Wes Roth - Luma AI Stuns the ENTIRE Industry Transcript.txt

0:00
so you've seen the insane text to video
0:02
AI models coming out open AI Sora was
0:05
probably the best looking one yet but
0:07
not available to the public another next
0:09
gen model that made its appearance was
0:11
called cing but we had to have a Chinese
0:14
mobile number to use it so I think a lot
0:17
of us assume that we're not going to see
0:19
the top-of-the-line AI video models
0:21
anytime soon at least ones that we can
0:23
get a Hands-On ones we can
0:26
use but Lumi makes an appearance and
0:30
it's good really good but most
0:33
importantly it's available to use now
0:36
well it looks like everyone on the
0:37
planet is trying it out right now so the
0:39
demand is SK High it took me oh about 3
0:43
hours to generate this one prompt so
0:46
expect to give it a day or two before it
0:47
can start generating stuff on the fly
0:50
but let me show you some stuff it can do
0:52
some of this stuff will be somewhat not
0:54
safe for work if you know what I mean
0:57
and the last one will be straight up
1:00
scary like horror movie scary it's
1:03
called the tales from the other side
1:06
I'll leave that one for last in case
1:07
you're watching this alone at night in
1:11
the dark because I'm going to say it now
1:14
ai generated horror fakes have the
1:16
potential to be the scariest things
1:18
we've ever seen here's another YouTube
1:21
Creator in the AI space his channel is
1:24
theoretically media and he got Early
1:26
Access to this I'll leave a link to his
1:29
video down below but he was able to do
1:31
some image to video so for example
1:34
creating an image in mid journey and
1:36
then animating that with Luma Ai and he
1:39
was also able to create some scenes from
1:41
his own prompts so this I think is a
1:44
really good non cherry-picked example of
1:48
what Luma AI can do here it is if you
Early Access Showcase
1:52
ever played Hitman this is your chance
1:54
to recreate that as your very own movie
1:56
Very Dynamic very action-packed the
1:59
second version from that same prompt um
2:01
you know obviously per prompt you get
2:03
two generations uh yielded this as a
2:05
result and yes while there is some
2:08
decoherence and so each time it
2:10
generates two different ones so I got to
2:12
say that first one is very very good I
2:15
mean the fingers right the hands are a
2:18
nightmare but I mean it does look like
2:20
he's looking down the sights of the gun
2:21
that he's shooting in the prompt they St
2:23
that the Assassin should be bald which
2:26
it nailed I got to say it's pretty good
2:28
there it is he's shooting the gun
2:30
leaning around the corners I love it
2:32
next is Pirates of the Caribbean or
2:35
something like it right so some sort of
2:37
a adventure on a pirate
2:41
ship pretty cool and this is possibly
2:44
one of my favorites so this is uh an
2:46
image that he made so he made a static
2:48
image I'm not sure if it was mid Journey
2:50
but taking that image and adding it into
2:52
Luma Ai and the dream machine is what
2:55
this thing is called so Luma AI is the
2:56
company this is the dream machine this
2:58
is what it does so you can you can see
3:00
here it animates the movements she's
3:02
looking kind of keeping eyes on the
3:05
camera and one thing that's very
3:06
interesting is that you're able to add
3:08
certain prompts to make it for example
3:11
for her to change positions so here's
3:13
one with her hands
3:15
folded that's what that looks like so
3:18
her folding her hands across her body
3:21
again the arms and everything else looks
3:24
insane and not very good but I got to
3:27
say the fact that you can say what the
3:29
position should be what the movement
3:30
should be and that gets animated from a
3:32
static image that's pretty exciting
3:34
right so I'll put the links down below
Dream Machine Features
3:36
so you can try this for yourself and in
3:38
a minute I'll show you the best of luma
3:41
Labs like the best Generations that it
3:44
can make so you can see what it's
3:46
capable of at if you're picking the best
3:48
examples so of course as you're
3:49
generating various examples various
3:51
samples some of them will be good some
3:54
of them will be bad so often times
3:55
picking the best shot will come down to
3:58
generating 100 shots with maybe slightly
4:00
different prompts and then picking out
4:02
the best ones that's similar to how we
4:05
often times approach mid journey and
4:06
these other things like magnific
4:08
upscaler you do multiple variations with
4:10
different parameters different settings
4:12
different prompts to hopefully capture
4:14
that perfect thing which is not too
4:16
dissimilar to how for example some
4:18
photographers approach photo shoots
4:20
right sometimes if you've ever seen a
4:22
professional photo shoot the
4:23
photographer will point the camera hold
4:24
down the button you hear that rapid
4:26
machine fire click that super fast
4:28
clicking of the camera as it's taking
4:30
dozens and dozens of shots and then
4:32
afterwards they go through it one by one
4:34
picking out the best ones some of the
4:36
newest phone cameras also do the same
4:38
thing right they take a little clip and
4:40
you can pick out the best the perfect
4:42
sort of still moment within that shot
4:43
and the reason that's important to
4:45
understand is because if you're trying
4:46
to produce something that's like
4:47
cinematic something that's like movie
4:49
quality yeah you might have to produce
4:52
hundreds of shots thousands perhaps have
4:54
some sort of a filtering process maybe
4:56
even do a little touch-ups here and
4:58
there but then once you put it together
5:00
the results can be stunning so let's
5:01
take a look at this and in a second I'll
5:03
show you what's possible if you're
5:04
willing to go through and find the best
5:06
shots so Luma labs and this is their
5:09
dream machine which is an AI model that
5:11
makes high quality realistic videos fast
5:13
from text and images So currently it's
5:15
taking they say 120 seconds to generate
5:17
One Clip but right now there's a huge CU
5:20
for how there a huge cue so it's taking
5:22
hours but you can see when it's in Q so
5:25
before it starts processing and they're
5:26
saying that it's a highly scalable and
5:28
efficient Transformer model train Direct
5:29
on videos making capable of generating
5:31
physically accurate consistent and
5:33
eventful shots and it sounds like they
5:35
will have more and more abilities to
5:38
customize the thing that you're trying
5:39
to capture as this thing gets rolled out
5:41
to more users they're going to be adding
5:42
these functionalities and it sounds like
5:44
there's a bit of a trade-off between how
5:46
eventful a shot will be versus
5:47
consistency versus how custom you want
5:49
it to be so if you want a long shot with
5:52
very specific things in it then it might
5:54
approach a point where the character on
5:56
the shot is not really doing anything
5:57
just standing still whereas if you let
5:58
it be a little bit creative I can go and
6:00
generate something with a lot of action
6:02
so there's a little bit of a trade-off
6:03
happening there we talked about this it
6:05
can create high quality video from both
6:08
text and images so you can upload your
6:10
favorite image and then have that be
6:12
brought to life so things incredibly
6:14
fast 120 frames in 120 seconds and here
6:18
they're talking about consistent
6:20
characters and this is something that I
6:21
did notice about the videos the
6:23
characters stay consistent dealing with
6:26
some of the previous text to video
6:28
models throughout the shop the character
6:30
can morph into something completely
6:33
different here whatever else you can say
6:35
I got to say they stay pretty consistent
6:37
there's still some glitches with the
6:38
fingers and hands but the character
6:40
doesn't like fundamentally flip and turn
6:43
into something completely different they
6:45
can also do a lot of the different
6:46
camera movements Zoom pan rotate Etc
6:50
that's going to be able to create a lot
6:52
of really stunning cinematic shots and
6:54
some like dramatic close-ups Etc the
6:57
fact that they can go from underwater to
6:58
above water with this little bear here
7:01
is uh pretty cool I got to say and I got
7:03
to give them credit for doing this so
7:05
they spell out their current limitations
7:07
right so they're saying what are we bad
7:09
at what is this model currently what
7:10
doesn't it do pretty well so morphing so
7:13
if a car has to morph into a different
7:15
color or a different car they're not
7:16
doing that too well quite yet movement
7:18
so this dog gliding along the the ground
7:21
so maybe if there's some sort of a
7:22
background foreground type of thing
7:23
maybe it's not that good text is
7:25
notoriously difficult to do for these
7:27
both image models and video models
7:30
and then Janis which I don't know what
7:32
that is I looked it up Janus it could be
7:35
referring to a god with two faces so
7:37
some sort of a two-faced deity so here
7:40
as you can see here's a polar bear maybe
7:42
the prompt was for it to have two front
7:44
ends two faces and basically I think
7:47
what you got to think of is like what
7:48
these models are trained on so if they
7:50
see tons of hours of footage of polar
7:52
bears they can reproduce a polar bear
7:55
how many hours of footage of two-headed
7:57
polar bears are there in the wild
7:58
probably not as many but the big message
8:01
here I think is that when Sora came out
8:04
there wasn't anything quite like it we
8:06
had pabs we had Runway ml they were good
8:10
but they weren't still a inspiring they
8:13
were they had that AI generated look so
8:15
you could make some cool things with it
8:17
some of the best stuff that I saw with
8:19
it lean more towards like mystical and
8:21
horror because just how weird the video
8:25
looked here I think we're definitely at
8:26
a point where a lot more thing SC be
8:29
done do you want something that's a
8:30
little bit more romantic do you want a
8:32
cinematic shot do you want an action
8:34
shot do you want like this whatever this
8:36
bear and Swine things are over here some
8:39
sort of a leather clawed badasses that
8:42
are walking away from an explosion like
8:44
you want to do that you can do that and
8:47
now we saw this sort of next step in in
8:49
these AI video models not just from Sora
8:52
which we couldn't get a hands- on
8:54
although people in the Hollywood have
8:55
been apparently have early access to it
8:57
and then we had cing which you had to
8:59
have a Chinese mobile number to use and
9:01
now finally this which now everyone can
9:05
use now you're limited to a certain
9:06
amount of generations here's the pricing
9:08
by the way so you get 3 free Generations
9:10
per month so each generation is two
9:13
different shots so two samples and you
9:15
can upgrade to have 120 Generations at
9:17
30 per month 400 at 100 bucks a month
9:20
and 2,000 Generations if you're willing
9:22
to Shell out 500 per month which again
9:25
if you're using this to create some sort
9:26
of features movies you're probably going
9:29
to need need that but with that said let
9:30
me show you some of the best of things
9:33
that people have created using this Luma
9:36
Labs dream machine if this was
9:37
interesting please make sure you're
9:39
subscribed because we're going to be
9:40
covering this in depth as soon as the
9:42
traffic demand dies down a little bit
9:44
and it's possible to generate these
9:46
videos we're going to be doing a deep
9:48
dive and seeing what's possible with
9:50
something like this seeing is it
9:52
possible to create our very own movie or
9:55
cartoon or anime of some sort so stay
9:58
tuned lots more than coming make sure
10:00
you're subscribed and enjoy
Demos and Showcases
10:04
[Music]
10:34
at this point life seemed to be a
10:37
NeverEnding
10:39
Circle again and
10:42
again I did the same
10:50
routine like that auto taxi that flies
10:53
round and round all changed when I met
10:57
her Yuki the shrin keeper she was trying
11:01
to stop the robot sent to destroy the
11:04
Tor I felt the urge to help her and
11:07
deactivated the
11:08
bot she thanked me and showed around the
11:11
shrine's
11:12
garden I knew the corporation would send
11:15
more Bots but at that moment nothing
11:18
else mattered only the shrine only Yuki
11:33
[Music]
11:41
[Applause]
11:44
[Music]
11:58
[Music]
12:12
[Music]
12:17
me
12:18
[Music]
13:10
[Music]
13:19
close your
13:23
eyes count
13:32
think of
13:36
home and
13:41
[Music]
13:44
open you're
13:46
never
13:48
[Music]
13:49
never coming home again
13:59
it's passed
14:02
[Music]
14:04
away it's gone
14:10
away it's
14:13
pass
14:16
[Music]
14:18
away it's
14:19
[Music]
14:27
gone hey my name is Abel and let me tell
14:32
you the story of
14:36
[Music]
14:44
f on the other side the other side you
14:51
never know
15:04
desperate wondering soul will come my
15:07
friend
15:13
for
15:18
the you never
15:26
[Music]
15:33
[Music]