, building a movie, you take left turns and right turns, and you turn the script into something it, it wasn't originally.
71
00:11:52.764 --> 00:12:03.644
Right. And that's part of the creative process. I don't think anybody is really expecting AI to do that. I mean, maybe, maybe on the social media side. I mean, I think in the, yeah, in a social media level- Yeah...
72
00:12:03.664 --> 00:12:14.664
like, there are a lot of tools and pushing out of, like, you know, the, your AI ag- director agent, or, like, here, give it a prompt, give it a script, and it'll, like, do all of the whole creative brief and generate a 30-second thing for you.
73
00:12:15.504 --> 00:12:18.184
I think that, like, people are pu- companies are pushing that
74
00:12:19.252 --> 00:12:29.472
Maybe that'll be some social niche genre or something, but yeah, I don't think that's gonna be at the highest level just- Well, that's not what we're gonna talk about on this podcast 'cause we're not interested in that, right?
75
00:12:29.472 --> 00:12:38.852
[laughs] If you're giving nuance to the haters who are never gonna get this deep into actually listening to the podcast, I'm assuming. Yeah, the, the, we've already lost them in the first 10 seconds.
76
00:12:38.852 --> 00:12:48.772
[laughs] So the gist of the essay is a, well, it's sort of the tale of two cities. So he's up in the Bay Area, they're, they're building the tech, they're building these powerful models, they're engineers.
77
00:12:49.352 --> 00:12:58.932
LA has the creative, creative talent and has all of the, the creative people, and he's, like, looking for a way to, like, sort of bridge these two worlds together. Mm-hmm.
78
00:12:59.012 --> 00:13:10.392
And also for less of the trying to push and resist the AI in general, and more, like, mess around, figure it out, so you could also define the rules and how it comes into play- Mm-hmm...
79
00:13:10.912 --> 00:13:23.632
on your terms and, like, be able to define how this works instead of letting it be defined- I mean-... for you or for the industry by tech and engineers up north. He's, he's a little... I love Katzenberg. Sorry, dude.
80
00:13:24.552 --> 00:13:30.932
You're a little late to the party, JK, 'cause, uh, we have hybrid setups like that, you know.
81
00:13:31.092 --> 00:13:44.692
And I, I'm looking at, you know, organizations like, uh, Netflix, and Fox Next, and Disney, and, like, these teams have well integrated with, uh, Bay Area tech, and they have the studio perspective on things.
82
00:13:45.292 --> 00:13:53.972
And then you also have your Asteria, your Promise, your Secret Level, and, like, all these AI, Wander and AI startups who are combining those two things pretty well.
83
00:13:54.032 --> 00:14:05.292
So I think the tale of two cities has already been mashed up, um, and maybe this is an opportunity for Jeffrey to, like, bring new perspective here. Yeah, I think it's early days of that mashup, and also, like...
84
00:14:05.312 --> 00:14:09.672
Well, look, I mean, on a union at, uh, end, there's still, like, a lot that's unclear and unsorted out.
85
00:14:10.192 --> 00:14:21.132
Yes, we know, like, studios and everything are messing with this, but a lot of it's still, like, a hush-hush or, like, in secret, and there's no, like, established guideline. Systemically, it hasn't integrated.
86
00:14:21.192 --> 00:14:28.352
Like, yeah, you're right. Like, at a union- Mm-hmm... level or at, like, a mass, mass scale level, um, AI is still not a thing.
87
00:14:28.552 --> 00:14:38.152
And it's not gonna be a thing for a long time because we know that the quality is not quite there yet. You're just exchanging one set of problems for another, and that's not really- Yeah... gonna help anybody, right?
88
00:14:38.212 --> 00:14:46.892
Like, if you're gonna adopt AI, it has to be inherently easier, better, faster, cheaper, and all that stuff, and we're just not there yet. This is a good, uh, anecdote.
89
00:14:46.992 --> 00:14:55.292
"Edison invented the motion picture but made terrible movies. It took Chaplin, Lloyd, Keaton, and so many others to make movies emotional." There you go. There you go. I think that was a good point. Yeah. Okay.
90
00:14:55.552 --> 00:15:06.612
Now [laughs] He looks the same. [laughs] He's looked the same for 30 years. That's incredible. He's, he's in good shape, man. He posted this essay on X, um, but it didn't really announce anything.
91
00:15:06.832 --> 00:15:12.312
But then he was on the Town, uh, podcast, and- Nice... they asked more questions.
92
00:15:12.332 --> 00:15:23.632
[laughs] So I wanna jump into a few quotes because there are a few things here too that stand out to me as, like, I don't know if that's the way it'll go. 5 billion people a day watch video.
93
00:15:24.192 --> 00:15:34.252
It is the single greatest way of information and communication and knowledge and entertainment. Like, it touches everything.
94
00:15:35.132 --> 00:15:47.932
What if you actually could now develop the tools that all of that can get elevated, all of it can move to a higher quality and a whole set of new things that none of us actually even are imagining today?
95
00:15:48.232 --> 00:15:58.112
That's what got me excited, and that's what I'm trying to figure out. Hmm. Okay. So the thing that stuck out to me there, 'cause he was talking about, like, yeah, people, 5 billion people watch video a day,
96
00:15:59.132 --> 00:16:05.492
but he brought up quality again and getting to, like, the higher quality level. Yeah. And I think that was Quibi's issue. Part, well, part of many of the issues.
97
00:16:05.492 --> 00:16:09.552
It, it was too high, it was too focused on quality when it should've been about the- It was too focused on quality...
98
00:16:09.672 --> 00:16:15.232
about the d- like, the moment and the distribution and the virality and all the thing that TikTok gets right.
99
00:16:15.252 --> 00:16:23.392
Well, look, and the thing he was right about Quibi was, right, he was right that, yeah, vertical videos would be a thing, and people would be into watching content on their phone.
100
00:16:24.492 --> 00:16:31.692
They just, the quality bar does not need to be that high, nor... I mean, this still goes back to, to, um,
101
00:16:32.672 --> 00:16:45.212
uh, Reed Hastings' quote of just like, yeah, the biggest, the bigger risk is just that people's quality bar of what's acceptable just gets lower, and they don't care about premium quality as much. Damn.
102
00:16:45.632 --> 00:16:51.712
Reed Hastings, smart man. Exactly. [laughs] You, we're guilty of this too 'cause you and I are cinema files, right?
103
00:16:51.872 --> 00:17:05.312
Like, we have our eye on quality all the time, almost too much to the point where we miss the opportunity or we don't see how something could be so big when it is, right? And I think Jeffrey the same way.
104
00:17:05.532 --> 00:17:16.672
He epitomizes the studio, right? Like, he's built studios, and so of course he's gonna have an eye on quality. He's still looking to build the DreamWorks pipeline, but for, like, everyone.
105
00:17:17.292 --> 00:17:29.572
But I don't know if that's- I don't think you need the quality for all 5 billion people. I think you do need the- Right... quality for, like, maybe 1% of that. Tongue, tongue to her. [laughs] Duh.
106
00:17:29.612 --> 00:17:38.932
Do you want me to sing it? Shut up. [laughs] Yeah, do it. No. I need, I need to program one of these buttons here to, uh, to have you here- Tongue tongue to tongue to her. All right, I did it.
107
00:17:38.932 --> 00:17:49.772
[laughs] Um- Yeah, it, it's, it's the perfect example. Oh, by the way, the, the coffee cup character, Ballerina Cappuccina, that's who it was. [laughs] Yeah. That's who's great. Yeah.
108
00:17:49.832 --> 00:18:01.212
It was different, and they had tools that were different, and there was- And you believe that the a- the Pixar of AI is one of these companies that is circulating around the Bay Area right now?
109
00:18:01.752 --> 00:18:10.952
No, I think I'm gonna have to start it. Oh, you do? Yes. I was gonna ask, what is your angle here? [laughs] I think I'm gonna have to start it. So that was the most of, like, what he's looking at, sort of.
110
00:18:10.972 --> 00:18:19.412
There's one other sound bite I wanna play of what he's looking at to build. But yeah, so he's like, like, so yeah, we've talked about what is gonna be the Pixar of AI. He thinks he's gonna have to start that company.
111
00:18:19.672 --> 00:18:31.792
That, but that still seems completely separate from, "I wanna build tools-" That make five billion people create better stuff, and I wanna create the Pixar of AI. Yeah, I, I don't, I don't think we...
112
00:18:31.992 --> 00:18:44.112
We don't, we don't really need it right now. I, I think what we need is, um, more the mashing of tech and Hollywood and figuring out what the new studio environment looks like.
113
00:18:44.472 --> 00:18:51.572
One more soundbite and then I wanna have this discussion about, yeah, what does that look like? Yeah. Mine is actually more I go back to my partner saying- Yeah... AI for creativity.
114
00:18:51.992 --> 00:19:01.312
I'd rather build the picks and shovels and provide them to everybody. Yeah. Like, it's not about having them be proprietary only to, to, for, for our use- Right...
115
00:19:01.332 --> 00:19:15.412
but actually the biggest opportunity would be if you could actually create a set of applications and tools and workflows that could be- That everybody has to pay you. Well, everybody's working on that, JK.
116
00:19:16.312 --> 00:19:22.292
Is he trying to build a model? Yeah. That's, that's the thing. [laughs] It's like, okay, if you say- Yeah... if he's doing the picks and shovels analogy- Yeah...
117
00:19:22.332 --> 00:19:30.532
and he's like, "Oh, we wanna build a tool that lets people create stuff." Well, Google's working on that. Runway's working on that. [laughs] Adobe's working on that. Right, and then you've got, then you've got- Yeah...
118
00:19:30.572 --> 00:19:39.332
Magnific, Arts- Artlist, Hixfield. Yeah. Um, all of these. Everybody's building picks and shovels, so how is he gonna build something different here? Right. And I don't even know if that's the play.
119
00:19:39.372 --> 00:19:48.152
I mean, well, so that, yeah, that's where I wanted to go. So I feel like if you're trying to build the Pixar of AI or you're trying to build a tool for everyone, those are completely separate things.
120
00:19:48.312 --> 00:19:57.452
And he was talking about like, oh, I don't want a gatekeeper, like, you know, have proprietary tools. Pixar has a lot of proprietary tools that, like, helps them make their scenes. RenderMan. Well, USD's an open source.
121
00:19:57.552 --> 00:20:02.772
Uh, Presto. Presto is their animation- Their- Yeah... animation system. Mm-hmm. Proprietary. Yeah.
122
00:20:02.832 --> 00:20:15.212
Yeah, I feel like, okay, there's either gonna be the tool set that is web focused, or you're gonna be a studio and you need to have proprietary systems or something that do- gives you an edge to do what you would do.
123
00:20:15.712 --> 00:20:24.752
Unless he's building a model or something that helps. This was a company that he invested in. Yeah, this was a while back, right? Yeah, a long time ago. May. Real-time AI video startup Reactor.
124
00:20:24.912 --> 00:20:28.692
Katzenberg's company, WonderCo, was an investor in it. I was trying to see, okay, what has he invested in?
125
00:20:28.712 --> 00:20:34.931
He invested in a bunch of just, like, Canva and the usual suspects, but this was the most AI-focused thing I could find that was public that he invested in.
126
00:20:35.392 --> 00:20:40.952
Obviously, real-time AI video would be a, an original model and would be useful. It would be a disruptor for sure.
127
00:20:41.252 --> 00:20:51.832
I think, like, Jeffrey's, uh, strength, in my opinion, just having worked under him [laughs] for so long, is that he has the leadership to pull off Shrek, right?
128
00:20:52.392 --> 00:21:00.552
Shrek wasn't necessarily about computer graphics at the time or all of the, uh, wonderful innovations that were happening with CG.
129
00:21:01.412 --> 00:21:10.832
Shrek was really about disrupting the Disney fairy tale format in a way that became a parody of all of the things that he made previously.
130
00:21:11.152 --> 00:21:22.352
And it, it was, uh, built for any audience, any part of the world, and it was massive, right? It was the highest grossing... Shrek 2 was the highest grossing animated movie for, like, decades after he made it.
131
00:21:23.152 --> 00:21:24.472
That's, I think that's his strength.
132
00:21:24.532 --> 00:21:39.492
So it's not necessarily building tech and tools, which quite frankly he's failed at several times, but building incredible movies and a studio system that can generate those movies and build those movies over and over again consistently, like a DreamWorks.
133
00:21:39.752 --> 00:21:48.312
Or revitalizing Disney Animation. Yeah. You can make the argument that no studio is sort of complete without its technology stack, and that's true.
134
00:21:48.352 --> 00:21:57.732
Like, uh, in the same way military needs a technology stack to be effective, a studio needs tech to be effective. But it's not the main point of a studio.
135
00:21:57.772 --> 00:22:05.992
The main point of a studio is to build IP, leverage IP to make money, and entertain the masses along the way. I think like- Yeah...
136
00:22:06.172 --> 00:22:21.392
i- if he applied himself in that way where, uh, maybe not so much focused on picks and shovels, but really focused on figuring out a feature film with AI that is effective, and he certainly knows the creatives that could pull it off.
137
00:22:21.532 --> 00:22:30.532
He just has to shepherd them in a way that he's done before. Yeah, and to be able to make something that is Shrek quality, but not $200 million. Yeah.
138
00:22:30.572 --> 00:22:45.352
And I, I'll tell you, I'll tell you this one personal, uh, Jeffrey story, uh, that y- I think will highlight how his brain works a little bit, which could be fundamentally, um, challenging in this environment that is moving very fast and very rapidly.
139
00:22:45.511 --> 00:22:59.132
So there was a big town hall meeting at DreamWorks a long time ago, and, uh, Jeffrey had just, uh, moved out of his Bel Air home and then had his new custom-built, uh, Bel Air home in an, like, a higher part of Bel Air, a more baller home, if you will.
140
00:22:59.372 --> 00:23:10.392
And, uh, we, we sort of, uh, you know, we're like, "Yeah, go Jeffrey, go," you know? "Be a billionaire." [laughs] And he's like, "I moved into this new house." I'm paraphrasing here. This was a long time ago.
141
00:23:10.872 --> 00:23:25.072
He's like, "I moved into this new house and I have this incredible view of LA from my infinity pool or whatnot, and I'm just looking out into this view that I've never had before. And then, uh, there is a, uh,
142
00:23:26.152 --> 00:23:36.992
a concrete wall where my dog had pissed, and the piss stain is like into the concrete wall." And he said he couldn't enjoy the view 'cause he kept looking at that concrete piss stain.
143
00:23:37.012 --> 00:23:45.832
[laughs] So, like, that's how his mind works. He's a perfectionist. He's gonna try to do things to the highest degree and to the best of his ability.
144
00:23:45.972 --> 00:23:58.272
But in a, in a environment like what we have with generative AI today, that is detrimental. You can't fixate on little things. You just have to move past it. Um, but I, I think that'll hinder him in some capacity.
145
00:23:58.492 --> 00:24:11.212
Unless he builds the studio and that is where it's like, okay, we are all together as a team to build the film at the highest quality using the best tools available, which includes AI, versus I'm gonna build a platform tool where, like, anyone can make stuff that looks good.
146
00:24:11.812 --> 00:24:24.472
Yes. Agree. [laughs] But hey, uh, I, I'm just happy that he's championing for AI. I mean, that, that is a very influential person in Hollywood. Like, like it or not, he's pulling a lot of strings behind the scenes.
147
00:24:24.492 --> 00:24:32.592
He, you know, he seems to know the heads of every studio, all of the actors, and all the guilds. Like, this is a very powerful figure in this industry.
148
00:24:33.504 --> 00:24:45.664
Um, so if he says accept and embrace AI, then that, that does mean a lot to a lot of people. Yeah. Yeah. I mean, it, yeah, it helps with the changing of the acceptance of the tools. Changing of the guards.
149
00:24:46.084 --> 00:24:49.864
Change [laughs] I don't know what to say change of the guards, but change of the tools. Changing of the guards.
150
00:24:50.044 --> 00:24:56.764
What I really mean by that is, like, we're gonna effectively- The change of the tools, you adapt or you don't adapt. I mean, you gave examples in this essay too you could read of, of like- Yeah...
151
00:24:57.004 --> 00:25:06.304
when they shifted from, uh, hand-drawn animation to, to 3D. That's changing of the guards. A lot of people adapted and some people just didn't adapt and retired. Yeah.
152
00:25:06.544 --> 00:25:19.844
Um, I was, I was a little too young to see this transition, but when DreamWorks f-first started with Prince of Egypt and a couple other movies that was still, like, hand cel-shaded, uh, hand-drawn, like, Disney movies,
153
00:25:20.764 --> 00:25:25.704
and then, uh, both DreamWorks and Disney and Pixar were going through the transition into 3D.
154
00:25:26.064 --> 00:25:38.924
And so a lot of the 2D artists had really nice, uh, fine art skills that were just not applicable 'cause they had to pick up a new software, mouse, keyboard, GPU, expensive computers, and, uh, a lot of them just,
155
00:25:39.904 --> 00:25:47.964
just didn't want it. They're like, "3D is, is not animation. 3D is not it. We're, we're, we're doing things our way, and this is animation. This is the way."
156
00:25:48.484 --> 00:26:00.124
And lo and behold, like, 25, 26 years later, right, what is animation today? It's 3D mainly. All right, next up. Opus. Opus. 5.5. Claude is back.
157
00:26:00.504 --> 00:26:13.464
All right, so if you've been following Claude or if you're a Claude Opus user- Me... you may have noticed that lately the Opus releases have been a little bit weird and funky and, like, they just...
158
00:26:14.164 --> 00:26:22.444
The vibe was off, for lack of a better word. Opus 5 really had a lot of online dislike for it. Oh, really? Why is that?
159
00:26:22.504 --> 00:26:33.744
Personally, when I would interact with it, I just found its responses were always, like, really convoluted and kind of just, like, word vomit and, like, confusing to understand. I thought 4.8 was really good.
160
00:26:33.824 --> 00:26:36.104
Uh, people loved 4.6. I thought 4.8 was still pretty good.
161
00:26:36.424 --> 00:26:45.364
And then Anthropic was saying, "Oh, well, 5 is sort of a new thing, and you should, like, get rid of all of your previous, like, skills, documents," uh, 'cause it, like, understands stuff.
162
00:26:45.644 --> 00:26:53.604
But it's just everyone was just like, "It's just not really great." And then in the meantime, Fable came out. Fable's very powerful and good, very expensive.
163
00:26:54.064 --> 00:27:00.904
And then Astra started, you know, taking all the buzz for everything it could do with- Right... especially with video editing and 3D modeling.
164
00:27:01.364 --> 00:27:18.144
5.5 comes out, not such a powerful name, but I have found it's like, I mean, for the stuff I'm doing, which is, like, basic tool building and then also 3D stuff, 5.5 has been, like, at Fable level, h- bare fraction of the cost.
165
00:27:18.644 --> 00:27:32.344
Quality's back. Responses are back. Uh, everything about it feels really good, and it's like, 3D wise, on par or better than Astra. There you have it, folks. Joey endorsed Opus 5.5. That's it.
166
00:27:32.724 --> 00:27:41.804
Now it's great because now I have both models, so when I burn the usage of one, I just jump to the other one and then until the other one resets. So, uh- [laughs] Yep. Good, good, good strategy.
167
00:27:42.144 --> 00:27:51.484
Diversifying burn, burn rates. Um, uh, Opus 5.5 integrates to Blender the same way GPT-6 Astra does, right? It could, yeah.
168
00:27:51.544 --> 00:27:58.664
I mean, and, and, you know, we talked about before, like, fi- I mean, you've been able to integrate with Blender. What it can do with it is a different story. Right.
169
00:27:59.704 --> 00:28:18.684
But Opus 5.5, all right, like in a specific example, I was trying to, like, match some, some shots that were shot with like a 3D environment that I'm trying to build behind them, and Astra did an okay job, but there were some shots where it was like definitely the perspective and sizing of like between going from like wide shot to close up, it wasn't quite getting it correct.
170
00:28:19.364 --> 00:28:30.064
And Opus 5.5 reviewed the same shots and was like, "Oh, like the blender lens has been at the same focal length, like the entire, for every single shot. Like all the focal lengths need to change and adjust."
171
00:28:30.684 --> 00:28:47.624
And it like was able to correct all of the focal lengths for every shot and- Nice... export new background plates that were correct. So it caught that, that Astra was having trouble with. Wow. It's outsmarting Astra.
172
00:28:48.024 --> 00:28:59.844
So now the other thing is like Anthropic does not still, you know, there's no native image generation, but people have been sh- doing crazy stuff of having it generate visuals, um, either with like, uh, 3DJS or other tools.
173
00:29:00.344 --> 00:29:03.824
Someone, Ryan Sale, built a- Oh, this is amazing. I love that...
174
00:29:03.944 --> 00:29:17.704
uh, yeah, he asked Opus 5.5 to explain camera focus by building an interactive lens lab, and it built this very cool looking, um, interactive website with a 3D display showing focal plane and how focus works with a lens, and it looks very beautiful.
175
00:29:18.224 --> 00:29:27.864
Wow. Magnific did an example where they gave a brief to both Opus 5.5 and GPT-6, uh, GPT-6 Soul.
176
00:29:27.964 --> 00:29:38.464
So yeah, also GPT-6 came out with Soul and Terra variations of, uh, the 6 model, which they're saying is like for most stuff, probably just as good as Astra and also way cheaper. Yeah.
177
00:29:38.504 --> 00:29:48.044
I bet Jeffrey, Jeffrey's looking at this like, "Oh, shit, we need to build [laughs] we need to build a studio." [laughs] Yeah. This is also why I, uh, I don't think the tool route is the best route.
178
00:29:48.124 --> 00:29:57.024
I don't think the picks and shovel analogy works in today's age. Anybody can build a pick and shovel. Yeah. Everyone has their own foundry to forge whatever you want. Uh- Ooh, ooh, I like that, Joey...
179
00:29:57.044 --> 00:30:05.764
whatever tool you want. We're all blacksmiths. There you go. There we go. Yeah, we're all blacksmiths. [laughs] We've all got a, we've all got a smelting- Yeah... whatever freak.
180
00:30:06.404 --> 00:30:17.464
My knowledge of blacksmithing is a bit freaky. We've all got a... We can all smelt and forge whatever tool we want. You're absolutely right. Yeah. The shovels are cheap. The axes are cheap.
181
00:30:17.784 --> 00:30:23.964
It's, it's really about building the log, log cabin. Go, go do that however you can. Yeah. Right? Yeah.
182
00:30:24.484 --> 00:30:35.544
This is not the best test for a [laughs] podcast, but I feel like the Opus 5.5 example of whatever it built using CDance is- Feels more cinematic to me. Yeah. More shot design, more shot variety. 100%. Yeah.
183
00:30:36.092 --> 00:30:46.672
Yeah, like the, the shot of, um, the reflection on his sunglasses, I thought that was, that was a really nice touch. Mm-hmm. Um, e- some even ma- may call it a creative touch, but- Ooh.
184
00:30:46.672 --> 00:30:54.912
[laughs] Again, that's, that's up to you. And yeah, this is a collection of other visuals and stuff that Opus 5.5 built, but using JavaScript or other coding tools.
185
00:30:55.112 --> 00:31:05.552
Um, or, I mean, you could always give it access to image and video generation and let it steer it. Okay. You, you've inspired me to go use Opus 5.5. I, I am now going, gonna go off.
186
00:31:05.632 --> 00:31:15.832
Yeah, this is, um, the best Opus in a while. Opus is strong, and the usage is, like, pretty good. It's not burning. Like, I've been using it instead of...
187
00:31:15.872 --> 00:31:28.852
Like, I still have Fable usage left on my weekly allotment, which normally I would burn through by, like, Tuesday, but Opus 5.5 has been able to do the same stuff at the same quality that Fable was doing for me. Yeah.
188
00:31:28.892 --> 00:31:38.932
I mean, I'm not doing, like, insane, crazy stuff, so I'm sure there's still uses for Fable. So you think, like- Makes sense... Fable is like the Arri Alexa and Opus 5.5 is like the Blackmagic Ursa- [laughs]...
189
00:31:38.952 --> 00:31:54.072
where it's, like, 90% of that for 10% of the cost? Yeah, that's a good analogy. Yeah. [laughs] Sure. [laughs] Or Fable is, uh... Yeah, Fable's the, um, the, the Arri 65. Like [laughs] Oh, oh, it's like the king. Yeah.
190
00:31:54.072 --> 00:32:06.152
Yeah. It's, like, big, but it's like, okay, well, but I don't need- Yeah... to shoot 65 mil- Okay... for everything. Then GBT-6 Astra is a Sony Venice 2. Sh- Like, it's also equally- Equally as good... crazy. Right.
191
00:32:06.672 --> 00:32:18.612
Yeah. Right. Yeah. [laughs] Uh, sure. We'll go with this analogy. [laughs] So is GBT-6 Sol, like, a equivalent of Fable 5.5? What are we... What is the difference between Sol and Astra? No. Sol...
192
00:32:18.752 --> 00:32:27.992
Astra is, like, the top tier. Astra is, I think, I would say Astra is their f- is the Fable equivalent. That's publicly- Right... released.
193
00:32:28.032 --> 00:32:36.712
I think they have other, you know, more powerful secret models, but the- Yeah, like the ones that they're trying to shut down... publicly... Right, yeah. [laughs] Publicly available models- [laughs]...
194
00:32:36.792 --> 00:32:50.432
uh, Astra and Fable are the frontier top models from, from OpenAI and Anthropic. And then, uh, below that, so I would say, like, Sol is the Opus equivalent. Opus, yeah, okay.
195
00:32:50.572 --> 00:33:01.252
So Opus, maybe Opus 5, but they don't have, like, a middle-of-the-road product like 5.5? They have a... Yeah, middle is, um, Terra is the other- Terra... Terra. That's right. Sol and- Yeah... it's Sol, Terra, then Astra.
196
00:33:01.312 --> 00:33:08.572
So GBT-6 now has a Sol and Terra model variations. I haven't used them. I don't know... You know, they're cheaper.
197
00:33:09.072 --> 00:33:21.452
Uh, you know, so it's like, can you get away doing what you were doing with Astra with Sol and Terra because you're gonna save, you know, usage. I, I would love to know, like, how they value engineer these models, right?
198
00:33:21.492 --> 00:33:28.392
And the, like, the way things are done with, like, cars, right, is, um... And this is terrible practice.
199
00:33:29.052 --> 00:33:41.272
So, like, BMW will make one engine that makes 300 horsepower, and they'll put it on the 5 Series, and then they'll, like, control it with the electric chip to make less horsepower and put it on the model, the 3 Series.
200
00:33:41.552 --> 00:33:49.912
But it's the same engine. They're just, like, value engineering it- Yeah... so, like, it makes sense at that price point. Like, it's just easier to make a bunch of the same engines and just sort of- Exactly...
201
00:33:49.932 --> 00:33:53.132
sort of, like, gate it. [laughs] Throttle it. Yeah, gate, gate it, yeah, and throttle it.
202
00:33:53.232 --> 00:34:06.252
And, uh, so I, I don't know if these are all fundamentally different models, or it's just all running Fable but compromised, cheaper versions of Fable, just so, like, it incentivizes more, uh, frequent usage.
203
00:34:06.532 --> 00:34:19.012
So somebody like you and me, who are, like, token conscious are, are gonna go for 5.5, whereas the ballers, the Jeffries of the world, are gonna use the- [laughs]... the Opus, right? Yeah. I don't know- Or Fable...
204
00:34:19.072 --> 00:34:30.192
how the models are... Yeah, I don't know how the variations of the models are constructed. Um- Yeah... also am, I'm still fuzzy on what effort level makes the most sense for what task. That is still- Yeah.
205
00:34:30.612 --> 00:34:32.312
If, if our viewers know- Yeah...
206
00:34:32.352 --> 00:34:44.272
this stuff, like, please educa- this, again, uh, we're just throwing up our hands here, but we'd love to know how the different price tiers work out, uh, uh, in terms of how the model is actually built and inferred.
207
00:34:44.692 --> 00:34:54.092
Yeah, and what effort levels you use, and if you get good results, especially related to Blender and, and, uh, Resolve usage, 'cause that's the bulk of my use cases for these things.
208
00:34:54.552 --> 00:35:04.912
I usually crank it up, and then I'm like, "Ah, I'm probably gonna burn through my usage more," but I wanna give it the best. [laughs] True. [laughs] All right, last one. New image model, Qwen Image 2.1, open weight. Yes.
209
00:35:05.772 --> 00:35:13.652
I don't have any generations. Damn it. You didn't do your homework, Addy. All right. Didn't do my homework. So let me look for some stuff, but yeah. Uh, vibe of this has been
210
00:35:14.772 --> 00:35:23.712
hearing, like, Nana Banana quality, but it's open weight and way cheaper. Yeah, and, uh, you can run it on Comfy, obviously. Um, maybe we can look at some examples online. Yeah.
211
00:35:23.772 --> 00:35:33.652
I thought Qwen Image was at 3, but yeah, I don't know what... 2. There might've been another model that's at 3. Mm. I know they have a new- Okay... text-to-speech model as well. Yeah, they should name everything Qwen.
212
00:35:34.012 --> 00:35:42.932
Qwen, or as, uh, iMessage autocorrect likes to call it, Owen. Owen. [laughs] It's, it's... Well, let's see. We got transparent layers. We got text detail. Text, yeah.
213
00:35:43.412 --> 00:35:52.692
So it's really made for graphics and animatics and animation. Image editing abilities. This is an example of, uh, a tree in front of the Seattle Space Needle, and then it has- Mm...
214
00:35:52.752 --> 00:36:04.712
removed the, just leave the leaves, and it does it pretty good. Mm, that's not so hard if you extract a depth map. Okay, man. Yeah. [laughs] This is hard. This one's hard. So it did this well.
215
00:36:04.872 --> 00:36:11.872
Isn't that the cast from Cheers? This is Cheers, yeah. So it's a bunch of photos of the- Yeah... Cheers cast, and they put them all into a portrait. Up to 10 reference images. Yeah.
216
00:36:11.872 --> 00:36:19.232
Yeah, almost kind of hosting- But look, I mean, I think the big headline of this is like, yeah, look, is this gonna be better than anything you've seen? No. But this is open weight. You could run this locally. It's free.
217
00:36:19.412 --> 00:36:32.312
Yeah. It's open source. You can run basically Nana Banana on your own. Yeah, this looks like, everything here looks, like, on par with, like, Nana Banana 2 quality. Yeah. Again, do you think, conspiracy theory
218
00:36:33.352 --> 00:36:46.124
Addy is saying this, do you think they're doing this, Alibaba's doing this to kind of, um- Pull users away from Nana Banana usage? Possibly. Or, I mean, just to have a, for the- Just to weaken their market share...
219
00:36:46.184 --> 00:36:57.444
Chinese companies to have a competitive model. Yeah. Um, yeah. I mean, why... Yeah, why not if you could, if you could do it and release it and gain adoption? Right.
220
00:36:57.524 --> 00:37:04.484
I mean, we talked about this before, like really the advantage that the Chinese companies have is just putting stuff out open and getting adoption. Yeah.
221
00:37:04.884 --> 00:37:16.344
Uh, I mean, the Chinese models are okay, like the image models specifically. They're just trained on Chinese, uh, people data, which is, makes sense. It's Chinese model.
222
00:37:17.084 --> 00:37:26.784
Um, so they don't necessarily do the American diversity very well. [laughs] You know, like if you put Addy next to Joey and try to generate that, uh, it, it- I-... it'll struggle a little bit more than what GPT can do...
223
00:37:26.804 --> 00:37:33.764
think if you have reference images, yeah. But I think if you're doing text- Yeah... to image- Yeah... and you don't give it specifics, I remember this like a long time ago with some of your- Mm-hmm...
224
00:37:33.884 --> 00:37:42.244
like early Kling, and it's just like convenience store, and then it would just be a very Chinese-looking convenience store, um- Right. Exactly... or a cow.
225
00:37:42.284 --> 00:37:49.724
Like if you don't very clearly specify what you want, it'll default to more- Yeah... um, yeah, Chinese-looking environments.
226
00:37:49.764 --> 00:38:03.824
In my opinion, I think the, the king of image models right now is GPT 2.5 Sunburst and Nana Banana 2, still really, really good and, and much faster in inference, uh, as I'm sure Google is working on the next image model.
227
00:38:04.464 --> 00:38:13.544
And then on the video side, I mean- Well, we need a new, uh, we need a new Pro update 'cause I still feel like Nana Banana Pro is better than 2. Oh, you think so? Yeah. Like I'll still just u- default to 2.
228
00:38:13.664 --> 00:38:27.184
Uh, image, GPT Image 2.5 is very good and using that more, but I still think- Yeah... between that and Nana Banana Pro over Nana Banana 2. Yeah. I, I mean, I don't really use the Qwen models too much.
229
00:38:27.804 --> 00:38:44.204
They do have Qwen 2512, which is not Image 2.1. Like Qwen has multiple image models. That was really good when I last tested that about a few months ago. I remember that mouthful of a name. What was unique about it?
230
00:38:45.124 --> 00:38:58.324
Uh, so 2511 I think was text to image, and 2512 was, uh, multi-reference to image. I don't know why- Is this an extension of the same model and they just cleaned up the naming because that's very confusing?
231
00:38:58.424 --> 00:39:10.904
[laughs] Perhaps. Perhaps. Yeah. You're right. Yeah. I, I kind of preferred like the scientific names. It was easier to remember. Um- It sounds like a film stock. [laughs] Kodak tw- 7510... Kodak Qwen 2512. Yeah. Yeah.
232
00:39:11.304 --> 00:39:24.164
And then on the video side, I think, uh, most people agree that CDEN 2.5 is currently the king. Um, although my personal favorite is Wan 3.0. It's been solid for me. Okay. That's cool. Yeah. That's cool.
233
00:39:24.224 --> 00:39:41.144
Yeah, CDEN 2.5, good, just pricey, but it seems like they're putting more routes in that is like generate at a cheaper fast mode or a cheaper, uh, lower resolution mode, and then up res to what you want, but still
234
00:39:42.784 --> 00:39:48.984
be able to get the same output and not have like a new spin. So it seems like they've been refining that kind of workflow
235
00:39:50.104 --> 00:39:57.024
because it is so expensive to like do cheaper generations but then still adhere to like the vision that you want once you're ready to, to go. Yeah.
236
00:39:57.044 --> 00:40:04.504
They gotta capture the, the market that's cost conscious and still use their... It's the same argument that I made with Fable and Opus.
237
00:40:04.544 --> 00:40:18.224
Like you gotta have the CDEN Opus version of that model where, you know, we can mess around with it. Mm-hmm. Yeah. Built out this whole interface, this whole web mockup too. This is Qwen. Yeah. It looks very Google.
238
00:40:18.604 --> 00:40:30.124
Yeah. And- That's cool... detailed diagrams and stuff. Yeah, I mean, so e- everything it does is, is, is stuff you have seen before, but it, uh, you can run this locally. The power we have on our local GPUs. Sure.
239
00:40:30.204 --> 00:40:38.964
I heard something interesting, um, online, and I wanna hear your take on it. So- Okay... Adobe acquired Topaz, and we talked about it before.
240
00:40:39.384 --> 00:40:46.744
And one of the posts said that the reason they acquired Topaz, I, I mean, how, how could this person know, right?
241
00:40:46.784 --> 00:41:00.184
[laughs] Like they're not part of the deal, but I guess they're just speculating, is really good at running local tasks on GPUs. Mm-hmm. Like they have a downloadable, uh, application. I think it's called Astra, right?
242
00:41:00.963 --> 00:41:04.744
Where- Yeah. Topaz Video. I mean, that's, that's what they started on. That was what they were- Right. Right... they were the original. So it's like,
243
00:41:05.664 --> 00:41:16.784
uh, maybe Adobe acquired them for that capability or, you know, like sort of the local GPU mo- like the comfy UI model, if you will. What are your thoughts on that? I think it's the other way around.
244
00:41:16.844 --> 00:41:26.984
I mean, I don't really think who's gonna be running local. Yeah. And also, how does that fit into, no shade to Adobe, but out of all of the companies that love to charge- Yeah... subscription fees. Exactly.
245
00:41:27.104 --> 00:41:36.164
Like where's the business model if it's local, right? [laughs] I don't see Adobe offering, offering options to run stuff local. [laughs] I agree. I agree. I just wanted to hear your take on that. Yeah.
246
00:41:36.524 --> 00:41:45.484
Uh, a lot of people are still fixated on local. Look, I mean, the local stuff's great, and I, I mean, I use the Topaz local models as well. I mean, and, and i- it's been a hybrid.
247
00:41:45.524 --> 00:41:50.084
It's been, they have their generative upscalers and, and sometimes they're great, sometimes they work really well,
248
00:41:51.164 --> 00:42:00.524
and then sometimes they don't, depending on the footage and, and sometimes just a regular, their, uh, GAN models like Gaia and stuff just works better for- That's true... the task at hand.
249
00:42:01.104 --> 00:42:12.304
And, uh, did you hear- I, I'm hoping that Adobe still keeps the dedicated Topaz Video app, like as an option. I hope so. Yeah. Or maybe it's just another thing on Creative Cloud, right? Like the- Or Creative...
250
00:42:12.384 --> 00:42:22.244
Right, right. Yeah. It's included with your subscription. It's part of your subscription. Yeah, maybe. Yeah. I heard that NVIDIA is not really releasing any new s- consumer GPUs next year. Did you hear that?
251
00:42:22.584 --> 00:42:32.104
I did not hear that. Uh, I heard Jensen's interview on New York Times. Okay. Where- Where- Yeah. [laughs]... it was interesting... he's like, "Shut down the labs." No, he was li- no, the opposite.
252
00:42:32.124 --> 00:42:36.904
He was like, "It's, uh, it's all..." He's like, uh, he's like, "This is alarmist talk. It's not helping anyone.
253
00:42:37.124 --> 00:42:44.464
Like the labs should not release dangerous things, but saying it's gonna destroy the world is not helping anyone." [laughs] That's true. Um- Yeah... okay.
254
00:42:44.604 --> 00:42:53.684
So, uh, maybe that's a more recent Jensen interview where he's like now come around that position, so possibly. Oh, yeah, it just came out this week. Yeah, it was w- Yeah... out with, uh, Ezra and Klein- Yeah...
255
00:42:53.724 --> 00:42:58.704
on the New York Times. Yep. Yeah. It's a good interview. Fun, fun times. All right. Wrap it up here.
256
00:42:59.164 --> 00:43:06.024
If you've been watching the show or even listening to it on Apple Podcast, Spotify, or YouTube and have not subscribed yet, please do so.
257
00:43:06.464 --> 00:43:16.264
That'll be a really big help for us, and if you wanna hear some of the things that we have not yet touched on, comment on them and we'll see that. Links for everything we talked about at denoisepodcast.com.
258
00:43:16.764 --> 00:43:19.184
Thanks for watching, and we'll catch you in the next episode.
1
00:00:00.120 --> 00:00:11.780
You can't deny the power that we now have with just a couple years of development. I can only imagine 10 years into this, the way VFX and traditional computer graphics have had two or three decades.
2
00:00:11.880 --> 00:00:28.100
Like, if AI gets a good solid decade in, where will we be? Welcome back to Denoised. Addy, how's it going? What's up, Joey? Long time no see. I know, in the virtual space. Okay, so since our last episode, we had
3
00:00:29.100 --> 00:00:35.940
a little, little, uh, comment, heat in the comments, and [laughs] a little, uh- Things got a little heated, didn't it? A little heated, yeah.
4
00:00:36.080 --> 00:00:50.080
So our last episode was about, well, the thumbnail was about how when I was at TIFF and I talked to people about hybrid AI, as soon as they heard the word AI, they assumed sloppy microdrama kinda stuff. Right.
5
00:00:50.180 --> 00:01:00.380
Uh, or just slop in general. The nuance of the episode was, as I explained that we were talking about shooting with real people and using AI to help production, then it was like, "Oh, okay, that sounds pretty cool."
6
00:01:00.820 --> 00:01:14.680
The comments were the exact [laughs] knee-jerk reactionary comments of, "Yeah, AI is slop." N- well, yeah, there is, there is, I think, a percentage of folks that'll just say e- you know, AI is slop.
7
00:01:14.720 --> 00:01:27.940
But then what I was shocked by was a lot of the commenters were M&A people, like, you know, VFX this, uh, studio that, like some of their screen names. And the comment is like, "Well, AI will never blank.
8
00:01:28.320 --> 00:01:40.040
AI will never..." Like, how... I don't understand how you could be so sure that it'll never do certain things, because [laughs] we're just, like, year two or year three into this thing.
9
00:01:40.880 --> 00:01:47.380
Yeah, there literally are things where it's like, yeah, I don't think it'll be that good at that. And then a week later... Even the past few weeks, the amount of progress- Yes...
10
00:01:48.060 --> 00:01:57.500
if, the things it can do with Blender and computer control has been, I would not have expected that or expected that- Exactly... this year Like, it surprises us, and we're skeptics, right?
11
00:01:57.600 --> 00:02:07.180
And if you follow the podcast journey from, like, a year ago to now, we have had [laughs] our doubts on certain things, and now we're, we're s- surprised ourselves, right? Like,
12
00:02:08.340 --> 00:02:26.260
but then again, I think the mindset is, is that we are on this journey into, um, watching this technology sort of unfold and empower us in making better creative choices, you know, unlocking financial opportunities that may not have existed with traditional VFX.
13
00:02:27.100 --> 00:02:37.240
And we're just in for the ride, right? Like, and we're covering it in our own lens. We're not championing for slop. We never have, [laughs] right? Like, we hate slop as much as...
14
00:02:37.300 --> 00:02:44.540
I mean, I'll laugh at it as a meme, but I, it's not consumable content to me. Yeah, it's not the, what I want my main entertainment to be. Yeah.
15
00:02:44.640 --> 00:02:50.120
And we're constantly, um, sort of, uh, keeping an open mind as this thing progresses.
16
00:02:50.220 --> 00:02:58.460
Of course, there are some things that happen in the AI space that you and I don't agree with, and we talk about it, especially on the ethics side, right?
17
00:02:58.480 --> 00:03:06.480
[laughs] Uh, like when it comes to, like, a Tilly Norwood and, and sort of the drama around that, like we're, that's distasteful, I think, in my opinion.
18
00:03:06.720 --> 00:03:14.200
But having said that, you can't deny the power that we now have with just, just a couple years of development.
19
00:03:14.440 --> 00:03:27.380
And I can only imagine 10 years into this, the way VFX and traditional computer graphics have had two or three decades. Like, if AI gets a good solid decade in, where will we be?
20
00:03:27.780 --> 00:03:38.500
Yeah, and the speed everything's going and where we're at just now, three to five years into when, like, we had crazy Will Smith eating spaghetti, and where we're at now. Yeah.
21
00:03:39.120 --> 00:03:48.660
Uh, and yeah, and it's always like, this is the worst it will ever be. Yeah. So again- Yeah... um, uh, uh, look, shout out to some of our regular listeners who sort of just commented on- There were-...
22
00:03:48.700 --> 00:03:54.420
on regular things Yeah, and supportive. Yeah. Uh, but- Or no, there were a lot of support comments in there- Yeah... from regular listeners, so we appreciate that. Yeah.
23
00:03:54.440 --> 00:04:01.520
But yeah, I found it amusing because it was like, oh, all the comments literally validated the exact premise of the episode, which was [laughs] people hear AI- Yeah...
24
00:04:01.580 --> 00:04:10.880
and just have a knee-jerk reaction And they immediately have a knee-jerk reaction. Right. You're only doing yourself a disservice if you're not curious about it, if you're not willing to test it out.
25
00:04:10.960 --> 00:04:20.560
I, I think if you just sort of face away from the entire thing and just pretend that this is gonna go away because it's wrong or whatnot, I don't think it is.
26
00:04:21.040 --> 00:04:32.060
I don't think it's gonna go away, regardless of how it was built. The question is, like, can we build, build it better, build it more ethically now that it's here and it's kinda proven itself? Yeah.
27
00:04:32.120 --> 00:04:41.000
And I, actually, we'll get into a lot of, I, this, I think, more when we talk about, uh, Katzenberg's essay, uh, in a minute. There was one comment, though, that I did find funny.
28
00:04:41.260 --> 00:04:50.060
"What does any of this have to do with creating art? Quote, 'Data, spreadsheets, outputs.' Do you even hear yourselves?" I think that's about Jeff. Oh. But [laughs] that was really- Okay, yeah.
29
00:04:50.200 --> 00:04:52.160
I, I think- Because [laughs]
30
00:04:52.220 --> 00:04:59.580
I thought it was funny though because I was thinking like, let me tell you, there's not a single film on Earth that exists that a spreadsheet was not involved- And I will correct that person-... to make it possible...
31
00:04:59.600 --> 00:05:08.420
because early animated movies were made with spreadsheets. Yes. Like animation tables are- Yeah. Yes, absolutely... r- they're still spreadsheets under the hood. So spreadsheets- Yeah... can make art, dude.
32
00:05:08.520 --> 00:05:22.700
[laughs] Or dedead. Deater, deader deadler. Yeah. Okay. Katzenberg. Yeah, my, my old, uh, boss's boss's boss's boss. [laughs] You work directly for Jeffrey Katzenberg? I wish.
33
00:05:22.780 --> 00:05:28.280
Uh, actually, um, yeah, I mean, I, I, I have a lot of personal anecdotes of Jeffrey Katzenberg.
34
00:05:28.340 --> 00:05:36.240
I worked at DreamWorks Animation during sort of the golden years in the early 2000s, uh, when Jeffrey was very involved in running the company. He,
35
00:05:37.220 --> 00:05:47.820
you would drive in, you would see his car already on the driveway, you know, as you're driving in, and then as you're driving out of DreamWorks going home, his car is still there. That guy was a workaholic.
36
00:05:47.860 --> 00:05:54.320
He was there day in and day out. Even on some weekends, you would see him. Uh, he had a team of assistants, and, um,
37
00:05:55.780 --> 00:06:03.300
I was in my 20s then, and I, um, I had a couple of, like, meetings and dinners with him here and there, and, um-
38
00:06:04.624 --> 00:06:20.504
As much as people like to bag on his sort of set of choices and, like, I guess you could say, quote-unquote, "ruthless style of leadership," I found his work ethic and sort of his sense of focus really inspiring.
39
00:06:20.624 --> 00:06:25.864
So he's actually one of my heroes [chuckles] from my early days of my career. Yeah. That's awesome.
40
00:06:25.944 --> 00:06:37.824
I mean, yeah, he did a lot of car- good stuff like, uh, so I mean, Jeffrey Katzenberg, uh, part- initially in Disney during their '90s revivement of, like, the golden age of Disney. He brought Disney back to life. Yeah.
41
00:06:37.924 --> 00:06:53.904
Like, Disney was going through crazy malaise, uh, during the '80s, and then in the '90s, Katzenberg was behind The Lion King, Aladdin, you know, um, all of, like, the hits that you associate the classic Disney era with, uh, with the 2D animated look.
42
00:06:54.264 --> 00:07:00.324
Mm-hmm. That's primarily- Lion King, Aladdin... yeah, his brainchild. Side thing. Did you listen to the Acquired podcast? No. Was he on there?
43
00:07:01.224 --> 00:07:12.504
No, but they did a two-part Disney history, Disney episodes that were, uh, excellent. Um, they- I am a big fan of the Acquired podcast, [chuckles] so I have to go back- Yeah... and catch that. Yeah.
44
00:07:12.604 --> 00:07:20.844
Uh, they're the most recent ones. They were, yeah, in total, it's like eight hours, nine hours, and I s- [chuckles] I still wanted more. [chuckles] Tell me more.
45
00:07:20.844 --> 00:07:29.004
[chuckles] But yeah, part two is on the modern age, the post, uh, Dis- Walt Disney era, and that's where it gets into a few more about Katzenberg. Yeah. That's like Michael Eisner, Bob Iger.
46
00:07:29.024 --> 00:07:30.864
Eisner, Katzenberg, and, um, Frank Wells.
47
00:07:31.124 --> 00:07:42.164
Fascinating sort of view of leadership and how the, how a massive organization like that evolved and kept up with its times rather than sort of just dither away and die, kinda like an IBM.
48
00:07:42.184 --> 00:07:54.344
It was on the verge of dying multiple times. [chuckles] Multiple times. Yeah. Okay. Katzenberg, Disney, then left Disney, started DreamWorks, then left, then started Quibi. [chuckles] RIP Quibi.
49
00:07:54.384 --> 00:08:04.984
Uh, he actually started a couple of other startups before, uh, Quibi. Um, I had the pleasure of working on a platform called Pitch, P-T-C-H.
50
00:08:05.704 --> 00:08:15.104
This was in 20- Back when it was cool to, uh, omit, omit the, uh, vowels from the name. Yeah. [chuckles] Yeah. This, this, this was like, um- It's like the-... pre-TikTok, pre-Vine. Yeah.
51
00:08:15.184 --> 00:08:29.064
Pitch was the first image and video social media platform, and it was designed really well, and, uh, Jeffrey was so- just so... He was, like, 10 years ahead of his time. This was like 2011, 2012, something like that.
52
00:08:29.724 --> 00:08:42.844
And as you know, TikTok and everything really hit big, uh, in the, in the late 2010s, right? And so I think the limitation was really on the network and the phone itself rather than the, the platform. That makes sense.
53
00:08:42.844 --> 00:08:54.624
Because our phones weren't able to really download video and play back video well, and then, you know, we're around 3G or 4G then, which was not conducive. Yeah. The early iPhones couldn't shoot video. Yeah.
54
00:08:54.664 --> 00:09:07.624
I mean, this, this pl- It was- This- So, uh, Pitch actually had, um, iPhone or mobile editing on its platform then. Like, you could shoot a video and edit it in your phone, which was pretty novel at the time.
55
00:09:08.404 --> 00:09:18.944
That's cool. Yeah. Other startups, K- uh, Quibi. Then I didn't realize he's been in the Bay Area for, like, 10 years. Um, and so he ha- he's had a venture fund, and he's been investing in various startups.
56
00:09:19.264 --> 00:09:24.724
And then this week, he posted a pretty long essay on X called The World Is Changing: AI for Creativity.
57
00:09:25.144 --> 00:09:35.704
The essay didn't really announce anything, but it was sort of, like, his observations and where he's seen AI and how it's being used in creativity. Now, three years ago, he did say something to the effect of
58
00:09:36.724 --> 00:09:49.244
that he predicted three years ago, in three years, AI would either replace or be able to do 90% of what, uh, it would take to do a high, uh, high-quality animated film.
59
00:09:49.884 --> 00:10:01.124
We're not quite there yet, but I think we're pretty close to that. Just so our haters are aware, [chuckles] we just said we're not quite there yet. We're getting close. Look, well, A, I think we talked about this.
60
00:10:01.184 --> 00:10:11.924
Animation's the e- Animation's the easiest one that could be- It's a easier target, for sure... assisted. Yeah. Assisted first. Yeah. Yeah. Because you don't have the uncanny valley issue. Uh, it works well with styles.
61
00:10:11.964 --> 00:10:20.984
We know animation works pretty well- Yeah... with AI assist. Animation is almost there. So to quote, "I asked one of the leading AI models creators a question that had been challenging me for months.
62
00:10:21.064 --> 00:10:27.004
What is the difference between reasoning and creating? Its answer changed how I think about almost everything happening in the industry."
63
00:10:27.224 --> 00:10:35.364
So he's looking at it like right now, AI models are really good at reasoning, solving things, predicting, not good at creativity, not good at creating. Yeah.
64
00:10:35.624 --> 00:10:44.044
We talk about creating is the act of creating something net new from your old experiences, and this is what we're so good at, right? Mm-hmm.
65
00:10:44.084 --> 00:10:53.464
Like, if you think about Tarantino or whoever, I just saw Once Upon a Time in Hollywood [chuckles] very recently, so, like, I'm thinking, uh, I'm in my Tarantino head.
66
00:10:53.584 --> 00:11:02.604
Um, like, he comes up with those ideas from all of the thousands of movies that he's watched. He's, like, a massive cinema fanatic, right?
67
00:11:02.664 --> 00:11:14.564
Like, so that database is in his head, and he is pulling threads from this movie, that spaghetti Western, and that, you know, w- sci-fi or whatnot. But then the thing that he's creating is net new,
68
00:11:15.464 --> 00:11:30.004
and this is what AI models just can't do yet, is just effectively create net new things. But I don't think they have to, because that's why we're here. You know, we can essentially have AI to help us with that.
69
00:11:30.504 --> 00:11:39.964
Yeah, and that's where we get it. The, the, the human, the person is still the creative driver using these tools. Quote, "Creating is fundamentally generative. It produces something that didn't exist before.
70
00:11:40.304 --> 00:11:52.264
And crucially, there's no single right answer waiting to be found." Exactly. Like, when you are, uh, building a movie, you take left turns and right turns, and you turn the script into something it, it wasn't originally.
71
00:11:52.764 --> 00:12:03.644
Right. And that's part of the creative process. I don't think anybody is really expecting AI to do that. I mean, maybe, maybe on the social media side. I mean, I think in the, yeah, in a social media level- Yeah...
72
00:12:03.664 --> 00:12:14.664
like, there are a lot of tools and pushing out of, like, you know, the, your AI ag- director agent, or, like, here, give it a prompt, give it a script, and it'll, like, do all of the whole creative brief and generate a 30-second thing for you.
73
00:12:15.504 --> 00:12:18.184
I think that, like, people are pu- companies are pushing that
74
00:12:19.252 --> 00:12:29.472
Maybe that'll be some social niche genre or something, but yeah, I don't think that's gonna be at the highest level just- Well, that's not what we're gonna talk about on this podcast 'cause we're not interested in that, right?
75
00:12:29.472 --> 00:12:38.852
[laughs] If you're giving nuance to the haters who are never gonna get this deep into actually listening to the podcast, I'm assuming. Yeah, the, the, we've already lost them in the first 10 seconds.
76
00:12:38.852 --> 00:12:48.772
[laughs] So the gist of the essay is a, well, it's sort of the tale of two cities. So he's up in the Bay Area, they're, they're building the tech, they're building these powerful models, they're engineers.
77
00:12:49.352 --> 00:12:58.932
LA has the creative, creative talent and has all of the, the creative people, and he's, like, looking for a way to, like, sort of bridge these two worlds together. Mm-hmm.
78
00:12:59.012 --> 00:13:10.392
And also for less of the trying to push and resist the AI in general, and more, like, mess around, figure it out, so you could also define the rules and how it comes into play- Mm-hmm...
79
00:13:10.912 --> 00:13:23.632
on your terms and, like, be able to define how this works instead of letting it be defined- I mean-... for you or for the industry by tech and engineers up north. He's, he's a little... I love Katzenberg. Sorry, dude.
80
00:13:24.552 --> 00:13:30.932
You're a little late to the party, JK, 'cause, uh, we have hybrid setups like that, you know.
81
00:13:31.092 --> 00:13:44.692
And I, I'm looking at, you know, organizations like, uh, Netflix, and Fox Next, and Disney, and, like, these teams have well integrated with, uh, Bay Area tech, and they have the studio perspective on things.
82
00:13:45.292 --> 00:13:53.972
And then you also have your Asteria, your Promise, your Secret Level, and, like, all these AI, Wander and AI startups who are combining those two things pretty well.
83
00:13:54.032 --> 00:14:05.292
So I think the tale of two cities has already been mashed up, um, and maybe this is an opportunity for Jeffrey to, like, bring new perspective here. Yeah, I think it's early days of that mashup, and also, like...
84
00:14:05.312 --> 00:14:09.672
Well, look, I mean, on a union at, uh, end, there's still, like, a lot that's unclear and unsorted out.
85
00:14:10.192 --> 00:14:21.132
Yes, we know, like, studios and everything are messing with this, but a lot of it's still, like, a hush-hush or, like, in secret, and there's no, like, established guideline. Systemically, it hasn't integrated.
86
00:14:21.192 --> 00:14:28.352
Like, yeah, you're right. Like, at a union- Mm-hmm... level or at, like, a mass, mass scale level, um, AI is still not a thing.
87
00:14:28.552 --> 00:14:38.152
And it's not gonna be a thing for a long time because we know that the quality is not quite there yet. You're just exchanging one set of problems for another, and that's not really- Yeah... gonna help anybody, right?
88
00:14:38.212 --> 00:14:46.892
Like, if you're gonna adopt AI, it has to be inherently easier, better, faster, cheaper, and all that stuff, and we're just not there yet. This is a good, uh, anecdote.
89
00:14:46.992 --> 00:14:55.292
"Edison invented the motion picture but made terrible movies. It took Chaplin, Lloyd, Keaton, and so many others to make movies emotional." There you go. There you go. I think that was a good point. Yeah. Okay.
90
00:14:55.552 --> 00:15:06.612
Now [laughs] He looks the same. [laughs] He's looked the same for 30 years. That's incredible. He's, he's in good shape, man. He posted this essay on X, um, but it didn't really announce anything.
91
00:15:06.832 --> 00:15:12.312
But then he was on the Town, uh, podcast, and- Nice... they asked more questions.
92
00:15:12.332 --> 00:15:23.632
[laughs] So I wanna jump into a few quotes because there are a few things here too that stand out to me as, like, I don't know if that's the way it'll go. 5 billion people a day watch video.
93
00:15:24.192 --> 00:15:34.252
It is the single greatest way of information and communication and knowledge and entertainment. Like, it touches everything.
94
00:15:35.132 --> 00:15:47.932
What if you actually could now develop the tools that all of that can get elevated, all of it can move to a higher quality and a whole set of new things that none of us actually even are imagining today?
95
00:15:48.232 --> 00:15:58.112
That's what got me excited, and that's what I'm trying to figure out. Hmm. Okay. So the thing that stuck out to me there, 'cause he was talking about, like, yeah, people, 5 billion people watch video a day,
96
00:15:59.132 --> 00:16:05.492
but he brought up quality again and getting to, like, the higher quality level. Yeah. And I think that was Quibi's issue. Part, well, part of many of the issues.
97
00:16:05.492 --> 00:16:09.552
It, it was too high, it was too focused on quality when it should've been about the- It was too focused on quality...
98
00:16:09.672 --> 00:16:15.232
about the d- like, the moment and the distribution and the virality and all the thing that TikTok gets right.
99
00:16:15.252 --> 00:16:23.392
Well, look, and the thing he was right about Quibi was, right, he was right that, yeah, vertical videos would be a thing, and people would be into watching content on their phone.
100
00:16:24.492 --> 00:16:31.692
They just, the quality bar does not need to be that high, nor... I mean, this still goes back to, to, um,
101
00:16:32.672 --> 00:16:45.212
uh, Reed Hastings' quote of just like, yeah, the biggest, the bigger risk is just that people's quality bar of what's acceptable just gets lower, and they don't care about premium quality as much. Damn.
102
00:16:45.632 --> 00:16:51.712
Reed Hastings, smart man. Exactly. [laughs] You, we're guilty of this too 'cause you and I are cinema files, right?
103
00:16:51.872 --> 00:17:05.312
Like, we have our eye on quality all the time, almost too much to the point where we miss the opportunity or we don't see how something could be so big when it is, right? And I think Jeffrey the same way.
104
00:17:05.532 --> 00:17:16.672
He epitomizes the studio, right? Like, he's built studios, and so of course he's gonna have an eye on quality. He's still looking to build the DreamWorks pipeline, but for, like, everyone.
105
00:17:17.292 --> 00:17:29.572
But I don't know if that's- I don't think you need the quality for all 5 billion people. I think you do need the- Right... quality for, like, maybe 1% of that. Tongue, tongue to her. [laughs] Duh.
106
00:17:29.612 --> 00:17:38.932
Do you want me to sing it? Shut up. [laughs] Yeah, do it. No. I need, I need to program one of these buttons here to, uh, to have you here- Tongue tongue to tongue to her. All right, I did it.
107
00:17:38.932 --> 00:17:49.772
[laughs] Um- Yeah, it, it's, it's the perfect example. Oh, by the way, the, the coffee cup character, Ballerina Cappuccina, that's who it was. [laughs] Yeah. That's who's great. Yeah.
108
00:17:49.832 --> 00:18:01.212
It was different, and they had tools that were different, and there was- And you believe that the a- the Pixar of AI is one of these companies that is circulating around the Bay Area right now?
109
00:18:01.752 --> 00:18:10.952
No, I think I'm gonna have to start it. Oh, you do? Yes. I was gonna ask, what is your angle here? [laughs] I think I'm gonna have to start it. So that was the most of, like, what he's looking at, sort of.
110
00:18:10.972 --> 00:18:19.412
There's one other sound bite I wanna play of what he's looking at to build. But yeah, so he's like, like, so yeah, we've talked about what is gonna be the Pixar of AI. He thinks he's gonna have to start that company.
111
00:18:19.672 --> 00:18:31.792
That, but that still seems completely separate from, "I wanna build tools-" That make five billion people create better stuff, and I wanna create the Pixar of AI. Yeah, I, I don't, I don't think we...
112
00:18:31.992 --> 00:18:44.112
We don't, we don't really need it right now. I, I think what we need is, um, more the mashing of tech and Hollywood and figuring out what the new studio environment looks like.
113
00:18:44.472 --> 00:18:51.572
One more soundbite and then I wanna have this discussion about, yeah, what does that look like? Yeah. Mine is actually more I go back to my partner saying- Yeah... AI for creativity.
114
00:18:51.992 --> 00:19:01.312
I'd rather build the picks and shovels and provide them to everybody. Yeah. Like, it's not about having them be proprietary only to, to, for, for our use- Right...
115
00:19:01.332 --> 00:19:15.412
but actually the biggest opportunity would be if you could actually create a set of applications and tools and workflows that could be- That everybody has to pay you. Well, everybody's working on that, JK.
116
00:19:16.312 --> 00:19:22.292
Is he trying to build a model? Yeah. That's, that's the thing. [laughs] It's like, okay, if you say- Yeah... if he's doing the picks and shovels analogy- Yeah...
117
00:19:22.332 --> 00:19:30.532
and he's like, "Oh, we wanna build a tool that lets people create stuff." Well, Google's working on that. Runway's working on that. [laughs] Adobe's working on that. Right, and then you've got, then you've got- Yeah...
118
00:19:30.572 --> 00:19:39.332
Magnific, Arts- Artlist, Hixfield. Yeah. Um, all of these. Everybody's building picks and shovels, so how is he gonna build something different here? Right. And I don't even know if that's the play.
119
00:19:39.372 --> 00:19:48.152
I mean, well, so that, yeah, that's where I wanted to go. So I feel like if you're trying to build the Pixar of AI or you're trying to build a tool for everyone, those are completely separate things.
120
00:19:48.312 --> 00:19:57.452
And he was talking about like, oh, I don't want a gatekeeper, like, you know, have proprietary tools. Pixar has a lot of proprietary tools that, like, helps them make their scenes. RenderMan. Well, USD's an open source.
121
00:19:57.552 --> 00:20:02.772
Uh, Presto. Presto is their animation- Their- Yeah... animation system. Mm-hmm. Proprietary. Yeah.
122
00:20:02.832 --> 00:20:15.212
Yeah, I feel like, okay, there's either gonna be the tool set that is web focused, or you're gonna be a studio and you need to have proprietary systems or something that do- gives you an edge to do what you would do.
123
00:20:15.712 --> 00:20:24.752
Unless he's building a model or something that helps. This was a company that he invested in. Yeah, this was a while back, right? Yeah, a long time ago. May. Real-time AI video startup Reactor.
124
00:20:24.912 --> 00:20:28.692
Katzenberg's company, WonderCo, was an investor in it. I was trying to see, okay, what has he invested in?
125
00:20:28.712 --> 00:20:34.931
He invested in a bunch of just, like, Canva and the usual suspects, but this was the most AI-focused thing I could find that was public that he invested in.
126
00:20:35.392 --> 00:20:40.952
Obviously, real-time AI video would be a, an original model and would be useful. It would be a disruptor for sure.
127
00:20:41.252 --> 00:20:51.832
I think, like, Jeffrey's, uh, strength, in my opinion, just having worked under him [laughs] for so long, is that he has the leadership to pull off Shrek, right?
128
00:20:52.392 --> 00:21:00.552
Shrek wasn't necessarily about computer graphics at the time or all of the, uh, wonderful innovations that were happening with CG.
129
00:21:01.412 --> 00:21:10.832
Shrek was really about disrupting the Disney fairy tale format in a way that became a parody of all of the things that he made previously.
130
00:21:11.152 --> 00:21:22.352
And it, it was, uh, built for any audience, any part of the world, and it was massive, right? It was the highest grossing... Shrek 2 was the highest grossing animated movie for, like, decades after he made it.
131
00:21:23.152 --> 00:21:24.472
That's, I think that's his strength.
132
00:21:24.532 --> 00:21:39.492
So it's not necessarily building tech and tools, which quite frankly he's failed at several times, but building incredible movies and a studio system that can generate those movies and build those movies over and over again consistently, like a DreamWorks.
133
00:21:39.752 --> 00:21:48.312
Or revitalizing Disney Animation. Yeah. You can make the argument that no studio is sort of complete without its technology stack, and that's true.
134
00:21:48.352 --> 00:21:57.732
Like, uh, in the same way military needs a technology stack to be effective, a studio needs tech to be effective. But it's not the main point of a studio.
135
00:21:57.772 --> 00:22:05.992
The main point of a studio is to build IP, leverage IP to make money, and entertain the masses along the way. I think like- Yeah...
136
00:22:06.172 --> 00:22:21.392
i- if he applied himself in that way where, uh, maybe not so much focused on picks and shovels, but really focused on figuring out a feature film with AI that is effective, and he certainly knows the creatives that could pull it off.
137
00:22:21.532 --> 00:22:30.532
He just has to shepherd them in a way that he's done before. Yeah, and to be able to make something that is Shrek quality, but not $200 million. Yeah.
138
00:22:30.572 --> 00:22:45.352
And I, I'll tell you, I'll tell you this one personal, uh, Jeffrey story, uh, that y- I think will highlight how his brain works a little bit, which could be fundamentally, um, challenging in this environment that is moving very fast and very rapidly.
139
00:22:45.511 --> 00:22:59.132
So there was a big town hall meeting at DreamWorks a long time ago, and, uh, Jeffrey had just, uh, moved out of his Bel Air home and then had his new custom-built, uh, Bel Air home in an, like, a higher part of Bel Air, a more baller home, if you will.
140
00:22:59.372 --> 00:23:10.392
And, uh, we, we sort of, uh, you know, we're like, "Yeah, go Jeffrey, go," you know? "Be a billionaire." [laughs] And he's like, "I moved into this new house." I'm paraphrasing here. This was a long time ago.
141
00:23:10.872 --> 00:23:25.072
He's like, "I moved into this new house and I have this incredible view of LA from my infinity pool or whatnot, and I'm just looking out into this view that I've never had before. And then, uh, there is a, uh,
142
00:23:26.152 --> 00:23:36.992
a concrete wall where my dog had pissed, and the piss stain is like into the concrete wall." And he said he couldn't enjoy the view 'cause he kept looking at that concrete piss stain.
143
00:23:37.012 --> 00:23:45.832
[laughs] So, like, that's how his mind works. He's a perfectionist. He's gonna try to do things to the highest degree and to the best of his ability.
144
00:23:45.972 --> 00:23:58.272
But in a, in a environment like what we have with generative AI today, that is detrimental. You can't fixate on little things. You just have to move past it. Um, but I, I think that'll hinder him in some capacity.
145
00:23:58.492 --> 00:24:11.212
Unless he builds the studio and that is where it's like, okay, we are all together as a team to build the film at the highest quality using the best tools available, which includes AI, versus I'm gonna build a platform tool where, like, anyone can make stuff that looks good.
146
00:24:11.812 --> 00:24:24.472
Yes. Agree. [laughs] But hey, uh, I, I'm just happy that he's championing for AI. I mean, that, that is a very influential person in Hollywood. Like, like it or not, he's pulling a lot of strings behind the scenes.
147
00:24:24.492 --> 00:24:32.592
He, you know, he seems to know the heads of every studio, all of the actors, and all the guilds. Like, this is a very powerful figure in this industry.
148
00:24:33.504 --> 00:24:45.664
Um, so if he says accept and embrace AI, then that, that does mean a lot to a lot of people. Yeah. Yeah. I mean, it, yeah, it helps with the changing of the acceptance of the tools. Changing of the guards.
149
00:24:46.084 --> 00:24:49.864
Change [laughs] I don't know what to say change of the guards, but change of the tools. Changing of the guards.
150
00:24:50.044 --> 00:24:56.764
What I really mean by that is, like, we're gonna effectively- The change of the tools, you adapt or you don't adapt. I mean, you gave examples in this essay too you could read of, of like- Yeah...
151
00:24:57.004 --> 00:25:06.304
when they shifted from, uh, hand-drawn animation to, to 3D. That's changing of the guards. A lot of people adapted and some people just didn't adapt and retired. Yeah.
152
00:25:06.544 --> 00:25:19.844
Um, I was, I was a little too young to see this transition, but when DreamWorks f-first started with Prince of Egypt and a couple other movies that was still, like, hand cel-shaded, uh, hand-drawn, like, Disney movies,
153
00:25:20.764 --> 00:25:25.704
and then, uh, both DreamWorks and Disney and Pixar were going through the transition into 3D.
154
00:25:26.064 --> 00:25:38.924
And so a lot of the 2D artists had really nice, uh, fine art skills that were just not applicable 'cause they had to pick up a new software, mouse, keyboard, GPU, expensive computers, and, uh, a lot of them just,
155
00:25:39.904 --> 00:25:47.964
just didn't want it. They're like, "3D is, is not animation. 3D is not it. We're, we're, we're doing things our way, and this is animation. This is the way."
156
00:25:48.484 --> 00:26:00.124
And lo and behold, like, 25, 26 years later, right, what is animation today? It's 3D mainly. All right, next up. Opus. Opus. 5.5. Claude is back.
157
00:26:00.504 --> 00:26:13.464
All right, so if you've been following Claude or if you're a Claude Opus user- Me... you may have noticed that lately the Opus releases have been a little bit weird and funky and, like, they just...
158
00:26:14.164 --> 00:26:22.444
The vibe was off, for lack of a better word. Opus 5 really had a lot of online dislike for it. Oh, really? Why is that?
159
00:26:22.504 --> 00:26:33.744
Personally, when I would interact with it, I just found its responses were always, like, really convoluted and kind of just, like, word vomit and, like, confusing to understand. I thought 4.8 was really good.
160
00:26:33.824 --> 00:26:36.104
Uh, people loved 4.6. I thought 4.8 was still pretty good.
161
00:26:36.424 --> 00:26:45.364
And then Anthropic was saying, "Oh, well, 5 is sort of a new thing, and you should, like, get rid of all of your previous, like, skills, documents," uh, 'cause it, like, understands stuff.
162
00:26:45.644 --> 00:26:53.604
But it's just everyone was just like, "It's just not really great." And then in the meantime, Fable came out. Fable's very powerful and good, very expensive.
163
00:26:54.064 --> 00:27:00.904
And then Astra started, you know, taking all the buzz for everything it could do with- Right... especially with video editing and 3D modeling.
164
00:27:01.364 --> 00:27:18.144
5.5 comes out, not such a powerful name, but I have found it's like, I mean, for the stuff I'm doing, which is, like, basic tool building and then also 3D stuff, 5.5 has been, like, at Fable level, h- bare fraction of the cost.
165
00:27:18.644 --> 00:27:32.344
Quality's back. Responses are back. Uh, everything about it feels really good, and it's like, 3D wise, on par or better than Astra. There you have it, folks. Joey endorsed Opus 5.5. That's it.
166
00:27:32.724 --> 00:27:41.804
Now it's great because now I have both models, so when I burn the usage of one, I just jump to the other one and then until the other one resets. So, uh- [laughs] Yep. Good, good, good strategy.
167
00:27:42.144 --> 00:27:51.484
Diversifying burn, burn rates. Um, uh, Opus 5.5 integrates to Blender the same way GPT-6 Astra does, right? It could, yeah.
168
00:27:51.544 --> 00:27:58.664
I mean, and, and, you know, we talked about before, like, fi- I mean, you've been able to integrate with Blender. What it can do with it is a different story. Right.
169
00:27:59.704 --> 00:28:18.684
But Opus 5.5, all right, like in a specific example, I was trying to, like, match some, some shots that were shot with like a 3D environment that I'm trying to build behind them, and Astra did an okay job, but there were some shots where it was like definitely the perspective and sizing of like between going from like wide shot to close up, it wasn't quite getting it correct.
170
00:28:19.364 --> 00:28:30.064
And Opus 5.5 reviewed the same shots and was like, "Oh, like the blender lens has been at the same focal length, like the entire, for every single shot. Like all the focal lengths need to change and adjust."
171
00:28:30.684 --> 00:28:47.624
And it like was able to correct all of the focal lengths for every shot and- Nice... export new background plates that were correct. So it caught that, that Astra was having trouble with. Wow. It's outsmarting Astra.
172
00:28:48.024 --> 00:28:59.844
So now the other thing is like Anthropic does not still, you know, there's no native image generation, but people have been sh- doing crazy stuff of having it generate visuals, um, either with like, uh, 3DJS or other tools.
173
00:29:00.344 --> 00:29:03.824
Someone, Ryan Sale, built a- Oh, this is amazing. I love that...
174
00:29:03.944 --> 00:29:17.704
uh, yeah, he asked Opus 5.5 to explain camera focus by building an interactive lens lab, and it built this very cool looking, um, interactive website with a 3D display showing focal plane and how focus works with a lens, and it looks very beautiful.
175
00:29:18.224 --> 00:29:27.864
Wow. Magnific did an example where they gave a brief to both Opus 5.5 and GPT-6, uh, GPT-6 Soul.
176
00:29:27.964 --> 00:29:38.464
So yeah, also GPT-6 came out with Soul and Terra variations of, uh, the 6 model, which they're saying is like for most stuff, probably just as good as Astra and also way cheaper. Yeah.
177
00:29:38.504 --> 00:29:48.044
I bet Jeffrey, Jeffrey's looking at this like, "Oh, shit, we need to build [laughs] we need to build a studio." [laughs] Yeah. This is also why I, uh, I don't think the tool route is the best route.
178
00:29:48.124 --> 00:29:57.024
I don't think the picks and shovel analogy works in today's age. Anybody can build a pick and shovel. Yeah. Everyone has their own foundry to forge whatever you want. Uh- Ooh, ooh, I like that, Joey...
179
00:29:57.044 --> 00:30:05.764
whatever tool you want. We're all blacksmiths. There you go. There we go. Yeah, we're all blacksmiths. [laughs] We've all got a, we've all got a smelting- Yeah... whatever freak.
180
00:30:06.404 --> 00:30:17.464
My knowledge of blacksmithing is a bit freaky. We've all got a... We can all smelt and forge whatever tool we want. You're absolutely right. Yeah. The shovels are cheap. The axes are cheap.
181
00:30:17.784 --> 00:30:23.964
It's, it's really about building the log, log cabin. Go, go do that however you can. Yeah. Right? Yeah.
182
00:30:24.484 --> 00:30:35.544
This is not the best test for a [laughs] podcast, but I feel like the Opus 5.5 example of whatever it built using CDance is- Feels more cinematic to me. Yeah. More shot design, more shot variety. 100%. Yeah.
183
00:30:36.092 --> 00:30:46.672
Yeah, like the, the shot of, um, the reflection on his sunglasses, I thought that was, that was a really nice touch. Mm-hmm. Um, e- some even ma- may call it a creative touch, but- Ooh.
184
00:30:46.672 --> 00:30:54.912
[laughs] Again, that's, that's up to you. And yeah, this is a collection of other visuals and stuff that Opus 5.5 built, but using JavaScript or other coding tools.
185
00:30:55.112 --> 00:31:05.552
Um, or, I mean, you could always give it access to image and video generation and let it steer it. Okay. You, you've inspired me to go use Opus 5.5. I, I am now going, gonna go off.
186
00:31:05.632 --> 00:31:15.832
Yeah, this is, um, the best Opus in a while. Opus is strong, and the usage is, like, pretty good. It's not burning. Like, I've been using it instead of...
187
00:31:15.872 --> 00:31:28.852
Like, I still have Fable usage left on my weekly allotment, which normally I would burn through by, like, Tuesday, but Opus 5.5 has been able to do the same stuff at the same quality that Fable was doing for me. Yeah.
188
00:31:28.892 --> 00:31:38.932
I mean, I'm not doing, like, insane, crazy stuff, so I'm sure there's still uses for Fable. So you think, like- Makes sense... Fable is like the Arri Alexa and Opus 5.5 is like the Blackmagic Ursa- [laughs]...
189
00:31:38.952 --> 00:31:54.072
where it's, like, 90% of that for 10% of the cost? Yeah, that's a good analogy. Yeah. [laughs] Sure. [laughs] Or Fable is, uh... Yeah, Fable's the, um, the, the Arri 65. Like [laughs] Oh, oh, it's like the king. Yeah.
190
00:31:54.072 --> 00:32:06.152
Yeah. It's, like, big, but it's like, okay, well, but I don't need- Yeah... to shoot 65 mil- Okay... for everything. Then GBT-6 Astra is a Sony Venice 2. Sh- Like, it's also equally- Equally as good... crazy. Right.
191
00:32:06.672 --> 00:32:18.612
Yeah. Right. Yeah. [laughs] Uh, sure. We'll go with this analogy. [laughs] So is GBT-6 Sol, like, a equivalent of Fable 5.5? What are we... What is the difference between Sol and Astra? No. Sol...
192
00:32:18.752 --> 00:32:27.992
Astra is, like, the top tier. Astra is, I think, I would say Astra is their f- is the Fable equivalent. That's publicly- Right... released.
193
00:32:28.032 --> 00:32:36.712
I think they have other, you know, more powerful secret models, but the- Yeah, like the ones that they're trying to shut down... publicly... Right, yeah. [laughs] Publicly available models- [laughs]...
194
00:32:36.792 --> 00:32:50.432
uh, Astra and Fable are the frontier top models from, from OpenAI and Anthropic. And then, uh, below that, so I would say, like, Sol is the Opus equivalent. Opus, yeah, okay.
195
00:32:50.572 --> 00:33:01.252
So Opus, maybe Opus 5, but they don't have, like, a middle-of-the-road product like 5.5? They have a... Yeah, middle is, um, Terra is the other- Terra... Terra. That's right. Sol and- Yeah... it's Sol, Terra, then Astra.
196
00:33:01.312 --> 00:33:08.572
So GBT-6 now has a Sol and Terra model variations. I haven't used them. I don't know... You know, they're cheaper.
197
00:33:09.072 --> 00:33:21.452
Uh, you know, so it's like, can you get away doing what you were doing with Astra with Sol and Terra because you're gonna save, you know, usage. I, I would love to know, like, how they value engineer these models, right?
198
00:33:21.492 --> 00:33:28.392
And the, like, the way things are done with, like, cars, right, is, um... And this is terrible practice.
199
00:33:29.052 --> 00:33:41.272
So, like, BMW will make one engine that makes 300 horsepower, and they'll put it on the 5 Series, and then they'll, like, control it with the electric chip to make less horsepower and put it on the model, the 3 Series.
200
00:33:41.552 --> 00:33:49.912
But it's the same engine. They're just, like, value engineering it- Yeah... so, like, it makes sense at that price point. Like, it's just easier to make a bunch of the same engines and just sort of- Exactly...
201
00:33:49.932 --> 00:33:53.132
sort of, like, gate it. [laughs] Throttle it. Yeah, gate, gate it, yeah, and throttle it.
202
00:33:53.232 --> 00:34:06.252
And, uh, so I, I don't know if these are all fundamentally different models, or it's just all running Fable but compromised, cheaper versions of Fable, just so, like, it incentivizes more, uh, frequent usage.
203
00:34:06.532 --> 00:34:19.012
So somebody like you and me, who are, like, token conscious are, are gonna go for 5.5, whereas the ballers, the Jeffries of the world, are gonna use the- [laughs]... the Opus, right? Yeah. I don't know- Or Fable...
204
00:34:19.072 --> 00:34:30.192
how the models are... Yeah, I don't know how the variations of the models are constructed. Um- Yeah... also am, I'm still fuzzy on what effort level makes the most sense for what task. That is still- Yeah.
205
00:34:30.612 --> 00:34:32.312
If, if our viewers know- Yeah...
206
00:34:32.352 --> 00:34:44.272
this stuff, like, please educa- this, again, uh, we're just throwing up our hands here, but we'd love to know how the different price tiers work out, uh, uh, in terms of how the model is actually built and inferred.
207
00:34:44.692 --> 00:34:54.092
Yeah, and what effort levels you use, and if you get good results, especially related to Blender and, and, uh, Resolve usage, 'cause that's the bulk of my use cases for these things.
208
00:34:54.552 --> 00:35:04.912
I usually crank it up, and then I'm like, "Ah, I'm probably gonna burn through my usage more," but I wanna give it the best. [laughs] True. [laughs] All right, last one. New image model, Qwen Image 2.1, open weight. Yes.
209
00:35:05.772 --> 00:35:13.652
I don't have any generations. Damn it. You didn't do your homework, Addy. All right. Didn't do my homework. So let me look for some stuff, but yeah. Uh, vibe of this has been
210
00:35:14.772 --> 00:35:23.712
hearing, like, Nana Banana quality, but it's open weight and way cheaper. Yeah, and, uh, you can run it on Comfy, obviously. Um, maybe we can look at some examples online. Yeah.
211
00:35:23.772 --> 00:35:33.652
I thought Qwen Image was at 3, but yeah, I don't know what... 2. There might've been another model that's at 3. Mm. I know they have a new- Okay... text-to-speech model as well. Yeah, they should name everything Qwen.
212
00:35:34.012 --> 00:35:42.932
Qwen, or as, uh, iMessage autocorrect likes to call it, Owen. Owen. [laughs] It's, it's... Well, let's see. We got transparent layers. We got text detail. Text, yeah.
213
00:35:43.412 --> 00:35:52.692
So it's really made for graphics and animatics and animation. Image editing abilities. This is an example of, uh, a tree in front of the Seattle Space Needle, and then it has- Mm...
214
00:35:52.752 --> 00:36:04.712
removed the, just leave the leaves, and it does it pretty good. Mm, that's not so hard if you extract a depth map. Okay, man. Yeah. [laughs] This is hard. This one's hard. So it did this well.
215
00:36:04.872 --> 00:36:11.872
Isn't that the cast from Cheers? This is Cheers, yeah. So it's a bunch of photos of the- Yeah... Cheers cast, and they put them all into a portrait. Up to 10 reference images. Yeah.
216
00:36:11.872 --> 00:36:19.232
Yeah, almost kind of hosting- But look, I mean, I think the big headline of this is like, yeah, look, is this gonna be better than anything you've seen? No. But this is open weight. You could run this locally. It's free.
217
00:36:19.412 --> 00:36:32.312
Yeah. It's open source. You can run basically Nana Banana on your own. Yeah, this looks like, everything here looks, like, on par with, like, Nana Banana 2 quality. Yeah. Again, do you think, conspiracy theory
218
00:36:33.352 --> 00:36:46.124
Addy is saying this, do you think they're doing this, Alibaba's doing this to kind of, um- Pull users away from Nana Banana usage? Possibly. Or, I mean, just to have a, for the- Just to weaken their market share...
219
00:36:46.184 --> 00:36:57.444
Chinese companies to have a competitive model. Yeah. Um, yeah. I mean, why... Yeah, why not if you could, if you could do it and release it and gain adoption? Right.
220
00:36:57.524 --> 00:37:04.484
I mean, we talked about this before, like really the advantage that the Chinese companies have is just putting stuff out open and getting adoption. Yeah.
221
00:37:04.884 --> 00:37:16.344
Uh, I mean, the Chinese models are okay, like the image models specifically. They're just trained on Chinese, uh, people data, which is, makes sense. It's Chinese model.
222
00:37:17.084 --> 00:37:26.784
Um, so they don't necessarily do the American diversity very well. [laughs] You know, like if you put Addy next to Joey and try to generate that, uh, it, it- I-... it'll struggle a little bit more than what GPT can do...
223
00:37:26.804 --> 00:37:33.764
think if you have reference images, yeah. But I think if you're doing text- Yeah... to image- Yeah... and you don't give it specifics, I remember this like a long time ago with some of your- Mm-hmm...
224
00:37:33.884 --> 00:37:42.244
like early Kling, and it's just like convenience store, and then it would just be a very Chinese-looking convenience store, um- Right. Exactly... or a cow.
225
00:37:42.284 --> 00:37:49.724
Like if you don't very clearly specify what you want, it'll default to more- Yeah... um, yeah, Chinese-looking environments.
226
00:37:49.764 --> 00:38:03.824
In my opinion, I think the, the king of image models right now is GPT 2.5 Sunburst and Nana Banana 2, still really, really good and, and much faster in inference, uh, as I'm sure Google is working on the next image model.
227
00:38:04.464 --> 00:38:13.544
And then on the video side, I mean- Well, we need a new, uh, we need a new Pro update 'cause I still feel like Nana Banana Pro is better than 2. Oh, you think so? Yeah. Like I'll still just u- default to 2.
228
00:38:13.664 --> 00:38:27.184
Uh, image, GPT Image 2.5 is very good and using that more, but I still think- Yeah... between that and Nana Banana Pro over Nana Banana 2. Yeah. I, I mean, I don't really use the Qwen models too much.
229
00:38:27.804 --> 00:38:44.204
They do have Qwen 2512, which is not Image 2.1. Like Qwen has multiple image models. That was really good when I last tested that about a few months ago. I remember that mouthful of a name. What was unique about it?
230
00:38:45.124 --> 00:38:58.324
Uh, so 2511 I think was text to image, and 2512 was, uh, multi-reference to image. I don't know why- Is this an extension of the same model and they just cleaned up the naming because that's very confusing?
231
00:38:58.424 --> 00:39:10.904
[laughs] Perhaps. Perhaps. Yeah. You're right. Yeah. I, I kind of preferred like the scientific names. It was easier to remember. Um- It sounds like a film stock. [laughs] Kodak tw- 7510... Kodak Qwen 2512. Yeah. Yeah.
232
00:39:11.304 --> 00:39:24.164
And then on the video side, I think, uh, most people agree that CDEN 2.5 is currently the king. Um, although my personal favorite is Wan 3.0. It's been solid for me. Okay. That's cool. Yeah. That's cool.
233
00:39:24.224 --> 00:39:41.144
Yeah, CDEN 2.5, good, just pricey, but it seems like they're putting more routes in that is like generate at a cheaper fast mode or a cheaper, uh, lower resolution mode, and then up res to what you want, but still
234
00:39:42.784 --> 00:39:48.984
be able to get the same output and not have like a new spin. So it seems like they've been refining that kind of workflow
235
00:39:50.104 --> 00:39:57.024
because it is so expensive to like do cheaper generations but then still adhere to like the vision that you want once you're ready to, to go. Yeah.
236
00:39:57.044 --> 00:40:04.504
They gotta capture the, the market that's cost conscious and still use their... It's the same argument that I made with Fable and Opus.
237
00:40:04.544 --> 00:40:18.224
Like you gotta have the CDEN Opus version of that model where, you know, we can mess around with it. Mm-hmm. Yeah. Built out this whole interface, this whole web mockup too. This is Qwen. Yeah. It looks very Google.
238
00:40:18.604 --> 00:40:30.124
Yeah. And- That's cool... detailed diagrams and stuff. Yeah, I mean, so e- everything it does is, is, is stuff you have seen before, but it, uh, you can run this locally. The power we have on our local GPUs. Sure.
239
00:40:30.204 --> 00:40:38.964
I heard something interesting, um, online, and I wanna hear your take on it. So- Okay... Adobe acquired Topaz, and we talked about it before.
240
00:40:39.384 --> 00:40:46.744
And one of the posts said that the reason they acquired Topaz, I, I mean, how, how could this person know, right?
241
00:40:46.784 --> 00:41:00.184
[laughs] Like they're not part of the deal, but I guess they're just speculating, is really good at running local tasks on GPUs. Mm-hmm. Like they have a downloadable, uh, application. I think it's called Astra, right?
242
00:41:00.963 --> 00:41:04.744
Where- Yeah. Topaz Video. I mean, that's, that's what they started on. That was what they were- Right. Right... they were the original. So it's like,
243
00:41:05.664 --> 00:41:16.784
uh, maybe Adobe acquired them for that capability or, you know, like sort of the local GPU mo- like the comfy UI model, if you will. What are your thoughts on that? I think it's the other way around.
244
00:41:16.844 --> 00:41:26.984
I mean, I don't really think who's gonna be running local. Yeah. And also, how does that fit into, no shade to Adobe, but out of all of the companies that love to charge- Yeah... subscription fees. Exactly.
245
00:41:27.104 --> 00:41:36.164
Like where's the business model if it's local, right? [laughs] I don't see Adobe offering, offering options to run stuff local. [laughs] I agree. I agree. I just wanted to hear your take on that. Yeah.
246
00:41:36.524 --> 00:41:45.484
Uh, a lot of people are still fixated on local. Look, I mean, the local stuff's great, and I, I mean, I use the Topaz local models as well. I mean, and, and i- it's been a hybrid.
247
00:41:45.524 --> 00:41:50.084
It's been, they have their generative upscalers and, and sometimes they're great, sometimes they work really well,
248
00:41:51.164 --> 00:42:00.524
and then sometimes they don't, depending on the footage and, and sometimes just a regular, their, uh, GAN models like Gaia and stuff just works better for- That's true... the task at hand.
249
00:42:01.104 --> 00:42:12.304
And, uh, did you hear- I, I'm hoping that Adobe still keeps the dedicated Topaz Video app, like as an option. I hope so. Yeah. Or maybe it's just another thing on Creative Cloud, right? Like the- Or Creative...
250
00:42:12.384 --> 00:42:22.244
Right, right. Yeah. It's included with your subscription. It's part of your subscription. Yeah, maybe. Yeah. I heard that NVIDIA is not really releasing any new s- consumer GPUs next year. Did you hear that?
251
00:42:22.584 --> 00:42:32.104
I did not hear that. Uh, I heard Jensen's interview on New York Times. Okay. Where- Where- Yeah. [laughs]... it was interesting... he's like, "Shut down the labs." No, he was li- no, the opposite.
252
00:42:32.124 --> 00:42:36.904
He was like, "It's, uh, it's all..." He's like, uh, he's like, "This is alarmist talk. It's not helping anyone.
253
00:42:37.124 --> 00:42:44.464
Like the labs should not release dangerous things, but saying it's gonna destroy the world is not helping anyone." [laughs] That's true. Um- Yeah... okay.
254
00:42:44.604 --> 00:42:53.684
So, uh, maybe that's a more recent Jensen interview where he's like now come around that position, so possibly. Oh, yeah, it just came out this week. Yeah, it was w- Yeah... out with, uh, Ezra and Klein- Yeah...
255
00:42:53.724 --> 00:42:58.704
on the New York Times. Yep. Yeah. It's a good interview. Fun, fun times. All right. Wrap it up here.
256
00:42:59.164 --> 00:43:06.024
If you've been watching the show or even listening to it on Apple Podcast, Spotify, or YouTube and have not subscribed yet, please do so.
257
00:43:06.464 --> 00:43:16.264
That'll be a really big help for us, and if you wanna hear some of the things that we have not yet touched on, comment on them and we'll see that. Links for everything we talked about at denoisepodcast.com.
258
00:43:16.764 --> 00:43:19.184
Thanks for watching, and we'll catch you in the next episode.
