part 7

0:00

S… Speaker 3 (part 7)

Let's discuss the following statements.

0:00

S… Speaker 3 (part 7)

Okay,

0:03

S… Speaker 2 (part 7)

so nothing much viva.

0:06

S… Speaker 2 (part 7)

What's happening is in terms of writing

0:11

S… Speaker 2 (part 7)

and post -production.

0:17

S… Speaker 2 (part 7)

What's happening with video generation models?

0:20

S… Speaker 2 (part 7)

See, there is a lot of different video generation models.

0:24

S… Speaker 2 (part 7)

There is Seed

0:28

S… Speaker 2 (part 7)

Dance and lots of other things.

0:30

S… Speaker 2 (part 7)

And there are these common places where you can integrate all these

0:34

S… Speaker 2 (part 7)

video models also.

0:35

S… Speaker 1 (part 7)

Correct.

0:36

S… Speaker 2 (part 7)

It's all about control.

0:40

S… Speaker 2 (part 7)

and control and achieve and I think

0:44

S… Speaker 2 (part 7)

then it will disrupt something.

0:46

S… Speaker 2 (part 7)

We are seeing inconsistencies with the location,

0:50

S… Speaker 2 (part 7)

character and they are basically definitely becoming

0:54

S… Speaker 2 (part 7)

better and better.

0:58

S… Speaker 2 (part 7)

They're becoming better and better.

1:00

S… Speaker 2 (part 7)

But having said that,

1:02

S… Speaker 2 (part 7)

these generative models,

1:04

S… Speaker 2 (part 7)

I don't know if it is even aligned to that perspective.

1:06

S… Speaker 2 (part 7)

That is

1:10

S… Speaker 2 (part 7)

a lot of,

1:14

S… Speaker 2 (part 7)

that's a question that's being targeted by all these video

1:18

S… Speaker 2 (part 7)

generation models.

1:20

S… Speaker 2 (part 7)

They are making a group of agents

1:24

S… Speaker 2 (part 7)

also doing their own custom pipelines.

1:26

S… Speaker 2 (part 7)

Custom pipelines,

1:28

S… Speaker 2 (part 7)

again,

1:29

S… Speaker 2 (part 7)

it is operating from the cloud.

1:32

S… Speaker 2 (part 7)

It's not operating from local servers.

1:38

S… Speaker 2 (part 7)

These outputs,

1:39

S… Speaker 2 (part 7)

video generation models,

1:41

S… Speaker 2 (part 7)

they are coming out in H .264 containers which are impact compressions.

1:45

S… Speaker 2 (part 7)

So, they are 10 bit rate,

1:47

S… Speaker 2 (part 7)

10 Mbps or 12 Mbps bit rates.

1:49

S… Speaker 2 (part 7)

My theatrical projection means you have a sequential

1:53

S… Speaker 2 (part 7)

DNG format.

2:03

S… Speaker 2 (part 7)

We run into a lot of data,

2:05

S… Speaker 2 (part 7)

13 stops of latitude in the cinema technical language.

2:09

S… Speaker 2 (part 7)

Now image generation models are doing that.

2:13

S… Speaker 2 (part 7)

Image scaling models are there.

2:17

S… Speaker 2 (part 7)

Can we build a pipeline where we can take these H .264

2:22

S… Speaker 2 (part 7)

videos,

2:24

S… Speaker 2 (part 7)

break them down into image sequences,

2:27

S… Speaker 3 (part 7)

PNGs,

2:28

S… Speaker 2 (part 7)

and the image upscaling,

2:30

S… Speaker 2 (part 7)

DNG sequence.

2:31

S… Speaker 2 (part 7)

I think it will be like a batch processing.

2:34

S… Speaker 2 (part 7)

I think it's going to be really,

2:37

S… Speaker 2 (part 7)

really disruptive for the film industry.

2:39

S… Speaker 1 (part 7)

I don't think we have enough GPUs to be able to pull this off.

2:42

S… Speaker 3 (part 7)

That's the problem.

2:45

S… Speaker 1 (part 7)

That is probably the problem.

2:47

S… Speaker 1 (part 7)

But I think first of all,

2:49

S… Speaker 1 (part 7)

Tarun,

2:49

S… Speaker 1 (part 7)

huge fan.

2:50

S… Speaker 1 (part 7)

Thanks,

2:51

S… Speaker 1 (part 7)

man. If I have to share a personal story,

2:54

S… Speaker 1 (part 7)

you know,

2:55

S… Speaker 1 (part 7)

Kida Cola has become that movie where after we get drunk,

2:58

S… Speaker 1 (part 7)

if we have to show someone a new Telugu movie,

3:00

S… Speaker 1 (part 7)

that's the movie.

3:01

S… Speaker 3 (part 7)

Right?

3:01

S… Speaker 1 (part 7)

So it's,

3:03

S… Speaker 1 (part 7)

I mean,

3:03

S… Speaker 1 (part 7)

my girlfriend has made me watch that movie like 10 times.

3:05

S… Speaker 1 (part 7)

But she'll be very happy to know that

3:10

S… Speaker 1 (part 7)

we caught up here.

3:14

S… Speaker 1 (part 7)

Chandana.

3:14

S… Speaker 1 (part 7)

But I think one is that I'm pleasantly surprised bro

3:18

S… Speaker 1 (part 7)

with the conversation that we just had.

3:20

S… Speaker 4 (part 7)

Right.

3:21

S… Speaker 1 (part 7)

I don't know if this is the case across,

3:23

S… Speaker 1 (part 7)

but I don't expect it.

3:26

S… Speaker 1 (part 7)

that you will be talking about content pipelines,

3:28

S… Speaker 1 (part 7)

you'll be talking about agent orchestration.

3:30

S… Speaker 1 (part 7)

So you put me in a spot.

3:34

S… Speaker 1 (part 7)

But sadly,

3:35

S… Speaker 1 (part 7)

with Seed Dance 2 .0,

3:37

S… Speaker 1 (part 7)

the one that we saw,

3:38

S… Speaker 1 (part 7)

right? That just came out.

3:40

S… Speaker 1 (part 7)

The one that is out for us is actually a distilled version of the product.

3:44

S… Speaker 1 (part 7)

The full version of the product is not out yet.

3:46

S… Speaker 1 (part 7)

Because one of the issues that ByteDance found is

3:51

S… Speaker 1 (part 7)

when they tried to convert an image into a video.

3:54

S… Speaker 1 (part 7)

They heard the voice of that same person that

3:58

S… Speaker 1 (part 7)

they didn't feed.

3:59

S… Speaker 1 (part 7)

Based on the data set.

4:01

S… Speaker 1 (part 7)

So it's like Chiranjeev got a photo of you and a version

4:07

S… Speaker 1 (part 7)

of that came out.

4:08

S… Speaker 3 (part 7)

That is insane man.

4:10

S… Speaker 1 (part 7)

As a result they thought this could be a big problem.

4:14

S… Speaker 1 (part 7)

I had to launch it around a couple of

4:18

S… Speaker 1 (part 7)

months back.

4:19

S… Speaker 1 (part 7)

Now they're slowly launching out,

4:21

S… Speaker 1 (part 7)

which is a heavy restricted,

4:23

S… Speaker 1 (part 7)

heavy guardrail restricted version.

4:26

S… Speaker 1 (part 7)

So,

4:27

S… Speaker 1 (part 7)

but with someone of your access,

4:30

S… Speaker 1 (part 7)

I think if I was you,

4:31

S… Speaker 1 (part 7)

I would not use a regular one.

4:34

S… Speaker 1 (part 7)

I would try to get in touch with ByteDance to get direct access

4:38

S… Speaker 1 (part 7)

of their no guardrail API.

4:41

S… Speaker 1 (part 7)

Right, which I again without any guardrails guardrails in Lake

4:45

S… Speaker 1 (part 7)

Kuna API without no restriction because they will so for

4:49

S… Speaker 1 (part 7)

example, there are companies like Heygen,

4:51

S… Speaker 4 (part 7)

right?

4:52

S… Speaker 1 (part 7)

Heygen,

4:53

S… Speaker 1 (part 7)

they got integration with SeedDance.

4:55

S… Speaker 1 (part 7)

They don't allow audio SeedDance doesn't

5:00

S… Speaker 1 (part 7)

Allow external audio that easily.

5:02

S… Speaker 1 (part 7)

But with HeyGen they allow because they follow two -step verification.

5:06

S… Speaker 1 (part 7)

So they don't want people to exploit it.

5:08

S… Speaker 1 (part 7)

You know what I mean?

5:09

S… Speaker 1 (part 7)

Hence,

5:10

S… Speaker 2 (part 7)

they're doing this.

5:11

S… Speaker 1 (part 7)

But there has to be a way that we should be able to get in touch directly with them.

5:14

S… Speaker 1 (part 7)

I'll also try to find a contact if I can because that is an unlock.

5:19

S… Speaker 1 (part 7)

The second unlock is consistency,

5:21

S… Speaker 1 (part 7)

character consistency and motion control and everything.

5:23

S… Speaker 1 (part 7)

It's already being solved.

5:24

S… Speaker 1 (part 7)

I think it is the biggest problem that all these videos and models are

5:28

S… Speaker 1 (part 7)

trying to solve.

5:29

S… Speaker 1 (part 7)

I think we are probably six months away from that.

5:31

S… Speaker 2 (part 7)

That's crazy.

5:32

S… Speaker 2 (part 7)

That's what I thought because conversations start and there

5:37

S… Speaker 2 (part 7)

are so many updates that it's a day -to -day update that we are looking at.

5:41

S… Speaker 2 (part 7)

No longer,

5:42

S… Speaker 2 (part 7)

the three -month -back conversation is outdated and every day the

5:46

S… Speaker 2 (part 7)

news is changing rapidly.

5:48

S… Speaker 1 (part 7)

But there are some ways to scale up.

5:52

S… Speaker 1 (part 7)

Videos also,

5:53

S… Speaker 1 (part 7)

I don't know if it will fit in the big screen,

5:56

S… Speaker 1 (part 7)

which is the theater screens until the level scale up.

5:59

S… Speaker 1 (part 7)

Because even on a video resolution level,

6:03

S… Speaker 1 (part 7)

best AI models are capped at 720p or 1080p.

6:06

S… Speaker 1 (part 7)

We are not at 4k yet.

6:07

S… Speaker 1 (part 7)

Not because they are not capable.

6:10

S… Speaker 1 (part 7)

It's purely because of GPUs.

6:12

S… Speaker 1 (part 7)

Cost.

6:13

S… Speaker 2 (part 7)

Yeah, the processing that's required.

6:15

S… Speaker 1 (part 7)

So there are models like LTX.

6:18

S… Speaker 1 (part 7)

which are open source.

6:19

S… Speaker 1 (part 7)

So,

6:20

S… Speaker 1 (part 7)

we can

6:25

S… Speaker 1 (part 7)

actually get 4k outputs directly.

6:27

S… Speaker 1 (part 7)

So,

6:29

S… Speaker 1 (part 7)

it is possible,

6:30

S… Speaker 1 (part 7)

but tech has to coincide to that point.

6:34

S… Speaker 1 (part 7)

And also access has to coincide to that point.

6:36

S… Speaker 1 (part 7)

But I think we are six months away.

6:38

S… Speaker 1 (part 7)

And I believe now that we have had this conversation,

6:40

S… Speaker 1 (part 7)

if someone is picking this up,

6:42

S… Speaker 1 (part 7)

it's going to be probably someone like you.

6:43

S… Speaker 1 (part 7)

So I'm looking forward.

6:45

S… Speaker 1 (part 7)

No,

6:46

S… Speaker 2 (part 7)

so that's what,

6:47

S… Speaker 2 (part 7)

there was one conversation with my ADs.

6:49

S… Speaker 2 (part 7)

And look,

6:50

S… Speaker 2 (part 7)

image scaling models,

6:51

S… Speaker 2 (part 7)

it's not only generally

6:55

S… Speaker 2 (part 7)

we get carried away by resolution,

6:57

S… Speaker 2 (part 7)

but theatrical projection is basically

7:01

S… Speaker 1 (part 7)

all about

7:02

S… Speaker 2 (part 7)

bitrate and data for that particular image.

7:06

S… Speaker 2 (part 7)

For example,

7:08

S… Speaker 2 (part 7)

you can have a 4K image,

7:10

S… Speaker 2 (part 7)

but it can be as low as 5MB.

7:12

S… Speaker 2 (part 7)

And the color information and the gradient information that that image

7:17

S… Speaker 2 (part 7)

has, millions of colors fill up.

7:19

S… Speaker 2 (part 7)

That's why we use expensive cameras like ARRI and RED because,

7:23

S… Speaker 2 (part 7)

you know, they have a lot of color information.

7:25

S… Speaker 2 (part 7)

Why,

7:26

S… Speaker 1 (part 7)

I don't know,

7:27

S… Speaker 2 (part 7)

it's a theatrical projection.

7:28

S… Speaker 2 (part 7)

It's all Dolby,

7:29

S… Speaker 2 (part 7)

Dolby Vision.

7:32

S… Speaker 2 (part 7)

You can literally see millions of colors.

7:35

S… Speaker 2 (part 7)

I mean, it's a phenomenal experience to watch theatrically.

7:39

S… Speaker 2 (part 7)

So I think the only way for theatres to survive is to provide that

7:43

S… Speaker 2 (part 7)

much of information for the common audience so that it's

7:47

S… Speaker 2 (part 7)

a visual experience that you cannot do on your digital screens.

7:51

S… Speaker 2 (part 7)

But I think we have a long way to go before we

7:55

S… Speaker 2 (part 7)

can process that amount of data through AI.

7:59

S… Speaker 2 (part 7)

The image upscaling models,

8:01

S… Speaker 2 (part 7)

they are doing it.

8:02

S… Speaker 2 (part 7)

Offlight,

8:04

S… Speaker 2 (part 7)

we are also seeing on Instagram,

8:05

S… Speaker 2 (part 7)

where they are fine detailing of old pictures.

8:08

S… Speaker 2 (part 7)

And you know,

8:09

S… Speaker 2 (part 7)

detailed textures generated,

8:11

S… Speaker 2 (part 7)

plus it's upscaling that image.

8:14

S… Speaker 2 (part 7)

The fact that it's upscaling an image,

8:17

S… Speaker 2 (part 7)

it's just 24 images a second.

8:20

S… Speaker 2 (part 7)

It's just like a few,

8:23

S… Speaker 2 (part 7)

as you said,

8:24

S… Speaker 2 (part 7)

hopefully a few months away.

8:25

S… Speaker 2 (part 7)

But my other concern is that all this is

8:29

S… Speaker 2 (part 7)

being done so that the average consumer can generate films or

8:33

S… Speaker 2 (part 7)

ideas at the click of a button.

8:35

S… Speaker 2 (part 7)

I still seem that it's far -fetched because of the processing power

8:39

S… Speaker 1 (part 7)

it takes.

8:39

S… Speaker 2 (part 7)

I think it's going to cost a bit.

8:43

S… Speaker 2 (part 7)

In the end,

8:44

S… Speaker 2 (part 7)

Rajivisar clearly made out a very good statement.

8:47

S… Speaker 2 (part 7)

He said, it's easy to make films like Bahubali,

8:51

S… Speaker 2 (part 7)

but it does not make sense to make a small film.

8:54

S… Speaker 1 (part 7)

That is true.

8:54

S… Speaker 2 (part 7)

So why would

9:01

S… Speaker 2 (part 7)

you go ahead and make a small film?

9:04

S… Speaker 2 (part 7)

So, okay,

9:05

S… Speaker 2 (part 7)

then the small filmmakers are still safe,

9:07

S… Speaker 2 (part 7)

I feel, but there's so many possibilities opening up.

9:10

S… Speaker 2 (part 7)

I'm really excited.

9:11

S… Speaker 1 (part 7)

Yeah.

9:12

S… Speaker 1 (part 7)

It's interesting.

9:13

S… Speaker 1 (part 7)

It's interesting.

9:16

S… Speaker 1 (part 7)

So, that is what I was trying to address.

9:19

S… Speaker 1 (part 7)

So, I was catching up with...

9:22

S… Speaker 1 (part 7)

Oh, you have time, no?

9:22

S… Speaker 2 (part 7)

Yeah,

9:23

S… Speaker 1 (part 7)

yeah, yeah.

9:24

S… Speaker 1 (part 7)

Go for it.

9:26

S… Speaker 1 (part 7)

I was catching up with Alex,

9:28

S… Speaker 1 (part 7)

who is the head of Meta AI.

9:30

S… Speaker 1 (part 7)

Facebook is the head of AI.

9:35

S… Speaker 1 (part 7)

It was the same conversation.

9:39

S… Speaker 1 (part 7)

The biggest bottleneck is at this point of time,

9:43

S… Speaker 1 (part 7)

cost of inference.

9:44

S… Speaker 1 (part 7)

Cost of processing and everything.

9:47

S… Speaker 1 (part 7)

Is the future actually local?

9:51

S… Speaker 1 (part 7)

All right,

9:52

S… Speaker 1 (part 7)

so you put a models are converging to a very

9:56

S… Speaker 1 (part 7)

good optimization so that at lower

10:00

S… Speaker 1 (part 7)

a memory you can do a lot more work right large

10:04

S… Speaker 1 (part 7)

models are becoming smaller and giving you the same kind of output so distillation

10:08

S… Speaker 1 (part 7)

of process only that is getting very very good yesterday I showed you the example of Gemma

10:12

S… Speaker 1 (part 7)

4 right and the same is kind of happening with video as well so

10:17

S… Speaker 1 (part 7)

first text

10:20

S… Speaker 1 (part 7)

Now the audio is almost happening.

10:23

S… Speaker 1 (part 7)

What we were expecting a year back.

10:25

S… Speaker 1 (part 7)

Now it has to happen with images and videos.

10:28

S… Speaker 1 (part 7)

Images also started.

10:29

S… Speaker 1 (part 7)

Now the next is video.

10:31

S… Speaker 1 (part 7)

With LTX Studio Japan.

10:33

S… Speaker 1 (part 7)

LTX is the first glimpse of that.

10:35

S… Speaker 1 (part 7)

Where you can actually generate a video out of a Mac.

10:38

S… Speaker 1 (part 7)

It's not great.

10:40

S… Speaker 1 (part 7)

But it's the footage that looks how it used to look with a Sora

10:44

S… Speaker 1 (part 7)

one.

10:45

S… Speaker 1 (part 7)

The newest model.

10:48

S… Speaker 1 (part 7)

Now, is it as good as Seed Dance?

10:49

S… Speaker 1 (part 7)

No. But maybe a year later,

10:51

S… Speaker 1 (part 7)

it'll be as good as Seed Dance.

10:53

S… Speaker 1 (part 7)

So, future,

10:53

S… Speaker 1 (part 7)

I think everybody will be having racks.

10:57

S… Speaker 1 (part 7)

That will be part of the cost of production.

11:00

S… Speaker 1 (part 7)

Or maybe as you pay money to actors and directors and

11:04

S… Speaker 1 (part 7)

producers, you rent out a rack as well as a part of the cost of production

11:08

S… Speaker 1 (part 7)

for the movie.

11:08

S… Speaker 1 (part 7)

One last

11:13

S… Speaker 1 (part 7)

question. What's happening in the industry with respect to AI adaption?

11:17

S… Speaker 1 (part 7)

What do you have to say?

11:19

S… Speaker 1 (part 7)

See, that's what,

11:20

S… Speaker 2 (part 7)

artificial intelligence in some way or form has always been a

11:24

S… Speaker 1 (part 7)

part of the film industry.

11:25

S… Speaker 2 (part 7)

It has been called different names.

11:27

S… Speaker 2 (part 7)

Photoshop,

11:28

S… Speaker 2 (part 7)

magic wand feature kind of computing.

11:31

S… Speaker 1 (part 7)

That's true.

11:32

S… Speaker 2 (part 7)

See,

11:34

S… Speaker 2 (part 7)

initially they used to,

11:36

S… Speaker 2 (part 7)

business and technology always converge

11:41

S… Speaker 2 (part 7)

and create compartmentalization.

11:44

S… Speaker 2 (part 7)

smart and green trend.

11:46

S… Speaker 2 (part 7)

Tech was,

11:49

S… Speaker 2 (part 7)

I mean, if we just cut it down to the basics,

11:51

S… Speaker 2 (part 7)

we've always had technology to assist us.

11:54

S… Speaker 2 (part 7)

And filmmaking has always been innovative and,

11:58

S… Speaker 2 (part 7)

you know, it's always been the edge of that.

12:00

S… Speaker 2 (part 7)

Because,

12:02

S… Speaker 2 (part 7)

you know, we're constantly working with VFX.

12:05

S… Speaker 2 (part 7)

What we deal with is video gravity.

12:09

S… Speaker 2 (part 7)

It's always been constantly evolving.

12:11

S… Speaker 2 (part 7)

Post -production,

12:12

S… Speaker 2 (part 7)

DI,

12:13

S… Speaker 2 (part 7)

obviously production

12:17

S… Speaker 2 (part 7)

intensive.

12:18

S… Speaker 2 (part 7)

It's always been around.

12:21

S… Speaker 2 (part 7)

For sure VFX,

12:24

S… Speaker 2 (part 7)

even in promotions,

12:28

S… Speaker 2 (part 7)

we had Trump.

12:28

S… Speaker 1 (part 7)

Oh yeah.

12:29

S… Speaker 2 (part 7)

And I was,

12:32

S… Speaker 2 (part 7)

I was plainly surprised.

12:34

S… Speaker 2 (part 7)

That is happening.

12:36

S… Speaker 2 (part 7)

And VFX,

12:38

S… Speaker 2 (part 7)

a lot of work has reduced.

12:42

S… Speaker 2 (part 7)

I mean, Rotoscope artist has a credit role.

12:46

S… Speaker 2 (part 7)

And that is going to reduce

12:50

S… Speaker 2 (part 7)

significantly.

12:51

S… Speaker 2 (part 7)

And a lot of work is going to be simplified for sure.

12:56

S… Speaker 1 (part 7)

Access.

12:57

S… Speaker 2 (part 7)

I don't think it's going to replace filmmaking altogether

13:01

S… Speaker 2 (part 7)

because definitely there are going to be a lot of constraints

13:05

S… Speaker 2 (part 7)

but it has to work in tandem.

13:07

S… Speaker 2 (part 7)

Definitely it's going to work in tandem.

13:09

S… Speaker 2 (part 7)

and some lower -rung jobs are in for a toss.

13:13

S… Speaker 2 (part 7)

And

13:25

S… Speaker 2 (part 7)

it's a part of evolution.

13:26

S… Speaker 2 (part 7)

I don't think I don't want to sit down

13:30

S… Speaker 2 (part 7)

on the road, because it's a part of

13:34

S… Speaker 2 (part 7)

evolution and you know it's going to happen.

13:36

S… Speaker 2 (part 7)

If you want the wheel to go,

13:38

S… Speaker 2 (part 7)

then the square has to become,

13:39

S… Speaker 2 (part 7)

gonna cut off his edges,

13:42

S… Speaker 1 (part 7)

girl. Well said.

13:43

S… Speaker 1 (part 7)

Crazy,

13:44

S… Speaker 1 (part 7)

crazy.

13:44

S… Speaker 1 (part 7)

So, Manadepu next?

13:46

S… Speaker 1 (part 7)

Yeah,

13:48

S… Speaker 2 (part 7)

I think it's going to happen.

13:52

S… Speaker 2 (part 7)

You're talking about E &E too,

13:53

S… Speaker 1 (part 7)

no? No, no.

13:55

S… Speaker 2 (part 7)

I am in Thailand right now.

13:57

S… Speaker 2 (part 7)

Scorching heat is coming.

13:58

S… Speaker 2 (part 7)

I wish if you are working with me,

14:01

S… Speaker 2 (part 7)

you will be able to get a photo of me.

14:04

S… Speaker 1 (part 7)

Do you want to talk about that?

14:07

S… Speaker 1 (part 7)

In our podcast,

14:10

S… Speaker 1 (part 7)

the first question is ENU2.

14:20

S… Speaker 2 (part 7)

I'm sure,

14:21

S… Speaker 2 (part 7)

I'm sure.

14:25

S… Speaker 1 (part 7)

We'll see you in the episode.

14:27

S… Speaker 2 (part 7)

Yes, why not?

14:28

S… Speaker 2 (part 7)

100%.

14:29

S… Speaker 2 (part 7)

It's always lovely speaking to you guys.

14:31

S… Speaker 1 (part 7)

Thank you very much.

14:32

S… Speaker 1 (part 7)

Means a lot.

14:32

S… Speaker 1 (part 7)

Thanks for the time also.

14:33

S… Speaker 1 (part 7)

Thank you,

14:35

S… Speaker 2 (part 7)

man. Thank you for this.

14:36

S… Speaker 2 (part 7)

Really looking forward to the future.

14:39

S… Speaker 2 (part 7)

Exciting and I'm glad I could speak to you.

14:41

S… Speaker 2 (part 7)

Vaibhav, Dilip and all of you guys.

14:44

S… Speaker 2 (part 7)

Thank you for having me.

14:45

S… Speaker 1 (part 7)

Thank you.

14:46

S… Speaker 1 (part 7)

Thanks a lot.

14:46

S… Speaker 1 (part 7)

Bye, man.

14:47

S… Speaker 1 (part 7)

Bye,

14:48

S… Speaker 1 (part 7)

Mamshi.

14:48

S… Speaker 1 (part 7)

See you, man.

14:49

S… Speaker 1 (part 7)

Bye.

14:50

S… Speaker 1 (part 7)

Insane. Wow, this was impressive.

15:00

S… Speaker 1 (part 7)

Someone who doesn't like or doesn't know about AI or he's not what

15:04

S… Speaker 1 (part 7)

you call adapting AI,

15:05

S… Speaker 1 (part 7)

if he's watching this content,

15:06

S… Speaker 1 (part 7)

what kind of level of thought process are you?

15:10

S… Speaker 1 (part 7)

In fact, how do you think about it?

15:11

S… Speaker 1 (part 7)

In your notes,

15:12

S… Speaker 1 (part 7)

in your remarkable document,

15:15

S… Speaker 1 (part 7)

30 % are you?

15:17

S… Speaker 1 (part 7)

But still,

15:18

S… Speaker 1 (part 7)

we are, we were so at it,

15:19

S… Speaker 1 (part 7)

and the content,

15:21

S… Speaker 1 (part 7)

I think this is the next episode.

15:22

S… Speaker 1 (part 7)

Crazy,

15:25

S… Speaker 1 (part 7)

isn't it?

15:27

S… Speaker 1 (part 7)

One industry or one of these,

15:30

S… Speaker 1 (part 7)

what different exposure do you have?

15:31

S… Speaker 1 (part 7)

True.

15:31

S… Speaker 2 (part 7)

Actually,

15:33

S… Speaker 2 (part 7)

Ben Affleck is not.

15:35

S… Speaker 1 (part 7)

Okay.

15:36

S… Speaker 1 (part 7)

It's good.

15:38

S… Speaker 2 (part 7)

There's a lot of time in Hollywood.

15:40

S… Speaker 2 (part 7)

In Hollywood,

15:50

S… Speaker 2 (part 7)

there are actors who have very good AI consciousness.

15:53

S… Speaker 2 (part 7)

There's awareness.

15:55

S… Speaker 2 (part 7)

In fact,

15:56

S… Speaker 2 (part 7)

Ben Affleck is an AI company.

15:57

S… Speaker 2 (part 7)

That's why it's a Netflix company.

16:03

S… Speaker 2 (part 7)

For me, I want to explain AI to normal people.

16:07

S… Speaker 2 (part 7)

First,

16:09

S… Speaker 2 (part 7)

I want to explain AI to Ben Affleck.

16:10

S… Speaker 2 (part 7)

So,

16:13

S… Speaker 2 (part 7)

what I want to say is,

16:13

S… Speaker 2 (part 7)

AI is the mean,

16:15

S… Speaker 2 (part 7)

the mean training.

16:16

S… Speaker 2 (part 7)

In AI,

16:18

S… Speaker 2 (part 7)

there will be 10 % best material,

16:21

S… Speaker 2 (part 7)

10 % worst material,

16:23

S… Speaker 2 (part 7)

80 % average.

16:25

S… Speaker 2 (part 7)

So,

16:26

S… Speaker 2 (part 7)

AI will be average.

16:30

S… Speaker 2 (part 7)

So, you know that you put a point in the 10 % point.

16:34

S… Speaker 2 (part 7)

That's a prompting.

16:36

S… Speaker 1 (part 7)

So,

16:37

S… Speaker 1 (part 7)

anything about this query?

16:40

S… Speaker 2 (part 7)

Actually upscaling is a lot of different problems.

16:44

S… Speaker 1 (part 7)

When he spoke about 24 frames per

16:48

S… Speaker 1 (part 7)

second, if we scale it up,

16:49

S… Speaker 1 (part 7)

we can put together to make it into a video.

16:51

S… Speaker 1 (part 7)

Technically,

16:52

S… Speaker 1 (part 7)

he is absolutely bang on.

16:53

S… Speaker 1 (part 7)

Upscaling is different.

16:55

S… Speaker 1 (part 7)

When we do,

16:56

S… Speaker 1 (part 7)

there are tools like Topaz,

16:57

S… Speaker 1 (part 7)

like you mentioned.

16:58

S… Speaker 1 (part 7)

The upscaling happens in the same way.

17:01

S… Speaker 1 (part 7)

Topaz is the one that our editors use.

17:09

S… Speaker 1 (part 7)

So, it's for video editing and now it has become

17:14

S… Speaker 1 (part 7)

AI first.

17:14

S… Speaker 1 (part 7)

We can probably show towards the end.

17:17

S… Speaker 1 (part 7)

Upskilling basically.

17:19

S… Speaker 1 (part 7)

So,

17:20

S… Speaker 1 (part 7)

it also does in the same way but what he did is very interesting.

17:24

S… Speaker 1 (part 7)

He told people

17:27

S… Speaker 1 (part 7)

Even if you know your art very well,

17:30

S… Speaker 1 (part 7)

you can apply AI there,

17:31

S… Speaker 1 (part 7)

learn AI and do it.

17:33

S… Speaker 1 (part 7)

Because the moment he said that upscaling and how it is 24

17:37

S… Speaker 1 (part 7)

frames per second,

17:38

S… Speaker 1 (part 7)

I can

17:42

S… Speaker 1 (part 7)

think of we can use Comfy UI.

17:45

S… Speaker 1 (part 7)

We can use Comfy UI and its image workflow.

17:48

S… Speaker 1 (part 7)

Basically he

17:52

S… Speaker 1 (part 7)

means, every second is divided into 24 frames

17:57

S… Speaker 1 (part 7)

per second.

17:59

S… Speaker 1 (part 7)

24 images,

17:59

S… Speaker 1 (part 7)

24 images put together just a

18:03

S… Speaker 1 (part 7)

1 second video.

18:06

S… Speaker 1 (part 7)

I want to use the video upscale and GPUs.

18:10

S… Speaker 1 (part 7)

I got it, okay.

18:11

S… Speaker 1 (part 7)

So,

18:11

S… Speaker 1 (part 7)

it's not an image.

18:12

S… Speaker 1 (part 7)

Because the image is upscale easy.

18:15

S… Speaker 1 (part 7)

It's not upscale.

18:17

S… Speaker 1 (part 7)

All the images are put together in the software.

18:20

S… Speaker 1 (part 7)

High quality video.

18:22

S… Speaker 1 (part 7)

High quality video.

18:23

S… Speaker 1 (part 7)

That is what Topaz Lab does.

18:25

S… Speaker 1 (part 7)

Right?

18:26

S… Speaker 1 (part 7)

But the interesting thing is Topaz Lab is probably not catering

18:31

S… Speaker 1 (part 7)

to his requirement of theoretical.

18:35

S… Speaker 1 (part 7)

I can imagine that you can build a comfy UI workflow for this.

18:39

S… Speaker 1 (part 7)

Comfy UI workflow is exactly this.

18:40

S… Speaker 1 (part 7)

Do you want to explain?

18:41

S… Speaker 2 (part 7)

So Comfy UI...

18:42

S… Speaker 1 (part 7)

Can we all just see?

18:43

S… Speaker 1 (part 7)

It's an open -source thing.

18:45

S… Speaker 1 (part 7)

So you have to look for Comfy UI.

18:46

S… Speaker 2 (part 7)

Comfy UI is open -source.

18:47

S… Speaker 2 (part 7)

In other words,

18:48

S… Speaker 2 (part 7)

if we use our local machine,

18:51

S… Speaker 2 (part 7)

it's heavy.

18:52

S… Speaker 2 (part 7)

So the intent

18:57

S… Speaker 2 (part 7)

is to load a cloud GPU server in the data.

19:00

S… Speaker 2 (part 7)

So the intent is to load a whole.

19:07

S… Speaker 2 (part 7)

Six months back.

19:08

S… Speaker 1 (part 7)

Six months back.

19:09

S… Speaker 2 (part 7)

Six months back.

19:10

S… Speaker 1 (part 7)

Six months back.

19:10

S… Speaker 1 (part 7)

Six months back.

19:12

S… Speaker 1 (part 7)

Six months back.

19:13

S… Speaker 2 (part 7)

Six months back.

19:14

S… Speaker 2 (part 7)

Six months back. Six months

19:24

S… Speaker 2 (part 7)

back.

19:30

S… Speaker 2 (part 7)

So, when you're doing that,

19:32

S… Speaker 2 (part 7)

you're doing that.

19:33

S… Speaker 2 (part 7)

You're doing it. You're doing it.

19:36

S… Speaker 2 (part 7)

You're doing it.

19:37

S… Speaker 1 (part 7)

You're doing it.

19:37

S… Speaker 2 (part 7)

You're doing it.

19:39

S… Speaker 2 (part 7)

You're doing it.

19:40

S… Speaker 2 (part 7)

You're doing it.

19:43

S… Speaker 2 (part 7)

You're doing it.

19:44

S… Speaker 2 (part 7)

Two months back.

19:46

S… Speaker 2 (part 7)

Two weeks.

19:47

S… Speaker 1 (part 7)

Two weeks.

19:48

S… Speaker 1 (part 7)

Two weeks.

19:49

S… Speaker 1 (part 7)

Two weeks. Two weeks. Two weeks.

19:49

S… Speaker 2 (part 7)

Two weeks.

19:51

S… Speaker 2 (part 7)

Two weeks.

Kokkuvõte

Küsi AI selle transkripti kohta