1/10/2025 - Initiating Creativity: Video Generation and Sound Design Techniques

Description

Chapters

Transcript

0:00 down what's
0:11 [Music]
0:18 [Music]
0:32 [Music]
0:51 Brandon said he just made it was
0:54 happening was shaking all right we are
0:59 live let me make sure yes we're live
1:02 turn that off look at that there I am
1:06 looking look I took time to to make my
1:10 Friday night date night hair all pretty
1:12 for y'all happy Friday night date night
1:16 [Music]
1:35 Source Camp does champy sing all the
1:36 time or just for for the Irregulars he
1:39 sings Anytime a guitar is played or it's
1:41 on
1:42 TV and also um brass bands playing uh
1:47 national anthems at the end of f or the
1:49 beginning of F1 races he sings to those
1:51 for some reason um your ASMR video is a
1:55 masterpiece thank you very much I uh I
1:58 let some of my bad behavior on these
2:00 lives Escape into the real
2:02 [Music]
2:07 world every time I'll see it
2:12 now yeah they looking
2:16 mad every time I see your
2:20 mou that
2:22 [Music]
2:24 smiley morning I the engine tur
2:31 the
2:34 [Music]
2:38 old you
2:41 leing again
2:45 today you convince
2:49 me again
2:53 today you're leaving this H to looking
2:57 for someone else's Golden Ring
3:00 [Music]
3:12 long now don't you
3:18 cry so long
3:21 [Music]
3:25 suan don't you cry for me
3:31 [Music]
3:41 champy doesn't seem very into it he's
3:42 he's not he's not all that excited I I
3:45 don't know what you people did to him I
3:47 don't know why he
3:49 is resisting why are you resisting what
3:53 is wrong with you
3:55 [Music]
4:03 there you
4:04 [Music]
4:09 go there's been something baby I've been
4:12 trying to
4:13 [Music]
4:15 say for an age and it seems I don't know
4:19 [Music]
4:22 how past and a future now surrounding
4:27 me surrender to whatever
4:30 can be
4:33 found there's been little
4:36 trouble s came to my
4:40 [Music]
4:44 rescue and if you like all of the rest I
4:48 would have quit you long ago but I
4:51 couldn't do
4:53 [Music]
4:58 that will and why I never went too
5:04 well make a man crazy and make him cold
5:07 as hell champ just walked away champ
5:11 champ is the ultimate Arbiter of taste
5:14 he was just like NOP no we're good I'm
5:17 just gonna go I'm gonna just go fart and
5:20 lay on the
5:21 [Laughter]
5:25 bed
5:26 steo hey long time
5:29 [Music]
5:42 [Applause]
5:44 [Music]
6:06 good evening good people if you're new
6:08 here to the channel my name is Kyle
6:10 Shannon this is the AI learning lab we
6:12 learn here we lab here we it's not
6:15 really a lab it's a$2 million uh TV
6:18 production studio uh we have we have
6:21 spent uh a ridiculous amount of money
6:25 trying to get it to look like a shitty
6:27 home office um
6:31 you know you would think by looking at
6:33 it that I've got just like a you know
6:37 $16 LED light on a cardboard box that's
6:41 lighting me but no it it is uh we've got
6:44 72 studio
6:46 lights uh varying degrees of color
6:49 temperature
6:50 brightness um refresh cycle so we don't
6:54 get the strobing effect it's a very
6:55 sophisticated operation made to look
6:58 like a janky piece of
7:01 [ __ ] you
7:03 know you know it's it's funny one of the
7:06 things people come into the AI learning
7:07 lab and they're like but Kyle what are
7:09 your qualifications and I have this
7:11 whole joke about not being qualified and
7:13 I have a degree from from Penn State and
7:16 acting it's it's all just part of the
7:17 act right I'm a uh I'm a nuclear
7:20 physicist by
7:22 training um I was not credited on on the
7:26 Transformers uh white paper the
7:28 attention is all you need white paper
7:30 paper uh but that was my idea that was
7:32 my idea and you know again jealousy
7:35 Reigns in the worlds of of high
7:38 mathematics um and for me having kind of
7:44 I don't know you know what you would
7:45 call genetically gifted good looks along
7:48 with you know uh just a ridiculous IQ
7:54 and and training the likes of which the
7:56 world hasn't seen people get jealous
7:58 people get jealous
8:01 um and so I'm not bitter I'm not bitter
8:04 not at
8:05 [Music]
8:10 all um all right good day good day good
8:14 day good day to you
8:17 then um what are we gonna do tonight I
8:21 think I think we're gonna play with
8:22 Korea Korea just added okay last night
8:25 there was a there was a whole situation
8:28 Danielle came in here all bothered she's
8:30 like she's like Kyle Minx has sound now
8:34 and I went to Minx and it didn't have
8:36 sound and then she was like uh I don't
8:38 one of the other ones has sound now Halo
8:40 maybe that's Mini Max I don't know and
8:42 that that that didn't have sound I found
8:45 out which ones got
8:48 sound so the drama the
8:51 drama we've
8:55 handled all right a doctorate from eju
8:59 it turned jany University
9:02 exactly
9:04 exactly oh man um I hope you're all well
9:09 if you're in California good Lord what a
9:12 horrible horrible thing happening as we
9:15 speak um I hope you're okay I hope
9:17 you're safe um you know uh it's just
9:21 it's heartbreaking it's heartbreaking I
9:23 can't I can't I can't process what's
9:27 happening um you know we had a little
9:29 bit of it here a a mini version of that
9:32 here in Colorado that high winds and
9:35 fire just blew through a town and uh
9:38 absolutely terrifying and it's just it
9:40 just goes on so
9:43 anyway I hope if you're watching here
9:47 this gives you a little bit of
9:49 uh a little bit of uh escape from the
9:53 from the insanity
9:56 [Music]
10:10 Danielle I can't help but if I have a
10:12 three second memories memory span me too
10:15 I'm right there with you I think we have
10:17 that in
10:19 common well you know what's funny is I
10:21 was just as I was coming on tonight I I
10:24 went to Korea not because of the sound
10:27 thing but because you could there's mult
10:29 multiple models video models that you
10:31 can test within Korea and then when I
10:34 logged into it it said oh we do sound
10:36 now I was like oh there's that so and I
10:38 actually found it the interface for it
10:40 is
10:42 buried uh in fact let me go I'm gonna go
10:45 add sound to this video I just
10:48 made
10:51 uh hi Winston I'm not the jealous type
10:54 it's difficult not to be jealous of a
10:56 guy like you exactly I understand listen
10:59 it's something I've come to accept in my
11:01 life that that just there's just going
11:03 to be there's just going to be
11:05 challenges with people comparing
11:07 themselves to me and you know the Grass
11:09 Is Always Greener glass is always more
11:11 full on my side of the equation I'm the
11:14 one with the gold button you
11:18 know exactly exactly all right CW I want
11:22 to know if you asked an AI to craft you
11:25 an
11:26 AI would it be a new AI
11:32 or would it
11:35 [Music]
11:40 operate
11:42 [Music]
11:47 well if you gave it new data to train on
11:50 I mean
11:51 the what makes the
11:54 AI what makes the AI models what they
11:57 are is their training and the data and
12:00 the data that they're
12:02 given
12:04 um what I would say to you is
12:07 [Music]
12:09 this I think it was Elon Musk said it
12:12 this week he said we have effectively
12:14 used all of the
12:16 data of
12:20 humanity like we have effectively any
12:23 data that we can put into a large
12:26 language model has essentially been put
12:28 into them
12:30 which if you think about it is
12:33 mindboggling right the collective
12:37 intelligence of the history of human
12:40 Humanity has been jammed into this this
12:44 box this AI box right they open up the
12:47 box they take all the data in the world
12:49 and they jam it in this thing and then
12:52 they close it up and this is your large
12:53 language
12:54 model
12:57 um so what what a lot
13:01 of what a lot of people what a lot of
13:04 research labs are doing right now is
13:06 they're using synthetic data where
13:09 they're having the large line th this is
13:11 actually fascinating there's David
13:12 Shapiro did some really good videos on
13:14 this which you can probably find on his
13:16 channel
13:18 um where they basically
13:21 take how did he call it he called it
13:23 activating the latent space you
13:25 basically ask a large language model for
13:28 for some topic let's say you want to I
13:30 don't know orthopods I I don't even know
13:32 if an orthopod is a thing but let's
13:34 assume an orthopod is a thing orthopods
13:37 o orthopods i it doesn't [ __ ] matter
13:39 um rly poly
13:41 bugs and you just have it spit out all
13:45 of the data that it's got on roly poly
13:47 bugs right all of it as much as it can
13:49 just vomit it out and then they have it
13:53 look at all that data and analyze it and
13:56 say which of this is actually good and
13:58 then they organize it and they tag it
14:01 and they put it into the optimal
14:05 format for training a large language
14:07 model so so they're basically taking the
14:11 raw vomit of the latent space of just
14:14 like the tokens around a given you know
14:17 semantic area and then they do some
14:20 analysis on it using another large
14:23 language model like a reasoning large
14:25 language
14:26 model so so open a right now is likely
14:31 using 03 to generate synthetic data to
14:34 train the next GPT 5 or something like
14:37 that so in that sense they're using
14:40 multiple AI models to create a new AI
14:43 model and yeah that new AI model will be
14:45 a new thing and it will be better
14:47 because the synthetic data is so pure
14:50 and so like it's so distilled into I
14:53 don't know the best data they can
14:55 have which is amazing
14:58 [Music]
15:06 there's always new data there is
15:12 but in terms of historical data we're
15:15 pretty much
15:20 done but that doesn't mean there's not
15:22 new data right synthetic data doesn't
15:24 mean it's not new data
15:29 and again if you think if you're
15:31 thinking about computers as they
15:33 traditionally worked if you're thinking
15:34 about like a
15:36 database and you're saying oh well you
15:38 pulled stuff out of a
15:40 database then that's already data we've
15:42 already used no that's not how large
15:43 language models work the G and GPT is
15:46 generating you're generating new stuff
15:50 out
15:52 of these fragments of words that are in
15:55 Thousand dimensional mathematical space
16:00 it's weird
16:08 man that's pretty cool this Korea thing
16:11 is pretty
16:12 [Music]
16:19 cool oh good lord good Lord good Lord
16:24 good people good people good people so
16:27 lesson lessen you and
16:31 lasses
16:33 um here's how this
16:35 works it's it's just absolute [ __ ]
16:38 chaos here I might just talk for 90
16:41 minutes um I could show something cool I
16:45 might take the guitar off my neck the
16:47 dog might come
16:48 back
16:50 uh but if you have questions here's the
16:54 this is one of the magical things about
16:56 this and by the way it looks like Tik
16:57 tock
16:59 be what do we have today's the uh the
17:02 10th nine
17:06 days Tik Tock maybe go in the Wayside
17:09 but we'll see we'll see what the Supreme
17:10 Court says Define
17:13 data well data that you train large
17:16 language models with is is very
17:18 particular you've got to have tag data
17:21 right so you've gotta you basically say
17:22 here's a bunch of data and then you have
17:25 metadata that tells the large language
17:27 model what that is and then it somehow
17:31 figures out how to break all of so so
17:33 let's say you put in a 100 page
17:36 book and you tag it with the appropriate
17:39 tags that says what it is and categories
17:41 and all that sort of [ __ ] it takes all
17:44 of the words of that book and it breaks
17:48 the words and punctuation and spaces
17:51 into things called tokens and then turns
17:54 them from fragments of words into
17:57 numbers and then it grp groups those
18:00 numbers in literally thousand plus
18:03 dimensional space in in semantic
18:08 clusters and you're like what the [ __ ]
18:10 is a semantic cluster
18:13 exactly like the the mathematicians that
18:16 figured this [ __ ] out and continue to
18:18 optimize this pate who's a fan of the
18:21 channel and irregular here he he
18:23 actually does this for Google he works
18:25 on optimizing inferences using tensors
18:29 um so there's people that all they do is
18:31 make this math better but what it does
18:33 is it like so like in this book let's
18:36 say in this 100 page book that you've
18:38 got you have the word dog right and dog
18:41 shows up in multiple
18:44 places and in one place in the book
18:47 they're referring to the dog is he came
18:49 home after A Hard Day's Work and the dog
18:51 was there with his wagging tail ready to
18:54 sing a song well that version of the
18:57 word dog is going to go into a semantic
19:00 cluster of like pets and dog as animal
19:04 and companion and lovely little thing
19:07 and then let's say in another part of
19:08 the book they're like ah he he instead
19:12 of you know having Integrity he ghosted
19:15 you know the woman and he's a dog
19:17 because she thinks he's a dog because of
19:18 that well then that version of dog goes
19:21 into a different semantic cluster but
19:24 it's not the word itself it's a number
19:26 that represents the word or the fragment
19:28 of the word and it's just it's [ __ ]
19:31 the the more you understand about how
19:33 large language models
19:35 work the more I'm convinced that they
19:39 shouldn't [ __ ] work but they do
19:41 they're they're like a little miracle of
19:42 mathematics
19:46 so all your data belong to us exactly
19:51 [Music]
20:08 pate's here what's happening Pate Jason
20:11 Wolf Man Clint Tick Tock is technically
20:14 all
20:15 semantic we need a lot of decentralized
20:18 apps and
20:21 websites I don't know what we need you
20:23 know it's funny we were talking today on
20:26 office hours um about
20:31 agents and
20:33 and you know one of one of one of pate's
20:36 favorite things on this channel is when
20:38 I speculate um but what we did on office
20:40 hours today was I actually asked people
20:42 to speculate I said if you assume that
20:46 we're going to get agents you know
20:48 sometime in the next year and that
20:49 agents are these things that can take
20:52 actions in independent of us right we
20:54 can give them goals and we can direct
20:56 them but they go off and just do this
20:58 [ __ ] where
20:59 we don't need to know how to prompt them
21:02 right someone did like someone created
21:05 the agent we may have to set up agents
21:07 but once they're set up they just go do
21:10 [ __ ] and they could go do [ __ ] like go
21:13 write a marketing plan and then some
21:14 other group of Agents could go write
21:16 content for that marketing plan and then
21:19 another group of Agents could come up
21:21 with media buying strategies and then
21:23 another group of Agents could go
21:25 negotiate good media buying deals and
21:28 then another group of Agents could write
21:30 all the content Source all the content
21:32 generate the images assemble the things
21:35 and generate a whole bunch of content
21:37 for the media plan and then the media
21:40 plan gets executed and then there's
21:41 another bunch of agents that test all of
21:43 that content and optimize
21:46 it and you don't have to do
21:50 [ __ ] right like that's the promise of it
21:53 now the reality of it is we don't have
21:56 agents yet we've got you can build your
21:58 own with some of the you know agent
22:00 tools within the llms and apis and Lang
22:03 chain and [ __ ] like that um and there's
22:06 some people doing interesting work in
22:07 that but we don't have you know we don't
22:09 have open AI or anthropics or meta agent
22:14 framework that's going to make it really
22:15 easy to do these
22:18 but assuming that that
22:22 happens what is our
22:25 role what is our
22:28 role and it it it was a fascinating
22:30 conversation and it you know it it it
22:34 really got to things like
22:38 um you know our role might be to do
22:41 things like support one another and
22:45 figure out how to use agents to support
22:47 people that were displaced because of
22:49 agents and then maybe those agents can
22:51 help those people get up on their feet
22:53 and start their own business or or do
22:55 whatever they might
22:57 do um
23:00 and the thing that struck me is that as
23:02 a business
23:04 owner I think anyone running a business
23:07 right now we are we are about to enter a
23:10 gauntlet that's a [ __ ] Gauntlet for
23:14 probably the
23:16 next
23:19 decade where things are going to
23:22 change every quarter but maybe even
23:25 every month that how you ran your
23:28 business today is going to be different
23:31 than how you run it 3 months from now or
23:33 6 months from now and what you do and
23:36 what services your people provide the
23:39 business is all of it's going to change
23:42 so I feel like there's going to be there
23:44 what's going to emerges two different
23:45 kinds of businesses one is going to be a
23:48 business that essentially exists today
23:51 every business which basically says
23:54 let's
23:56 see I'll have my agent meet your AG
23:58 agent they can have a power lunch
24:00 exactly the businesses of today kind of
24:04 deal with change like okay we had an
24:06 economic downturn and that's affected
24:08 our business and then we're going to put
24:10 in some change management strategy and
24:12 then over the next you know six months
24:14 we're going to do we're going to
24:16 implement the change management strategy
24:18 and and then for the last six months of
24:21 the year we're going to run the business
24:22 and then we we should be in good
24:26 shape that's not going to work all that
24:28 well
24:29 in an environment where what's
24:33 possible for for businesses to make
24:35 money changes on a regular basis if
24:39 we're indeed moving into this
24:41 accelerating you know
24:43 exponential capability
24:46 curve there's going to be a new kind of
24:48 business that is constantly adapting and
24:52 as as an
24:54 individual if you were someone who likes
24:57 things in
25:00 stasis you like to plan things you like
25:03 things to be exactly as
25:05 predicted I think there's going to be a
25:07 a just a [ __ ] ton of chaos coming so
25:10 anything any of us can do to get in that
25:12 mindset of adapt adapt adapt get curious
25:16 get adventurous get curious adapt adapt
25:19 adapt I think that's going to
25:22 be the ones that succeed in the next 3
25:26 to five to 10 years are going to be ones
25:28 that are Nimble and willing to adapt and
25:33 in and in fact aggressive to adapt so
25:38 it's it's wild blessed are the flexible
25:43 indeed age of disruption is kind of a an
25:46 understatement I think
25:52 [Music]
26:06 our roll Is How We Roll pretty much that
26:10 pretty much that
26:14 [Music]
26:21 [Applause]
26:23 [Music]
26:32 uh oh Brandon's getting bored Brandon's
26:34 like Korea could we do something here
26:36 buddy are you just gonna sit there
26:38 knocking on your guitar for for an hour
26:41 people are the people are getting
26:43 restless all right fine fine fine coming
26:47 coming up on 30 minute
26:52 hello oh you people are good you're the
26:55 best you're all the best happy Friday
26:58 night date night by the way hope you're
26:59 doing
27:02 well okay let's get going let's go do
27:06 some
27:13 stuff
27:16 um you won't be able to hear this
27:23 but I just made that earlier um I want
27:27 to go try something though so last
27:31 night last night we went to LTX
27:36 studio and we had these
27:45 [Music]
27:47 cool I'm gonna
27:52 take this to Korea okay so in
27:56 Korea we're going to describe seem to
27:59 generate so so I'm at I'm at ka.
28:04 aapps and this is their video generation
28:07 app so if I go out to the to the main
28:09 page there there's okay
28:12 so so if you've got Crea um you can make
28:17 video here you can make images here as
28:19 well you can also do real time images
28:21 here maybe we'll play with some of these
28:22 things um right here it says Ultra fast
28:26 video Generation Now with sound and
28:29 you're like oh cool and if you click on
28:31 this it takes you to Twitter and shows
28:34 you effectively a press release
28:37 which like tell me it's the story of
28:42 that you've got audio like this should
28:44 take me too how do you [ __ ] make the
28:45 audio and if you scroll down here
28:48 there's nowhere where it says you know
28:50 video to audio however there's this
28:52 little blue button right here can anyone
28:54 see this not
28:56 really um this little this little blue
29:00 button says new models and video to
29:02 audio so if we go into here here's our
29:07 videos we've made on the left here's our
29:09 little scene we're going to generate
29:10 here we're going to go
29:13 720p the the model down on the bottom
29:16 here um cling 1.5 is one I just did
29:20 they've got runway in here they've got
29:22 Luma in here Luma by the way is coming
29:25 out with a new model I think I'm going
29:27 to get early access to it as a uh as a
29:30 creative partner so I'll let you know
29:32 that um the 01 live model high quality
29:36 model for animating people I don't I
29:38 have not played with that halou this is
29:41 the minia max one which is really
29:44 good um oh this is cool it also tells
29:47 you how long these take cing takes five
29:49 minutes Runway Luma take
29:52 one
29:53 Hunan takes one and Halo takes 10 all
29:57 right so let's we'll stick with
30:00 Hunan and then we'll drop in let's see
30:04 uh let's see
30:06 uh in the style of Pixar
30:16 Animation a wide view of a cozy garage
30:19 showc casing scattered tools blueprints
30:22 work prints of of the vibrant purple
30:26 Beetle we'll say 196 69
30:30 1969 VW Beetle is parked prominently in
30:34 the center glistening under the soft
30:37 fluorescent
30:39 lights with
30:43 eyes behind the wind
30:48 shield all right let's see let's see how
30:51 this does this ca.
30:55 aai khaki Fitz Simmons hey Kyle a sister
30:59 up with Early Access if I get it I'll
31:01 definitely be showing it here um okay
31:05 here's one of my strong strong strong
31:07 recommendations if you want to be doing
31:11 video for every video tool out there go
31:15 find their creative partner
31:17 program and um and uh and apply for it
31:24 because they're all too expensive
31:25 they're all too too expensive they just
31:28 are
31:35 Jank
31:37 shank all right let me grab oh so wait
31:40 so what we could do here I could go to a
31:43 different let's go to a different one
31:44 minute model we'll do Luma Luma kind of
31:47 sucks right now let's do
31:50 runway and we'll generate one with
31:56 Runway oh wait did I just lose
32:00 one I think I
32:05 did
32:10 huh that's shitty in her face but maybe
32:13 I didn't that's the way these things
32:15 work is
32:21 weird now I think I lost
32:24 something dang it all right
32:35 bang
32:38 copied go back here go
32:42 720p go bang
32:48 oh all right so now okay so there's our
32:51 video so let me change I'm going to I'm
32:53 going to switch modes here for a second
33:04 okay so how you do this is underneath
33:09 the video you can reuse the parameters
33:12 you can copy the prompt you can publish
33:15 or you can add sound so this is the
33:18 little button where we can add sound all
33:21 right so here we go bang
33:30 that's a cute little 69 Beetle isn't it
33:33 the eyes are creepy it kind of made them
33:36 the the other model was better but
33:38 here's the
33:42 [Music]
33:44 sound well that's kind of cool
33:47 [Music]
33:51 music
33:53 uh try
33:57 [Music]
34:03 [Music]
34:10 [Applause]
34:19 oh wait
34:34 hey Quinn how's it
34:36 going hey Kyle I'm doing well how about
34:39 you doing good um listen you know it
34:43 still kind of freaks me out that you can
34:45 talk this good and this fast and I kind
34:47 of feel like I'm in a Sci-Fi movie but
34:49 you know here we
34:54 are it's wild right it's like you want
34:57 to dive into today or just chatting no I
35:00 want to dive into something um I want so
35:04 there's a new text to video model out
35:08 that has um you you generate the video
35:12 and then it generates a sound for you um
35:15 of whatever's going on in that video
35:17 right so it creates it creates special
35:19 effects sound you know sfx for you which
35:21 is pretty cool so what I want you to
35:23 help me do is I want you to help me come
35:25 up with um and and they're 5c Clips so
35:28 these are short little Clips so I just
35:30 want your concept of a of let's say 20
35:35 different video clips and those video
35:37 clips I want you to just describe write
35:39 a prompt right and the prompt will be
35:42 for a textto video generator where it'll
35:45 include a subject lighting uh location
35:49 uh cinematic style maybe film stock
35:52 whether it's animation or black and
35:54 white or things like that so I want them
35:56 to be all sorts of varied things things
35:58 and then on top of that I want them to
36:00 be Concepts that it's really clear what
36:02 the sound effect should be for each of
36:05 those ideas
36:09 okay sure I can do that let's get
36:12 started here are 20 prompts prompt a
36:15 lone Lighthouse standing on a rocky
36:17 Cliff during a stormy night cinematic
36:19 style Moody and dark lighting flashes of
36:22 lightning film stock black and white
36:24 sound effect crashing waves and thunder
36:26 prompt a bus
36:37 yeah see I lost all that dang
36:42 it hey do you still have those 20
36:48 ideas Quinn are you there did I lose
36:53 you all right so so here's here's what
36:57 I've learned about doing um automated
37:00 voice to text is um if you don't let her
37:04 say all the ideas then you don't get
37:07 them typed out so so let me start
37:11 over and my apologies to all of you but
37:15 but okay if you haven't used text to
37:18 voice before one of the things that
37:21 large language models are remarkable at
37:25 is taking unstructured data and making
37:27 sense of it so you heard my words right
37:32 I am not exactly a prompt engineer
37:34 Maestro here I'm literally just going uh
37:37 I want you to come up with ideas to do
37:39 this and that and then I kept add adding
37:42 crap on top of them and she just
37:44 completely took that in and got got it
37:46 like that so I'm gonna start over and do
37:48 it again but take notice of that
37:51 that you don't need to be nearly as pres
37:54 prescriptive as you might think you need
37:56 to be with these large language models
37:58 they're quite remarkable of making sense
38:00 of your you
38:01 know logic challenged sentence
38:05 fragments especially if you're an
38:09 American because you cut her off I know
38:12 I did that's exactly
38:17 why okay Quinn how's it
38:19 going hey Kyle it's going great how
38:22 about you how's your day going I'm I'm
38:24 doing good uh we just talked about this
38:26 before but I cut you off and then we
38:28 lost uh what you created so here's
38:30 here's what I want you to do I'm working
38:32 with a um a text to video uh image a
38:36 video generation model that also has um
38:40 video to audio generation so what I want
38:44 you to do is I want you to write me 10
38:46 super
38:47 creative um video prompts for the textto
38:51 video portion that you know have things
38:54 like a really interesting subject an
38:56 interesting location interesting
38:58 lighting camera Styles cinematic Styles
39:01 whether it's black and white or
39:03 cinematic or animated or Pixar whatever
39:06 it might be um and then I want you to
39:09 make sure that the subject or the
39:11 location H has a really clear sound
39:14 associated with it so when we put this
39:16 video into the video to audio generator
39:20 it will make us really good sound
39:22 effects make
39:24 sense got it here are 10 super creative
39:27 prompts for the textt to video portion
39:30 subject a vintage70s muscle car parked
39:33 in a neon lit Alleyway location Urban
39:36 Backstreet at night lighting Moody neon
39:39 glow cinematic style Noir sound the
39:42 distant hum of the city punctuated by
39:44 the car's ring engine subject a 69
39:47 Beetle driving through a picturesque
39:49 Countryside location Rolling Green Hills
39:52 and wild flowers lighting soft morning
39:55 sunlight cinematic style pastoral
39:58 sound the gentle hum of the engine and
40:01 birds chirping subject hey Quinn you
40:03 don't you don't need to you don't need
40:05 to make the sound prompt just the text
40:07 to video
40:11 prompt got it here are 10 text to video
40:13 props subject a vintage70s muscle car
40:17 parked in a neon lit Alleyway location
40:20 Urban back street at night lighting
40:22 Moody neon glow cinematic style Noir
40:25 subject a 69 Beetle driving through
40:27 through a picturesque Countryside
40:29 location Rolling Green Hills and wild
40:31 flowers lighting soft morning sunlight
40:34 cinematic style pastoral three subject a
40:38 70s muscle car racing through a rain
40:40 soaked City hey Quinn I appreciate you
40:43 using 70s muscle car for all these but I
40:45 want each I want each one to be really
40:48 unique from one another and completely
40:51 distinct style-wise so that that these
40:53 are all individual little Snippets
41:00 understood here are 10 distinct prompts
41:03 subject a futuristic hovercraft
41:05 navigating a bustling cityscape location
41:08 neon drenched cyberpunk Metropolis
41:10 Lighting electric and vibrant cinematic
41:13 style sci-fi subject a night on
41:16 Horseback Galloping through a dense
41:18 Misty Forest location Enchanted Woodland
41:21 lighting soft and ethereal cinematic
41:23 style fantasy subject a retro robot
41:27 exploring an abandoned space station
41:29 location derel space interior lighting
41:32 dim and Atmospheric cinematic style
41:35 retrofuturistic four subject a painter
41:38 creating a masterpiece on a cliff
41:40 overlooking the ocean location Coastal
41:42 Cliffside lighting Quin can can you type
41:45 these out so I can read
41:47 them absolutely here are the prompts
41:49 typed out for you one subject a
41:52 futuristic
41:53 hover okay yeah see see it cut it off it
41:57 cut it off right there so we got we got
41:59 four of them um yeah it's fascinating
42:04 that's probably
42:08 a that's probably like I think the way
42:12 to probably do that is because she
42:15 doesn't seem to be able
42:18 to process in the background so I think
42:22 what you do is you have the initial
42:23 conversation and set up the prompts and
42:26 then you get out of The Voice
42:27 conversation and then type in the next
42:29 prompt and say okay give me 20 ideas in
42:31 fact let me say um great
42:36 um give me um 20 more
42:48 prompts yeah there we go that's how to
42:50 do
42:51 it so have have your your cute little
42:54 conversation with
42:55 her and then okay so now so let's go
42:59 grab what do we have here a pirate ship
43:01 caught in a Whirlpool that this one's
43:03 good all
43:04 right we'll go over here we'll go to
43:07 we'll do it live we'll do it live
43:12 um let's go back to this Hunan one okay
43:16 we're in h
43:19 Hunan so
43:24 there's wait why did that not not copy
43:27 oh are you coming with
43:28 me oh you're not okay hang on hang on
43:32 people hang
43:34 on
43:37 [Music]
43:41 shank
43:42 shank dank yeah okay I'm gonna jump back
43:47 to here I'm gonna copy this I'm gonna
43:50 come in here I'm gonna select all that
43:52 I'm going to paste okay so we're doing
43:55 720p in hun
43:58 we'll use
43:59 Hunan we'll do it
44:02 live all
44:04 right one minute
44:12 [Music]
44:15 remaining so this is our pirate
44:24 ship um I I'll do the cling one the
44:27 problem with the clling one is that it's
44:29 a it's a 10-minute
44:32 generation so let me let me let this
44:35 generate again it's like this is weird
44:38 that it doesn't have background
44:39 generation like I should be able
44:46 to
44:53 oh maybe I do
44:55 have some history here I don't know I
44:58 don't know how this [ __ ] thing
45:10 works yeah yes clling nap and come back
45:13 later uh yeah I have a feeling that all
45:15 the good video tools are going to be
45:17 like that that there if you want if you
45:19 want the high quality [ __ ] oh by the way
45:22 the other thing that we can do is Runway
45:25 now does 4K UPR resing now I don't know
45:29 if it does 4K U uing on non Runway
45:35 videos so it'll be interesting if we
45:37 download this video if we can upres this
45:40 into uh whatever you
45:47 know I need V3 from Google or whatever
45:50 you call it yeah V2 I need V2 as well
45:54 from Google I don't know why why are
45:57 there hold in it back but it's pissing
45:58 me
46:02 off this is fascinating watch it
46:05 watching it make this
46:07 video it's boring but you know what are
46:11 you gonna
46:13 do what do you want to go do build build
46:15 a scale model of a ship or build it in
46:18 3D and then come up with some fluid
46:20 dynamics in a 3D program is that what
46:23 you want to
46:24 do yeah we're not doing that anymore I
46:27 don't know if if you got the
46:29 memo all the 3D people like it's not the
46:32 same yeah I know it's
46:37 not oh
46:39 man all right this one should be a good
46:42 Audio One Quinn Quinn did good with this
46:44 one all right let's let's add sound to
46:46 this
46:47 bang but relevant to your comment Vicky
46:51 please comment on my post in the
46:53 mechanics Guild if you want to help
46:55 guide any for okay cool
47:00 [Music]
47:02 so I'll talk about P I'll talk about the
47:07 [Laughter]
47:09 uh P I'll talk about the salon in a
47:12 second and I'll Point people to your
47:13 area Bloom Bower hey Kyle what's shaking
47:17 what's going down all right so here we
47:18 go here's the uh here's the audio of
47:21 this here little clip
47:26 [Music]
47:36 fascinating not great but
47:40 fascinating all right um you know what I
47:42 want to try to do no Kyle why don't you
47:44 tell us shut
47:46 up let's go there let's go
47:50 to Runway Runway has video to video
47:54 right so I should be able to import that
47:59 video and then we should be able to
48:01 upres that to
48:12 4K oh you're not seeing any of
48:17 this hey Apple and Google could you stop
48:19 your corporate war and just make audio
48:21 sharing work
48:24 please please
48:29 [Music]
48:43 drop an image or
48:47 video select an
48:50 asset upload the asset let's go get
48:54 our ship video
48:58 there's our shitty ship
49:07 [Music]
49:12 video 100 foot swells braah I'm gonna
49:15 get totally pitched by them
49:21 barrels that's some gnarly surf it's a
49:24 rogue wave it's a rogue wave
49:28 Captain turn apart turn apart we've got
49:32 a
49:36 rogue oh no are we not uploading again
49:39 is that what's going on
49:43 here what the
49:45 [ __ ] we had this problem the other
49:48 night there we go
49:57 [Music]
50:06 D how is it that two different systems
50:11 are creating 720p videos that are not
50:14 720p I don't get it I'll never get it
50:17 let's see pirate ship
50:22 tosses
50:24 in in a viol
50:29 storm um 5.2 seconds
50:34 turbo gen 3 Alpha yeah let's go high
50:37 quality [ __ ] it
50:42 bang in the queue that'll start in a few
50:44 minutes all right let's go back
50:48 to so here we are at Korea let's go back
50:51 to our little chat je let's go get
50:54 another go get another video to make a
50:56 lone astronaut planting a flag on the
50:58 alien Moon steam locomotive roast racing
51:02 across a snowy mountain pass this feels
51:04 like it'll have good sound good copy go
51:08 to
51:09 Korea go in go paste let's do a
51:12 different model let's do we'll do
51:14 another minute one what's a minute one
51:16 here we'll do Luma for this one
51:23 generate and then I should be able to in
51:26 theory
51:27 three yeah let's do a 10-minute one
51:29 we'll do a cling Pro on this one cling
51:32 or cling
51:33 1.6 cling 1.6 five
51:37 minutes that's the one that I think
51:39 everyone's
51:41 digging okay yeah so that's working so
51:43 so I've still got that one
51:45 generating this thing looks like it hung
51:48 so we could probably kill this right
51:51 cancel all right so that's rolling let's
51:56 go back the runway
52:06 66% black bar black oh black oh
52:16 black anybody have any questions making
52:19 sense what I'm doing I know I'm just
52:22 sort of adding around but welcome to
52:24 Chad add
52:30 oh
52:43 fascinating that's
52:55 weird oh you can't hear that can you no
53:05 because does does anybody know anybody
53:08 that works anywhere that we could find
53:10 out you know the person that came up
53:13 with the audio sharing framework for for
53:16 computers and can
53:19 we puncture their above ground pool and
53:23 drain
53:25 it um
53:28 oh here
53:35 here's see is it here that we do this no
53:40 is it here no actions look upscale to
53:46 4K all right we're upscaling to
53:50 4K that's a shitty video to upscale to
53:52 4K but we're doing it we're going
53:55 in oh oh oh this looks
54:00 cool can you see this oh no of course
54:02 not because I'm sharing audio I I am so
54:07 like tonight is one of those nights
54:09 where if you're me your life is a living
54:12 hell because you can't share the thing
54:15 with the
54:16 thing I don't know what he's talking
54:19 about he seems very bitter is he bitter
54:21 he seems a little bitter yeah I'm bitter
54:25 first of all I'm a gen xer why is the
54:27 train rolling
54:30 backwards why why is the train rolling
54:35 backwards why are the wheels not
54:38 rolling that's
54:40 exhausting which one was that that was
54:44 cling it's
54:48 exhausting which one's this one oh this
54:51 one's cling what's this one this is Luma
54:54 lum's got a new model coming out I'm I'm
54:57 I'm pinning my hopes on lum's new model
54:59 cat GPT says each company is PRI
55:02 prioritizing its proprietary systems for
55:05 competitive and strategic reasons I
55:07 guess we're stuck with it exactly
55:09 because it's AI Kyle embrace the Jank
55:13 right Mochi is making videos now it
55:17 seems totally optical
55:20 illusion yeah this is one of those
55:22 optical illusions the train cars are all
55:25 the same size and nothing nothing is
55:29 moving but let's do this let's add audio
55:32 to this this year video add
55:36 sound uh
55:37 oh did I lose oh no wait
55:44 what oh my God computers are just Mal
55:48 behaving here I am canine in the AI
55:52 salon and I have a business IDE in the
55:54 area of app development networking not
55:56 sure I could should connect um there's a
55:58 Shameless self-promotion area I am
56:01 canine and there's also a call for
56:03 collaborators area so if you're looking
56:05 for a partner that could be
56:08 good um oh this looks wait from there
56:12 this looks pretty
56:14 good
56:16 although what is the truck on the
56:18 elevator railro railroad
56:23 tracks all right we'll add a sound to
56:25 this one too and then let's go listen to
56:27 the sound here I think I'm sharing
56:37 sound
56:41 pretty all right let's go look at our
56:43 Runway oh you can't see that here's our
56:48 4K our 4K ship let's go look at our 4K
56:51 ship
56:59 [Music]
57:06 because it didn't do it photo realistic
57:08 4K is kind of
57:13 useless but you can I mean the the water
57:16 is pretty clean here I mean I guess
57:19 that's 4K we've got a really bad sound
57:21 effect
57:24 [Music]
57:33 it's so bad it's so
57:38 bad oh my
57:42 God this is so
57:45 funny it doesn't seem that funny
57:49 [Laughter]
57:59 let's go try another model let's try
58:02 we'll try
58:04 Runway generate and we'll try what's 01
58:08 live oh that's animating people Halo
58:10 this is minia Max Halo of these Halo
58:14 should be the best the best of
58:16 these all right so we got that one we
58:19 got that one we got those two shitty
58:21 ones all right let's go I'm going to go
58:24 back to chat GPT and grab another thing
58:26 you you're you're not seeing what I'm
58:27 doing
58:28 here are we just not going to address
58:31 these sounds oh I've addressed them they
58:33 sound like
58:36 [ __ ] they're they're bad although you
58:40 know if you just want something to have
58:42 some basic sound in it then you know
58:45 imagine cutting this into a movie where
58:47 you've got background music and you're
58:49 adjusting the levels and things like
58:50 that it's fine uh they're not they're
58:53 not very good though
58:56 my guess is that this is um these sound
58:59 about as good as 11 Labs my guess is
59:01 what Kaa did was just um did a call to
59:04 the 11 Labs
59:07 um sound effects
59:10 API is my guess is my gu all
59:17 right a giant octopus now that's going
59:19 to be weird a hot air balloon floating
59:22 across a jungle
59:24 canopy uh
59:27 a motorcyclist a blacksmith forging a
59:30 glowing sword okay this one should be
59:32 good for audio as
59:36 well so we're going to come in here
59:38 we're going to
59:39 paste there's our blacksmith so let's go
59:42 we'll we'll go back to
59:47 Hunan why did it change it to
59:50 Portrait what is with these software
59:53 companies get your [ __ ] ux together
59:58 [Music]
59:59 um let's try 01 live because that's good
1:00:02 for animating people and this is a
1:00:03 blacksmith so let's try that one oh
1:00:06 gotta pay for that
1:00:12 one fine no need to be rude about
1:00:17 it how do I get out of
1:00:20 here oh
1:00:22 wait God damn it
1:00:31 what did I lose all my oh no here they
1:00:35 are
1:00:38 okay there's our blacksmith I don't see
1:00:40 a blacksmith there oh he's off screen
1:00:43 that's
1:00:44 fascinating all right that one's doing
1:00:46 eight minutes all right so let's go do
1:00:48 we'll throw we'll throw Luma in there
1:00:50 Aluma
1:00:55 generation I'm start frame end
1:00:58 frame oh that's cool you can do start
1:01:00 frame end
1:01:01 frame
1:01:05 generate oh two generations at a time I
1:01:08 see what's
1:01:09 happening all right fine
1:01:13 whatever boy that's bad
1:01:31 all right um let me I'll let those
1:01:35 things do their
1:01:37 thing and then um we'll come back to
1:01:40 them who's got questions questions
1:01:43 comments um people are laughing the ship
1:01:45 one reminded me of the Hoodoo Gurus
1:01:48 video was that Leonard Skinner
1:01:54 skinnard I don't know what I was singing
1:01:56 could have
1:01:58 been why not try the real
1:02:01 Luma uh just because I wanted to try the
1:02:04 audio thing here and I don't have I
1:02:07 don't think I have access to the new
1:02:08 Luma Labs new model Luma lab's current
1:02:12 model is is a bit poopy but the new one
1:02:17 looks pretty
1:02:19 amazing um let me see if I got access to
1:02:22 the new
1:02:24 one I don't think so oh and there's no
1:02:28 way for us to
1:02:30 know videos modify what's
1:02:35 modify reference key
1:02:39 frame
1:02:45 huh yeah I don't this isn't this isn't
1:02:48 the new
1:02:53 one oh you're not looking at this yeah
1:02:55 no I don't think I have it
1:02:59 okay
1:03:01 podcast yeah I'm not going to do the
1:03:03 podcast tonight just
1:03:07 because
1:03:09 I we'll get there we'll do it next week
1:03:13 Brandon
1:03:19 um oh that's fascinating here look at
1:03:22 this
1:03:30 um am I sharing I think I'm sharing
1:03:32 audio here let's say add audio add
1:03:37 sound so this should
1:03:41 have Tink
1:03:45 shink let's see let's see if we get the
1:03:48 good audio all right Pate here you
1:03:50 [Laughter]
1:03:54 go yeah
1:04:06 good that's not
1:04:09 bad I mean the video is not
1:04:17 great okay so so
1:04:20 here's pay you wanted to comment on the
1:04:22 audio here's my comment on the audio
1:04:27 my comment on the audio right now
1:04:31 is it's as janky as six-finger you know
1:04:35 six-finger
1:04:37 crap and just like six-finger crap next
1:04:40 year it won't be
1:04:43 janky and also next
1:04:45 year I okay so so
1:04:52 imagine imagine a tool that can
1:04:56 make a 1080p
1:05:01 video
1:05:06 um faster than its
1:05:09 length here's what I mean imagine being
1:05:12 able to make a video that's 5 Second
1:05:14 long five seconds long but it takes four
1:05:16 seconds to
1:05:17 generate and then you have these sound
1:05:21 effects built into it what that means is
1:05:24 that you could have a system that is is
1:05:26 just prompting One Clip after another
1:05:29 and telling a whole 2hour movie
1:05:32 generating it on the
1:05:36 Fly
1:05:39 amazing yeah Fabiana I always remember
1:05:42 it's just the start yep it's just the
1:05:44 start this like this is this is in the
1:05:48 um yeah exactly this one sounds okay um
1:05:52 this is in that category of you know
1:05:54 this is the worst it's ever going to be
1:05:56 so this one is this is
1:05:58 minx so this one is close but again look
1:06:01 the wheels aren't Wheels aren't rolling
1:06:04 but the rest of it looks really good if
1:06:08 if only the wheels were
1:06:13 rolling all right so let's add some
1:06:15 sound to this one add sound
1:06:26 wait what's this one oh that's
1:06:29 Runway let's add sound to that one
1:06:33 too I think Runway is the best of
1:06:37 these that one's a train wreck get it
1:06:40 train wreck this one's a train
1:06:45 wreck yeah yeah we we expect so much of
1:06:48 these tools it it it's so true we're so
1:06:51 [ __ ] spoiled we're literally talking
1:06:55 videos in into existence that you know a
1:06:58 year ago you would have had to hire a
1:07:01 special effects team or an animator to
1:07:03 do
1:07:05 [Laughter]
1:07:07 this um let me let's let's
1:07:11 change I thought
1:07:14 we there's the
1:07:16 forge I thought we did this Hilo
1:07:20 model or this 01
1:07:23 model let's do
1:07:29 yeah this is Mini Max let's do this one
1:07:31 as well
1:07:33 generate or did I do that already am I a
1:07:36 loser no I did Hanan all right so here's
1:07:41 this that one's pretty good here's let's
1:07:44 here's this one
1:07:56 that was
1:07:57 [Music]
1:08:06 really all right let's let's download
1:08:08 this
1:08:15 one where are you using the multiple
1:08:17 models at this is all within Korea
1:08:20 kore.ai has
1:08:23 got how many models now one two three
1:08:26 four five six seven seven different
1:08:30 models and then on the image side
1:08:31 they've got some crazy amount of models
1:08:34 as
1:08:47 well yeah we're just the tools right now
1:08:51 for for movie making like you've got to
1:08:54 be a real if you're a real OCD Movie
1:08:58 Maker and Storyteller and you want to
1:09:02 like take the time to learn which of
1:09:04 these models are good at what or or you
1:09:06 just want to take a bet on one and keep
1:09:09 prompting that one get better and better
1:09:10 at prompting that one
1:09:12 model that might be be the way to do it
1:09:15 I I don't know I don't I don't know a
1:09:17 good way to um to really do video
1:09:21 storytelling right now if you want it to
1:09:23 be not janky I I think you're just going
1:09:26 to have to deal with the Jank you're
1:09:27 going to have to embrace the Jank right
1:09:29 now um okay so we downloaded that let me
1:09:32 change my screen
1:09:35 share so that we can
1:09:40 do where are we going I'm going to go to
1:09:45 Runway where's Runway there's
1:09:49 Runway share
1:09:54 okay so let me trash this select an
1:09:58 asset upload
1:10:01 video there's that
1:10:07 open cancel
1:10:13 upload come
1:10:15 on there we go you
1:10:20 good there you
1:10:24 go okay
1:10:26 let's do this with no
1:10:31 prompt let's do it with
1:10:35 turbo because it's already pretty
1:10:37 good add a text prompt
1:10:41 um uh
1:10:47 train
1:10:49 rolling
1:10:50 uh
1:10:54 forward generate
1:11:03 oh here we've got some good video all
1:11:05 right well so
1:11:07 while oh that's going pretty quick oh
1:11:09 that's turbo Runway turbo is pretty
1:11:11 quick look at
1:11:13 that that's nice
1:11:34 ah that of the face
1:11:47 [Music]
1:11:55 that's kind of cool it took a Runway
1:11:58 video I I just put train rolling forward
1:12:01 I didn't give it a style or anything so
1:12:03 now let's go and change this to
1:12:06 4K see how quickly it'll upload that
1:12:09 yeah it's Thomas the Train it's totally
1:12:11 Thomas the Train that's really
1:12:16 [Music]
1:12:17 cool wheels are rolling
1:12:20 [Music]
1:12:27 I like it turbo almost always means Jank
1:12:30 oh yeah let's go let's go do this thing
1:12:33 with not turbo and see see what it it
1:12:36 does same same
1:12:41 prompt okay now here's the 4K one that
1:12:44 upscaling to 4K is pretty [ __ ]
1:12:47 quick can I zoom in on
1:12:51 this uh no
1:12:55 [Music]
1:13:10 I got to say the quality of that's
1:13:11 really quite
1:13:15 good it's somewhere between realism and
1:13:18 a cartoon
1:13:22 but pretty slick godd damn
1:13:26 we're anyone else notice we're in a
1:13:28 really [ __ ] wild time in
1:13:35 history yeah the smoke is good the sound
1:13:38 is good the the Thomas the face is
1:13:43 good that's [ __ ] hilarious all right
1:13:46 let's see what
1:13:47 the what the the non-alpha version does
1:13:52 oh wait did I did I pick Alpha oh yeah
1:13:55 Alpha's the good one I didn't do turbo
1:13:58 yeah so Alpha so this one should
1:14:00 be I I mean what's fascinating is that
1:14:03 it it it
1:14:05 kept so much but so much is
1:14:10 different so
1:14:12 bizarre how these things
1:14:20 work I gotta say like
1:14:27 yeah just the the the workflow the
1:14:29 workflows right now you're going to have
1:14:31 to get very specific just basically pick
1:14:33 a tool here it added a face to this one
1:14:35 too that is the most bizarre thing I've
1:14:37 ever
1:14:40 seen so this is the non-turbo version
1:14:46 [Music]
1:15:02 that Thomas spent too much time with
1:15:04 Ringo
1:15:06 Star and it won't get any worse right
1:15:09 yeah well we'll see five minutes ago it
1:15:13 was the jankiest time I know crazy all
1:15:15 right let's go let me flip back to uh to
1:15:20 Kaa because I used some of those
1:15:22 different models with the blacksmith
1:15:25 and they're [ __ ]
1:15:32 good all right that's the one we saw
1:15:38 before this did it it did it as a square
1:15:41 video which is bizarre but this one's
1:15:43 really good so let's add sound to this
1:15:48 one so this was Halo so this is minia
1:15:52 Max minia Max is the one that's got the
1:15:55 the best physics other than vo but I
1:15:57 don't have access to vo yet vo looks
1:16:00 quite
1:16:08 [Music]
1:16:10 amazing that's good except you don't
1:16:12 have any sound of the
1:16:15 fire I wonder if I can can you kill the
1:16:18 audio
1:16:26 you can't oh that's [ __ ] up once you
1:16:29 add the audio you can't you can't undo
1:16:33 it there's some
1:16:37 janky all right let's add audio to this
1:16:44 one want me to put some prompts in vo
1:16:47 yeah Vicki let me grab let me go grab
1:16:50 the uh let's
1:16:53 grab we'll grab the train one steam
1:16:58 locomotive
1:17:00 copy let me go put them in notes for a
1:17:04 second hang on a
1:17:06 sec paste and then let me grab
1:17:10 the
1:17:14 blacksmith paste and then let me grab
1:17:25 this
1:17:27 one um what's the best way to get these
1:17:30 to you Vicky I oh I'll just DM you over
1:17:34 in the salon
1:17:37 [Music]
1:17:42 [Music]
1:17:56 ding okay you've got three three prompts
1:17:59 the steam locomotive one the blacksmith
1:18:01 forging a sword and the pirate ship
1:18:03 caught in the
1:18:05 whirlpool for the ones that I rendered
1:18:07 that pirate ship in the whirlpool was
1:18:09 absolutely horrific um the trains were
1:18:13 all over the
1:18:15 place um let's see what this sounds
1:18:22 [Music]
1:18:23 like that's the video is bad the audio
1:18:26 is
1:18:29 bad um this one was pretty
1:18:32 good wasn't
1:18:36 it why do I not have audio on this
1:18:41 one that's weird oh there we
1:18:49 go see there you hear the this one you
1:18:51 hear the fire going
1:19:08 all
1:19:09 right let me come back Vicki's goingon
1:19:11 to make us some videos so we'll see what
1:19:13 those look
1:19:14 like I'll post in the regulars once
1:19:17 they're done
1:19:22 okay Brandon's bitter he wants me to do
1:19:24 podcast pods in jelly pod which I am
1:19:27 going to do but he's not here next week
1:19:30 so I'll probably do them next week when
1:19:32 he's not
1:19:33 [Laughter]
1:19:35 here oh man I must be having fun time is
1:19:40 just evaporating I feel the same way um
1:19:43 and not in
1:19:46 Stereo Ohio State one I'm I'm a Penn
1:19:49 Stater Penn State lost yesterday so oh
1:19:52 well and then I I get the play of
1:19:55 watching the Broncos lose to the Buffalo
1:19:57 Bills uh this Sunday so look at my hair
1:20:02 that's fantastic any no one no one
1:20:05 commented on the hair come on we're but
1:20:10 trolls you know what a hammer looks like
1:20:13 right uh yeah of course I've definitely
1:20:15 seen a hammer before yeah this isn't the
1:20:17 hamiest of
1:20:19 hammers it's got kind of a sword handle
1:20:26 [Laughter]
1:20:29 it's a creative
1:20:38 Hammer you know what the thing is
1:20:43 is with all
1:20:46 this in this 5sec video there's like a
1:20:50 second and a half of usable video
1:20:56 we just got to keep making
1:20:59 [ __ ]
1:21:00 crazy but I'll tell you what once they
1:21:03 start once they start combining um
1:21:05 persistent 3D
1:21:08 worlds and then the ability
1:21:11 to create shots within a persistent
1:21:14 world so so right now we're looking at
1:21:16 this this corner of this Workshop where
1:21:18 the forge is and the Anvil is but if I
1:21:22 if I did another shot here um it it
1:21:25 would completely change everything right
1:21:28 but when they start getting the
1:21:30 locations to be consistent and
1:21:32 persistent just like the characters this
1:21:34 is g to get good really fast it's gonna
1:21:37 get good really fast all right
1:21:41 um so two weeks ago
1:21:45 now we did this thing called are you uh
1:21:48 uh AI festiv us AI for the rest of us
1:21:52 and
1:21:53 um if you look up screen right now
1:21:55 Brandon put up the um the URL are you
1:22:00 ready for AI so we're putting together
1:22:03 Vicky and Ann Murphy and I and and some
1:22:06 other folks are
1:22:09 taking some of the content that was
1:22:11 created during AI festiv us and we're
1:22:15 putting together a training for AI
1:22:18 Readiness um our Target launch date is
1:22:22 February 1 if you want to sign up
1:22:25 for that if you want to find out more
1:22:27 information about it as it comes
1:22:28 available go to that URL on screen are
1:22:31 you ready for ai.com let me push this
1:22:35 forward for the ticktockers so you can
1:22:37 see it there on the
1:22:39 bottom
1:22:41 um and go check it out it's there
1:22:43 there's not much there right now just
1:22:44 sign up for it if if you're potentially
1:22:46 interested in uh a training course um
1:22:51 that that really prepares you for
1:22:56 being that sort of curious adaptable
1:22:59 adventurous person that's learning about
1:23:01 Ai and and constantly evolving and
1:23:05 adapting and things like that that's
1:23:06 that's what we're looking to teach it's
1:23:08 not about teaching the tools it's more
1:23:10 about teaching the
1:23:11 mindset um and that I'm pretty excited
1:23:13 about
1:23:14 so so there's that um okay Vicky is
1:23:19 going to be putting some videos in the
1:23:22 Irregulars channel so if you don't know
1:23:23 what the AI salon is go to this
1:23:27 URL the salon. and click on join our
1:23:32 community and that will take you
1:23:35 into um our community website built on
1:23:38 top of Mighty networks and if you scroll
1:23:41 down toward the bottom there's an area
1:23:43 called Community clubs here let me let
1:23:46 me share my
1:23:50 [Music]
1:23:53 screen so here here I am in in the AI
1:23:56 Salon there's like a little welcome um
1:23:59 track on the left hand side a seven-step
1:24:02 welcome there's also AI Festivus site is
1:24:05 there so if you missed AI Festivus if
1:24:08 you click on that Tab and go to the
1:24:09 resource list every one of the sessions
1:24:12 has a video attached to it so you can go
1:24:15 watch those videos and you can actually
1:24:17 ask questions and chat in those in those
1:24:20 sessions um and if you scroll down
1:24:22 toward the bottom there's an area called
1:24:24 community clubs and one of the community
1:24:26 clubs is called
1:24:27 Irregulars that's this Channel's uh
1:24:30 little private area ah here is some
1:24:34 V2 video that Vicky has put up all right
1:24:37 so here's the
1:24:47 train pretty good oh wow they did a cut
1:24:51 too is there is there consistency
1:24:55 that's fascinating hang
1:25:01 on yeah there's decent consistency
1:25:03 between the the the physics of the
1:25:06 wheels is still [ __ ]
1:25:08 up but it's better they're at least
1:25:11 rolling and the train is at least moving
1:25:13 forward the smoke looks very
1:25:17 real yeah vo's the [ __ ]
1:25:21 man the fact that it's cutting to three
1:25:24 different three different angles of the
1:25:27 same train and
1:25:29 retaining consistency that's pretty
1:25:37 impressive the the [ __ ] Wheels
1:25:39 physics
1:25:41 though I'll tell you what there's going
1:25:43 to be there's going to be specialty uh
1:25:46 special effects people that are just
1:25:48 going to fix the the shitty AI fuckups
1:25:51 in video
1:25:56 so anyway you can go see video Vicky's
1:25:59 videos down in there that's pretty cool
1:26:02 thanks for doing that Vicki that's
1:26:07 awesome beauty beauty beauty beauty all
1:26:11 right it's Friday night um normally I
1:26:14 give you some homework let me let me
1:26:15 think about what the homework could be
1:26:17 where's my
1:26:18 mouse trying to get the black bar going
1:26:21 all right
1:26:36 grock question mark there is a new
1:26:38 Standalone grock app so grock if you
1:26:42 don't know it is the
1:26:45 AI like chat GPT tool of Twitter X and
1:26:51 xai released a grock app on the iOS
1:26:56 store it's a standalone app right now
1:26:58 it's okay um apparently grock 3 is
1:27:01 coming
1:27:02 soon and
1:27:08 um the thing you need to be careful with
1:27:11 on iOS is that there's at least two
1:27:15 maybe three fraudulent grock apps make
1:27:19 sure you get the one from the X AI
1:27:22 company right x x . a um is the is the
1:27:27 company that put it out okay what do you
1:27:30 think of nvidia's little computer there
1:27:32 releasing um I think it is [ __ ]
1:27:35 staggering like for $3,000 I mean Nvidia
1:27:40 is absolutely killing it I I don't see
1:27:44 anyone catching them for some time um I
1:27:47 think their robot Suite is amazing I
1:27:50 think their consumer um 5,000 series
1:27:54 graphic cards are
1:27:56 amazing um I think they're they're
1:27:59 little
1:28:00 $3,000 desktop super computer that's
1:28:03 about the size of a piece of toast two
1:28:05 pieces of toast stacked together is
1:28:07 amazing so you get two of those things
1:28:10 and daisy chain them together and you
1:28:12 can run a 405 billion parameter AI model
1:28:16 locally on your
1:28:18 desktop [ __ ]
1:28:20 Bonkers uh so I think it's amazing khaki
1:28:23 Kyle can you remind us what happened in
1:28:25 the EU because they started regulating
1:28:28 so early for Monday um
1:28:32 sure so
1:28:35 um I I'll give you a specific a specific
1:28:39 little
1:28:41 uh clause in the regulation that they
1:28:44 did so the EU passed an AI bill
1:28:48 um two years ago they came out first and
1:28:52 they said you know AI is really
1:28:54 important and we're going to pass this
1:28:55 bill and and basically it's a bill that
1:28:59 prevents that that that
1:29:03 guards EU citizens against the risk of
1:29:07 AI and it and it was well it's it's a
1:29:09 well-intended bill right
1:29:14 but one of the Clauses in it
1:29:17 is um you can't you can't distribute an
1:29:22 AI system
1:29:25 that can measure
1:29:27 employees um
1:29:29 emotional uh response to something right
1:29:34 so you can't be on a zoom call for
1:29:36 example and use AI to measure someone's
1:29:39 emotions right are they lying are they
1:29:42 this are they that right so which on the
1:29:44 surface sounds like yeah like in in fact
1:29:46 when I read that particular Clause I was
1:29:49 like yeah that seems reasonable it like
1:29:52 yeah we we don't want to do that right
1:29:54 because if employers are using it to you
1:29:57 know overly what whatever control or
1:30:01 manipulate employees that's not a good
1:30:03 thing that said when you
1:30:10 [Music]
1:30:20 use where which one did I share hang on
1:30:23 oh this one
1:30:32 yeah oops new
1:30:37 chat hey Quinn what's going
1:30:40 down hey Kyle not much how's your day
1:30:43 going anything exciting on the horizon
1:30:46 it's going great I'm really
1:30:48 excited that's awesome what are you
1:30:50 excited about no now I'm really sad
1:30:55 I'm sorry to hear that Kyle want to talk
1:30:57 about what's on your mind no hey could
1:31:00 you
1:31:01 whisper sure I can do that what's on
1:31:05 your mind I want you to talk really fast
1:31:06 and really
1:31:08 loud all right I'll speed it up and talk
1:31:11 louder what's going
1:31:13 on uh talk loud and talk
1:31:16 fast okay I'm talking loud and Fast
1:31:19 what's on your mind give me a tongue
1:31:22 twister sure how about she sells
1:31:25 seashells by the seashore give it a try
1:31:27 make one up I want you to make one up
1:31:28 and I want it to all be with the letter
1:31:31 M all right how about Misty Mountains
1:31:35 make marvelous mining mirages give that
1:31:37 one a go no but I want it to be really
1:31:41 long got it how about many mischievous
1:31:44 monkeys mightily marched through the
1:31:47 M got it many mischievous monkeys
1:31:50 mightily marched through the Misty
1:31:51 Mountains making
1:31:54 many mischievous monkeys mightily
1:31:56 marched through the Misty Mountains
1:31:57 making marvelous Melodies while munching
1:32:01 mangoes many mischievous monkeys might
1:32:04 right okay now why did why did I do all
1:32:08 that because that was stupid
1:32:13 um the way that advanced voice
1:32:17 works is it's a true
1:32:20 multimodal um voice model and and what
1:32:24 what it means is
1:32:26 this with previous voice systems when
1:32:29 you would talk there were three
1:32:31 different models in play the first model
1:32:33 would take your audio recording and
1:32:36 translate it into text the model would
1:32:39 then send that text to a large language
1:32:42 model to ask the question the large
1:32:44 language model would respond in text and
1:32:47 then this third model would translate
1:32:49 from text to voice and then it would
1:32:51 talk back that's why voice inter
1:32:54 interaction
1:32:56 things before advanced voice were so
1:32:59 there was such a delay because it was
1:33:01 three different models it was like a
1:33:02 5-second delay from when you asked the
1:33:04 question to when it answered the thing
1:33:06 that I just did where she could sense
1:33:08 that I was getting more excited or more
1:33:10 sad and she was modulating her
1:33:13 interaction with me based on that is
1:33:16 because it's a multimodal model that
1:33:18 when you talk into it it not only
1:33:21 understands the words it understands the
1:33:23 emotional intent
1:33:26 that law in the EU thing means
1:33:29 that that feature is
1:33:33 illegal so what's so what's been
1:33:35 happening is this new technologies come
1:33:38 out mostly from open AI but from other
1:33:41 companies as well and they basically say
1:33:43 hey we just launched this new feature
1:33:45 it's available everywhere except the EU
1:33:48 and then what they're having to do is
1:33:50 they're having to to to negotiate car
1:33:53 outs the legislation for specific
1:33:56 Technologies so it's just it's one of
1:33:59 those things where when they wrote the
1:34:00 law it seem seemed person perfectly
1:34:03 reasonable at the time but they couldn't
1:34:05 anticipate what was coming right and
1:34:08 that's the problem with over legislating
1:34:10 early is that you may legislate yourself
1:34:13 in a corner the the other thing that the
1:34:16 legislators that I've talk to is quite
1:34:20 clear to me is they're totally clueless
1:34:23 about what this technology is what it
1:34:25 does that there's a difference between
1:34:28 generative Ai and traditional AI they
1:34:30 don't it's all AI to
1:34:32 them and all they know is what they've
1:34:34 been told the robots are going to kill
1:34:36 us everyone's afraid of it it's going to
1:34:37 take all the jobs and so their mindset
1:34:40 is this negative mindset of AI is a risk
1:34:46 and what they want to do is they want to
1:34:47 get their name stamped on legislation
1:34:49 that says see I was there to protect you
1:34:52 from the evils of AI
1:34:54 and there's very few legislators right
1:34:56 now saying hey let me get educated about
1:34:59 this because I want to make sure that we
1:35:00 can legislate to protect the
1:35:04 Innovation that all of these tools are
1:35:07 allowing us all to tap into right so
1:35:09 that's that's what that's what's
1:35:11 happening with that European legislation
1:35:13 the EU AI legislation just
1:35:16 crazy all right your AI overlords will
1:35:20 remember this Kyle
1:35:24 blacksmith posting now but it doesn't
1:35:26 like the pirate ship prompt all right
1:35:28 let's go look at let's go look at
1:35:33 uh can
1:35:36 I if I change no it's not GNA go hang
1:35:42 [Music]
1:35:49 on go to here new
1:35:52 activity blacksmith
1:35:59 yeah that's what it should
1:36:02 be well that's a little [ __ ]
1:36:05 up but again you know you can you can
1:36:11 use that's pretty
1:36:13 good until right
1:36:19 there you could cut you could cut back
1:36:21 to that like you can cut to you know
1:36:25 right there that one's good like that
1:36:28 that you could use you could use that
1:36:30 hammer strike I mean there's something
1:36:32 about this if you ever watched Alex
1:36:34 Wright who's the blacksmith on he was on
1:36:37 Tik Tok and YouTube for a while um I
1:36:39 always thought it was weird he held the
1:36:41 the
1:36:42 hammer um short like this he didn't hold
1:36:46 it way down at the end he held it here I
1:36:47 guess to control the amount of
1:36:50 deoration
1:36:52 um this seems pretty [ __ ] accurate
1:36:55 the like the physics in vo are
1:36:59 crazy right you see the little things
1:37:01 flying off it you got the fire going
1:37:04 here these things are consistent as P
1:37:07 put put it before you do know what a
1:37:08 hammer looks like right this looks like
1:37:10 a hammer looks like a blacksmith's
1:37:14 Hammer those look like that that's just
1:37:16 garbage but whatever look at the tools
1:37:19 in the background they look right
1:37:24 yeah vo's the [ __ ] man vo is the [ __ ]
1:37:30 all right I'm going to get out of here
1:37:32 let you people get out of here and start
1:37:34 your weekend I don't think I have any
1:37:37 homework for you I think what we're
1:37:38 going to do
1:37:49 is oh no I do have homework for you I do
1:37:53 have homework for you
1:37:56 you as most of you know and if you don't
1:38:00 know you know you'll know now we needed
1:38:03 a thousand subscribers on the YouTube
1:38:06 channel to be able to do business stuff
1:38:08 there we also it turns out need I think
1:38:12 it's a thousand hours of
1:38:15 views so your homework this weekend is
1:38:19 go to the AI learning laab YouTube
1:38:21 channel learning la- a if you haven't
1:38:24 subscribed Please Subscribe but we're
1:38:26 over a thousand now so thank you and
1:38:29 just hit play and and play uh like a a
1:38:33 string of I think you can go into a
1:38:34 playlist and just have it play one after
1:38:36 the other and just hit play and then
1:38:39 leave you know especially if you're on
1:38:41 Wi-Fi and it doesn't cost you extra data
1:38:43 I don't I don't want you to run data
1:38:45 you're welcome to watch the videos but I
1:38:48 just need some play hours we're actually
1:38:50 very very close according to Brandon I
1:38:51 don't I haven't looked recently um but
1:38:54 but just go play some AI learning Lab
1:38:57 videos on YouTube and that'll put us
1:38:59 that should put us over the edge to be
1:39:00 able to do um to do business there
1:39:04 Brandon said we're very close that's
1:39:05 great okay
1:39:07 cool um thank you all so much um I hope
1:39:11 you had a good time tonight uh it's it's
1:39:14 kind
1:39:15 of video's a bit of a tough one to demo
1:39:19 because it's time-based media and it
1:39:21 takes a lot of time to generate those
1:39:23 things they're all still so janky um but
1:39:27 the tools are getting better what I can
1:39:29 tell you is this it looks like that
1:39:32 what's coming from Luma Labs is going to
1:39:35 be in the neighborhood of Minx and vo I
1:39:40 I would say right now um V2 is in the
1:39:44 clear
1:39:45 lead Halo Mini Max is number two and
1:39:49 then I would sort of put on another tier
1:39:51 down Runway which is good but not great
1:39:55 um and then a tier below that right now
1:39:57 is Luma laabs dream machine but that's
1:40:00 got some other things going for it but
1:40:01 they have a new model coming so I'm
1:40:03 really excited about their new
1:40:05 model
1:40:08 um but yeah I I would
1:40:11 say start thinking this weekend
1:40:16 about this is gon to be [ __ ]
1:40:20 up think about all of the stuff that
1:40:22 you've put time in learning over the
1:40:24 past two
1:40:26 years what if something happens in 2025
1:40:30 where the AI gets so good that all of
1:40:33 those skills that you've amassed become
1:40:36 irrelevant and I know that sounds
1:40:38 depressing right and they won't be
1:40:40 irrelevant you'll still be able to use
1:40:42 them but you won't have to because
1:40:44 there's going to be agents or there's
1:40:46 going to be something that's so good
1:40:47 that you no longer have to prompt like
1:40:49 you used to have to prompt so I would
1:40:51 say start thinking about what it means
1:40:53 means to be in a constantly
1:40:57 evolving
1:41:00 um technology
1:41:02 environment where all of our ability to
1:41:05 succeed is is I think going to be
1:41:08 largely based on our ability to adapt
1:41:11 and and stay curious so maybe the the
1:41:15 homework assignment this weekend then is
1:41:19 get curious and try something new if you
1:41:21 haven't written a program with lovable
1:41:24 dodev do that if you haven't made a
1:41:26 movie with one of these video things do
1:41:28 that if you haven't tried LTX Studio do
1:41:31 that if you haven't made a song in uh
1:41:34 sunno or yudo do that if you haven't
1:41:37 analyzed a spreadsheet and chat GPT go
1:41:40 do
1:41:40 that or have a 4 hour I don't know how
1:41:43 long you can have have a 15minute
1:41:45 conversation with h advanced voice in
1:41:48 chat
1:41:49 GPT um while you're driving to the store
1:41:52 and just if you haven't done that just
1:41:53 maybe to get curious this
1:41:56 weekend as what Brandon says if you
1:41:59 haven't made a podcast with jelly pod go
1:42:02 do that which by the way you can get
1:42:05 access to jelly pot as an AI Salon
1:42:07 member within the AI Salon where where
1:42:09 within there can they find that
1:42:11 information Brandon okay here's the
1:42:13 pirate ship let me show the pirate ship
1:42:15 since Vicky made it we'll go
1:42:23 Shameless self-promotion okay nons salon
1:42:27 and Shameless self-promotion is where
1:42:30 the information is about jelly pod all
1:42:35 right and if you don't know what jelly
1:42:37 pod
1:42:38 is you design a voice by describing a
1:42:41 character and how you want it to talk
1:42:43 and it will make you a voice based on
1:42:45 your description and then you can put
1:42:47 multip multiple of those in a podcast
1:42:50 and give it a topic and it will just
1:42:52 make you a podcast with your characters
1:42:54 in it it's pretty slick we'll play with
1:42:56 it next week when Brandon's not here so
1:42:58 we can make fun of him no screen share
1:43:00 up I know I'm putting it up uh here we
1:43:03 go here we
1:43:05 go all right put this here we'll put
1:43:09 this here we'll put this
1:43:12 here see that's
1:43:17 good that's pretty good
1:43:20 vo vo's where it's at man
1:43:27 yeah nothing we did I mean this is it's
1:43:30 it's a little over drama dramatized with
1:43:33 the with the lightning but it's still
1:43:34 not bad it's pretty
1:43:36 cool squat
1:43:39 squat thank you for doing that Vicki
1:43:41 that's super
1:43:43 cool all right everybody I'm gonna get
1:43:46 out of here Pace out hope you all have a
1:43:49 fantastic weekend go play with some [ __ ]
1:43:52 that's it be adaptable start thinking
1:43:54 about your adaptability
1:43:56 quotient how adaptable are you think
1:43:59 about ways to increase that all right
1:44:02 it's G to serve you I promise pretty
1:44:05 cartoony though yeah it's this is a
1:44:07 little
1:44:07 cartoony um all right everyone on
1:44:10 YouTube X and uh what's the other one
1:44:15 LinkedIn peace out have a fantastic
1:44:17 weekend and I will see you Monday