10/13/2025 - A Deep Dive into OpenAI's Sora 2 Prompting Guide and Building a Custom GPT Helper

Description

Chapters

Transcript

0:01 [Music]
0:02 Come here, buddy.
0:06 [Music]
0:18 [Music]
0:19 [Applause]
0:22 [Music]
0:50 Oh.
0:52 [Music]
1:07 [Applause]
1:11 What's up, Dr. J? What's shaking? What's
1:13 going down?
1:16 [Music]
1:33 Oh, open up your B and see like me.
1:38 [Music]
2:04 in a westerly
2:07 direction.
2:08 [Music]
2:11 This car is my train.
2:16 I've been driving. I've been wondering
2:20 what it is I'm running from again.
2:24 [Music]
2:26 Feel like an idiot old man
2:30 holding on to 29
2:34 and up on that horizon
2:39 is a California line.
2:47 [Music]
2:50 Up ahead of trucks carrying a wide load.
2:54 Preab house cut in half.
2:58 Cute little front door into windows. My
3:00 loss.
3:02 Ain't sure the cry should last.
3:06 You see, I broke a home up myself once.
3:10 As I stumbled to that door,
3:15 I read a note by the dawn light.
3:19 Said, "Don't you come around here
3:24 anymore."
3:27 Well, I've had enough.
3:29 [Music]
3:31 This fre
3:33 [Music]
3:35 was good with decisionions.
3:38 At least that's what I've been told.
3:41 [Music]
3:49 Oh,
3:51 is it meltdown Monday? Is it meandering
3:53 Monday? Is it hyperfocused Monday? We
3:56 don't know. And you know when we're
3:59 going to find out? I don't know.
4:00 Whenever I decide to stop playing this
4:02 stupid thing.
4:06 Good evening, good people. was shaking.
4:11 My Tik Tok feed just repeated that door
4:14 about 20 times. Oh, that's funny.
4:19 Tom Nodler,
4:21 so many simultaneous
4:24 something YouTube
4:27 let me have open at once.
4:30 [Music]
5:06 What's happening, Mr. It?
5:08 [Music]
5:17 So, it sounds like the uh she leads AI
5:20 create conference was a success.
5:23 Many uh
5:26 many meetings were had, many
5:28 presentations were presented,
5:30 relationships formed.
5:32 [Music]
5:47 [Music]
6:13 Tom Matler. It's been so long. I don't
6:15 think I've lost my irregular status.
6:17 Here's the deal about irregular status.
6:20 If you have ever been an irregular,
6:23 it's not about the label. You're just
6:25 irregular.
6:32 There are a lot of things in this world
6:33 you lose membership to. Being an
6:35 irregular is not one of them. Once an
6:38 irregular, always an irregular.
6:41 Dr. J. Dr. Jay actually confirms that
6:45 that is uh that has been clinically
6:47 confirmed. hurts.
6:49 [Music]
7:12 That's probably hurting people's ears.
7:18 it, you know, I went to that AI learning
7:20 lab and it was I I mean the fellow was
7:23 he was interesting. Um I I didn't I
7:27 didn't really learn anything per se, but
7:30 uh I I suppose learn he seemed like he
7:34 was taking a guitar lesson uh rather
7:37 than intelligence.
7:41 Um, so I left. I So I I I'm sure he
7:46 means well. Yeah. And so may maybe he
7:48 was a substitute teacher. It was
7:50 possible he was a substitute teacher.
7:54 [Music]
8:00 A serious eater. Okay, that's a that's a
8:04 name I can get behind.
8:07 Welcome, Serious Eater.
8:12 Sorry, I'm losing it. Hey, you know what
8:14 I just realized? No, Kyle, tell us,
8:16 please. Um, I don't have anything to
8:18 drink. I shall return. Uh, hold, please.
8:35 I shouldn't
9:17 What are you doing? Are you coming,
9:20 champ?
9:22 Oh,
9:22 >> he's getting water.
9:27 >> Come on.
9:32 Oh, good. Come on.
9:35 [Music]
9:44 >> Okay,
9:45 we're back. And I can hydrate,
9:49 which is important. Apparently,
9:52 if you live in Colorado, that becomes uh
9:55 becomes part of your job.
9:57 Ah, how's everybody doing? It's a It's a
10:02 It's a small but appreciative crowd
10:04 tonight.
10:16 Oh man.
10:19 Um, I want to show you some new stuff in
10:21 the AI salon. We're uh we're revamping
10:27 I can continue to hold my head high.
10:31 I hope you've been well, Tom. I hope all
10:33 all's going well.
10:36 Tobias Tobias with the uh irregular
10:40 appreciative. Very nice.
10:45 [Music]
10:47 Tell me what you guys want to do
10:48 tonight. If you want to look at [ __ ]
10:50 play with [ __ ]
10:52 talk about [ __ ]
10:54 you tell me what you want, and then I
10:57 will do my best to ignore it entirely.
11:01 All right. If anyone's new here, please
11:04 feel free to say hi, ask questions.
11:07 Sora,
11:09 Sora 2, we could go play with Sora, too.
11:12 I think OpenAI just put out a Sora 2 um
11:17 guide.
11:19 So maybe what we do, let me see. Sora 2
11:24 guide open AI. We could Sora 2 prompting
11:29 guide.
11:35 See, we could go we could go look at it.
11:43 But but then the better thing to do
11:48 Oh, you know what we could do tonight?
11:50 This could be fun. We will
11:54 we'll go look at the Sora 2 prompting
11:57 guide from OpenAI. We'll we'll look at
11:59 it. We'll read through it a little bit.
12:00 I'll comment on it. I'll probably make
12:02 fun of it or I'll or I'll put myself
12:06 down. I never prompt like that because
12:08 I'm a loser. One of those. I'll put them
12:10 down or I'll put me down. Can't be
12:12 neutral.
12:14 And then
12:17 we'll take that content and we'll go
12:21 over to uh to chat GPT
12:25 and we'll turn chat GPT into a Sora 2
12:28 prompting engine. And then you can put
12:31 in your lazy shitty prompt
12:34 and uh
12:37 and then it'll turn into a good prompt.
12:40 [Music]
12:55 [Music]
13:07 Freedom came my way that night
13:11 [Music]
13:14 just like a jet plane in and out of
13:18 sight.
13:19 I was hauling ass at a million miles an
13:23 hour. Yeah, I can I can see what you're
13:25 writing. Uh, it is a custom GPT I was
13:27 talking about.
13:32 I'm looking right at it as I'm singing.
13:34 I'm just like noting the things you're
13:36 writing. I should probably acknowledge
13:37 you once in a while.
13:40 [Music]
13:43 I can't open Sora in my laptop.
13:47 Are you in the US? It might be a a a
13:54 If you're in the US, then you just need
13:56 to go to sora.hatgpt.com.
14:07 Well, sore is a re resource hog. Oh,
14:10 that their website their website right
14:13 now, Mr. is horrible. It just crashes
14:16 all the time.
14:19 laugh so far. Yes. They did not get me.
14:22 Oh, that's right. That's right. You're
14:24 You're always like, "I'm still here."
14:28 Yeah. Well, I shouldn't be laughing.
14:33 Okay. Um Yeah. So, we can go do that. Um
14:37 I do want to show some uh
14:40 some of the new things in the in AI
14:42 Salon Mastermind.
14:46 [Music]
15:09 Th blue telescope
15:11 [Music]
15:13 looking at the world tonight. through a
15:16 blue telescope.
15:18 Wish I may wish I might not see what I
15:22 see.
15:24 Sheet metal on sheets of ice.
15:29 [Music]
15:32 Looking through this blue telescope
15:37 down a moon struck a road tonight.
15:41 [Music]
15:47 [Music]
15:55 [Music]
16:04 All right, let's get going. Let's get
16:06 going. Let's everybody get going. Let's
16:09 everybody get going.
16:13 All right. So, what I would love to turn
16:18 your attention to
16:20 is uh go on over to the AI salon
16:28 uh community website which is
16:30 community.thesalon.ai
16:32 or you can just go to the salon.ai and
16:34 say join our community if you're not
16:36 already a a community member. Um,
16:40 and we got some some all sorts of new
16:42 stuff going on here. So, we have a a
16:45 much cleaner experience and I'm I'm
16:47 quite excited about it. Um,
16:51 so
16:53 we've talked for a long time about the
16:54 cycle of AI readiness and it it was play
16:57 first, mindfully create, generously
16:59 lead. I've talked for the past I don't
17:02 know two months or so, month and a half,
17:04 two months that I feel like we're
17:06 entering a new arena when it comes to AI
17:10 that that it used to be enough to just
17:12 be curious about it. And what we're
17:15 starting to see now is we're starting to
17:16 see and you know we see a lot of them in
17:18 here is we're starting to see what I
17:21 would call AI professionals. So
17:22 professional, you know, people that are
17:25 using AI in a in a powerful professional
17:29 way. They're professionalizing their AI
17:32 game. Um, they tend to use lots of
17:35 tools. They tend to think of of AI, they
17:38 prompt AI like a producer where they're
17:40 the producer who's got the idea and
17:42 they're holding the definition of what
17:44 good looks like and then they use lots
17:46 of different tools to do that.
17:48 So we've updated that second stop in the
17:51 cycle instead of mindfully create to
17:54 create excellence.
17:56 So play first. So play with AI, figure
17:59 that [ __ ] out.
18:02 And then create excellence is once you
18:04 learn what some of these tools are, then
18:06 start to think about, you know, how do I
18:08 what do I want to do? Who do I want to
18:10 talk to? How do I want to impact them?
18:13 And then hold that vision for what good
18:15 looks like? and then figure out all the
18:18 different tools you need to duct tape
18:19 together to get to that vision. That's
18:21 create excellence.
18:23 And then generously lead is about
18:26 sharing what you're learning along the
18:28 way with others, sharing with others,
18:30 connecting with others, learning from
18:31 others, giving other people the
18:33 opportunity to lead. Right? Part of
18:35 generous leadership is giving. Part of
18:38 generous leadership is asking asking for
18:41 help because that gives someone else the
18:44 opportunity to generously lead to lead
18:47 you to give you something back. Um
18:52 and and when you do all of this and you
18:55 keep repeating that cycle, there's a
18:57 point at which you just your mindset
19:01 shifts and you kind of understand that
19:04 there's not much stopping us from from
19:06 achieving anything we want to achieve.
19:08 Right? This is AI when used properly,
19:11 when it's when you use it as as a a why
19:15 amplifier. What's your why? Right? You
19:18 know, um uh who is it? Simon Senic. Is
19:21 it Simon Senk? It's like it's all about
19:23 your why. AI your why, right? Why are
19:27 you doing this? What do you want to
19:28 achieve? And put that in there. Yeah.
19:30 Start with why is Simon Senic, right?
19:34 So, I I'll I'll bastardize that. Start
19:37 AI with Y or AI your Y, right?
19:41 Um, so anyway, so when you come to the
19:44 AI Salon website now, you'll see that
19:46 new graphic and we've got a new
19:47 explanation. So play first joyfully
19:50 exploring generative AI, gaining the
19:52 knowledge and inspiration to create
19:54 excellence by knowing what you want and
19:56 staying committed to the fidelity of the
19:58 idea, unleashing you to generously lead
20:02 others, enriching lives, redefining
20:03 what's possible, and unlocking a new era
20:06 of human achievement. So that's what the
20:08 salon's all about. If you haven't been
20:10 there, if you haven't been here, um
20:12 we've got a cool little uh new
20:15 reorganization.
20:16 And so we start with start your
20:18 adventure, which is now more simple. It
20:19 used to be seven steps, it's down to
20:21 five. We're probably going to try to get
20:22 that to three. We want to really tighten
20:24 up this experience.
20:26 And then we've got events and community
20:29 feed, right? So, so if you want to learn
20:32 about our events, it's right at the top.
20:35 The community feed, for those of you
20:37 that are not new here, the community
20:40 feed actually used to be the feed inside
20:43 of water cooler, but people didn't know
20:45 about it because nobody knows that
20:48 there's extra tabs. So, we broke it out
20:50 as a separate community feed. So,
20:52 tonight as you share stuff, I want you
20:55 to share it in the community feed. Okay?
20:58 Water cooler is still water cooler,
21:00 right? That's just the the ongoing chat.
21:02 You can just chat and ask questions and,
21:04 you know, troll each other, do whatever
21:06 you do there. But the community feed is
21:08 now broken out. So, anything you want to
21:11 share, anything. The whole the whole
21:13 reason we're calling it that and why we
21:15 put it there is that this can be
21:16 anything. This is sort of the anything
21:18 goes section. You want to put a song in
21:20 there, you want to put some new piece of
21:21 technology in there, you want to put a p
21:24 a piece you work on and get feedback on
21:26 it, that's the community feed. Then we
21:28 got salon announcements. And then we've
21:31 reorganized this the the community into
21:34 three main areas. Play and create, learn
21:37 and grow, and then advocacy and policy.
21:41 We've got uh some some partnerships and
21:44 and uh self-promotion stuff in there as
21:46 well. Um but those three main areas are
21:49 going to be the main areas. So play and
21:50 create is where you're kind of learning
21:52 and playing and interacting with each
21:54 other. Um, we've moved Danielle's Mad
21:56 Lib Challenge uh up out of the
21:59 mastermind. If you see the little crowns
22:01 here, those are part of the mastermind
22:03 uh subscription. That's a subscription
22:05 area of the salon. The things with the
22:07 little AI, what are those? Stars or what
22:10 are they called? Sparkles. The AI
22:12 sparkles. Um, those are free areas. So,
22:16 lots and lots of free areas. And then
22:18 the things with the with the crowns are
22:20 the mastermind area. All right.
22:23 and uh more and more to come. But go
22:26 check out the new thing. And what I
22:28 would encourage you to do right now, why
22:31 doesn't why don't we do this? Everyone
22:33 who's on the call, we got 24 people on
22:36 LinkedIn, we got 19 people on Tik Tok.
22:39 Why don't we jump in there into the
22:40 community feed and just post something?
22:42 Post something you've made in the past
22:44 week or just go go into the community
22:46 feed and and give it give it some life.
22:49 Now, here's the deal. Here's why the
22:50 community feed's there.
22:54 One of the pieces of feedback we got is
22:55 that people joining the community,
22:57 they're like there were all these very
22:59 specific things to put thing. They
23:01 didn't know what to do. They didn't know
23:03 where to go. So the community feed is
23:05 just just like sort of Facebook's
23:06 community feed, right? It's just like go
23:08 put anything there and let people start
23:10 interacting with you. So so go jump in
23:13 right now and over the next 15 20
23:15 minutes just go put some stuff in the
23:17 community feed. All right. Beautiful.
23:20 Beautiful. Beautiful. Beautiful.
23:22 Fantastic,
23:23 Bob. Okay, let's move over
23:29 to Sora
23:32 and
23:34 see if we can get Mr. It hooked up here.
23:38 But let's also I'll get hooked up here.
23:41 There's Vicki. Look at that. Vicki,
23:45 what's happening? Um,
23:49 so if you haven't been to the Sora
23:51 website, it's sora.chatgptt.com
23:56 and then I'm going to jump down to my
23:58 little area.
24:02 At the bottom bottom on the left is your
24:04 profile.
24:11 >> Hey man, you got that good CBD, man? For
24:13 sure.
24:15 Hey man, you got that good CBD, man?
24:17 Join the AI salon.
24:21 Join the AI salon and have some CBD.
24:26 Okay, so
24:29 here's Sora and and I can we we'll we'll
24:33 talk about that in a second, how to how
24:36 to do it, but I think we're going to go
24:38 over to just making stuff up. We need
24:41 tappers.
24:46 the twinkle, the AI twinkle. Yes, the
24:49 twinkle, the sparkle, twinkle, sparkle,
24:51 whatever it's called.
24:54 But I'm I'm I'm really excited about So
24:57 Andy Scarantino's been talking to a lot
24:59 of people including Danielle and Vicki
25:01 and Brandon and Liz and myself and a
25:04 bunch of other people. She's been
25:05 getting all sorts of ideas. And then she
25:08 she comes in, she has a lot of ideas as
25:11 well. Uh the AI salon is the fourth
25:12 community. She's kind of rearchitected
25:15 and and so we're really tightening
25:18 things up and really looking at what are
25:20 the bottlenecks that were keeping people
25:22 from coming back and participating and
25:24 things like that. So we're going to
25:25 clean up the infrastructure. Then we're
25:27 going to start doing lots and lots of
25:28 things to get new people in and more
25:30 people in and people more engaged and
25:32 I'm super excited about it. So, with
25:35 that, let's head over to um the OpenAI
25:41 Sora prompting guide
25:44 and let's look we'll we'll look through
25:46 this together
25:48 and then I think what we're going to do
25:50 is we're going to turn this into a
25:52 custom GPT. Now, if you're impatient, I
25:55 know there's a lot of Neurospicy people
25:57 on here.
25:59 When I mentioned that we were going to
26:01 turn this into a co custom GPT 15
26:03 minutes ago probably means that you've
26:05 already gone there and built your own
26:08 which is fine. In fact, go share that in
26:12 the community feed if you have. Um, and
26:14 then there's probably also a lot of
26:16 existing custom GPTs based on this
26:19 document. So, so what we're going to do
26:22 here in about 15 minutes after we go
26:24 through this thing is uh is not novel.
26:27 I'm sure lots of people have done it.
26:29 Um, but um, if you're if you're new to
26:33 chat GPT and if you're new to Sora, this
26:36 might be a good a good exercise to just
26:39 kind of pay attention, sit back and
26:41 enjoy.
26:43 All right, so before you prompt, think
26:44 of prompting like briefing a
26:46 cinematographer who's never seen your
26:48 story board. If you'll leave out
26:50 details, they'll improvise and you may
26:52 not get what you envisioned. By being
26:55 specific about what the shot should
26:58 achieve,
27:00 you give the model more control. Okay.
27:02 Yeah. Yeah. All right.
27:05 API parameters. Okay. So, this is this
27:07 is largely for the API, but that also
27:10 means that um
27:14 that this will some of these things will
27:16 work for non-API. The prompt controls.
27:19 Okay. So, model Sora 2 or Sora 2 Pro. I
27:24 think Sora 2 Pro is only available via
27:27 the API,
27:30 but maybe if you have a Pro account, you
27:32 have access to this as well in in the
27:34 Sora app. I don't know. It makes sense.
27:39 You can you can define the length. The
27:42 default value of a Sora video is 4
27:46 seconds, but you can defi define it as
27:48 4, 8, or 12. That's cool. Video res
27:52 resolution.
27:55 Okay. Video length
27:58 prompt anatomy that works. A clear
28:00 prompt. Okay. So, what's in the an
28:02 what's in the prompt? Shorter prompts
28:05 give the model more creative freedom.
28:08 Expect surprising results.
28:23 I saw I saw OpenAI
28:26 bought more data today for for Sora 2.
28:29 You mean more compute?
28:32 They said it was growing faster than
28:33 Chat GPT. Yeah. Um
28:37 the other thing I heard Sam Alman say
28:39 today is that um the rights holders,
28:44 this makes perfect sense. the rights
28:47 holders when when you when you copy
28:50 someone's text they get pissed or you
28:54 know when you recreate someone's text
28:55 they get pissed when you recreate
28:58 someone's video intellectual property
29:00 they get really pissed he said there's
29:03 there's a level of passion amongst the
29:05 rights holders uh for soru that is that
29:08 is shocking them as well but it also
29:11 makes sense they
29:17 What always surprises me is that OpenAI
29:20 says that
29:22 new adoption numbers surprise them.
29:26 Um, they're the industry leader. People
29:28 wait for them to do stuff and they keep
29:31 releasing highquality models. Even the
29:34 GPT5 release, which initially sounded a
29:37 bit like a flop, right? Like people were
29:40 like, "You took away my GPT4. I don't
29:42 like it. I don't like GBT5. Well, GPT5
29:45 is really good. They had a they had a
29:46 shitty routing model initially.
29:49 Um, but it's a really good model, but
29:52 they keep putting out
29:54 state-of-the-art
29:56 um products, features, so people are
29:59 going to use them. They have a weekly
30:00 user base. I don't know if you heard
30:02 this, their weekly user base has has
30:04 gone to um 800 million users.
30:09 It took the worldwide web I think 13 or
30:11 14 years to get to 800 million users. It
30:15 took Chat GPT a single company
30:20 two years three two and a half years
30:23 going on three years
30:26 800 million users the worldwide web it
30:29 was 13 years the the entire internet 13
30:34 years to get to 800 million users.
30:38 So that should put it in a little bit of
30:39 perspective.
30:41 Um,
30:44 and yeah, and and then on top of that,
30:48 they didn't launch Sora as a
30:50 professional tool. They launched Sora as
30:53 a meme maker, a meme generator, and they
30:56 launched a little Tik Tok competitor
30:58 where you can make funny videos of your
31:00 friends. So yeah, people are going to
31:02 use it. Sam, why is it a surprise?
31:09 Come on. Come on, people.
31:13 Pay attention to what you're doing.
31:17 You You put a consumer toy maker out
31:20 there, people are going to play with the
31:21 toy.
31:25 H
31:27 level four fan. Uh as Sam's mother, he's
31:31 draining my bank account for you people.
31:33 Exactly.
31:35 Um, oh wait, we should go let's go make
31:37 a Sora video. Hang on. We got to go back
31:39 to Sora for a second. Okay, we're gonna
31:42 we're gonna make a video here. Okay,
31:45 wait. Share this tab and then I'm going
31:48 to go Okay, I'm going to go uh Sama,
31:53 let's see. Sama
31:58 sits
32:01 on
32:02 Oh no. Did this site just crash?
32:07 Oh my god. Could you please fix this?
32:16 Unbelievable.
32:18 Hang on. I got to reshare.
32:28 Listen.
32:29 Listen. I don't think you know how hard
32:32 it is to be at OpenAI right now. They've
32:34 only raised $600 trillion dollar.
32:41 Like, are they ever going to fix this
32:44 site? Is this is this like the new video
32:46 upload on X?
32:49 Okay, let's go. Let's go make a video.
32:51 So, we're going to make Okay. Sama
32:53 sitting on sitting
32:57 on the roof
33:00 of Open AI
33:04 headquarters,
33:08 crying about how
33:12 he didn't think
33:16 so many people would make meme
33:22 videos
33:23 with Sora 2.
33:28 And then we'll go at Kyle Shannon.
33:30 That's me.
33:33 Kyle Shannon says uh tells Sam
33:38 uh
33:42 let's see. tells Sam,
33:45 um,
33:52 I told you that
33:56 if you made a tick tock
34:01 tick
34:02 tock like app, they
34:06 would bury you.
34:10 And at Samma says,
34:16 okay, at Samma says, "You're right,
34:18 Kyle.
34:23 You always are.
34:26 Wait, black bar. Am I not sharing the
34:29 right
34:30 Oh, pin. Black bar. Pin.
34:35 Um,
34:36 there's no pin right now. I'll come back
34:39 to it. You always are. Okay. So, let's
34:43 go make let's go make a Sam Alman video
34:45 that he's surprised that people are
34:47 using Sora. Jason, they use Sora too as
34:50 psychology play to get more people to
34:52 adapt to AI video in my opinion. Yeah,
34:55 Jason, I totally agree. I and and I
34:58 think it was I think it was actually
34:59 quite a brilliant marketing move
35:02 that
35:04 every launch before this what people
35:07 were talking about was the evilness of
35:10 AI video.
35:12 It's producing AI slop. It's stealing
35:14 jobs. It's intellectual property theft.
35:16 All that sort of stuff. There was some
35:17 of that about the sor stuff because you
35:20 know people were making full Simpsons
35:22 episodes and things like that with it.
35:25 But what most people are talking about
35:28 is the silly little videos. And so I
35:31 think that what this app has
35:33 successfully done is is disarm
35:38 some of the haters and some of the
35:41 people
35:43 who maybe
35:45 conceptually hated AI video
35:49 because everyone told them it was awful.
35:51 When they saw this, they were like,
35:52 "Well, that's kind of fun. I'd like to
35:54 make a funny picture of my sister, a
35:57 funny video of my sister, right? So,
36:00 yeah, I think it's pretty pretty
36:02 brilliant ploy on their part. Okay. Um,
36:05 their their servers are totally cooking.
36:07 It's it's like this thing's a mess.
36:10 Okay, let's go back to Oh crap. Did I
36:13 just do that? Yes, I did.
36:15 sora.jpe.com
36:17 at gp.com
36:22 sora guide opening eye. Here we go.
36:32 [Music]
36:34 Why does he always sing like that? It's
36:36 very annoying. All right, a short
36:38 prompt. In a '9s documentary style
36:41 interview, an old Swedish man sits in
36:43 the study and says, "I still remember
36:45 when I was young." I think that's how
36:47 most people will prompt this thing. Like
36:50 that or less. Most people are going to
36:52 put shitty stupid little prompts in the
36:54 chat. GPT,
36:56 the prompt will likely work well. '90s
36:59 documentary sets the style of the video.
37:01 Old Swedish man sits in a study,
37:03 describes the subject and the setting in
37:05 minor detail, letting the model take
37:07 creative liberties, and says, "I still
37:10 remember when I was young." Describes
37:12 the dialogue. Sora will likely be able
37:14 to follow this exactly. All right.
37:18 Going ultra detailed for complex
37:21 cinematic shots. So again
37:24 because I've noticed something
37:30 because their servers are cooking and
37:33 they were surprised at how much people's
37:35 are using Sora. It looks like they've
37:38 they've set the default for for the Sora
37:40 app to 4 seconds.
37:42 But I bet if you prompt it right, you
37:45 can get it to go longer than that.
37:47 That's my guess.
37:49 Sora 2's been so much fun for me and my
37:52 siblings. We've been reenacting core
37:55 childhood memories. Th This is the
37:57 thing, Danielle. you're you're striking
38:00 on something that I think is going to be
38:04 it's a thing that really excites me
38:09 is that
38:12 one of the things that AI does
38:14 masterfully
38:18 if you're willing to learn how to use it
38:21 is that it takes your ideas and it
38:24 amplifies them. Right? It lets you get
38:26 the ideas out of your head and and into
38:30 the world.
38:31 For the past 3 years, in order to do
38:34 that, we've had to like hang out on this
38:37 channel and, you know, learn this tool
38:40 and that tool and learn these features
38:42 and learn prompting and learn all the
38:44 stuff, right?
38:47 We're transitioning right now where
38:48 tools like Sora 2
38:51 and and Sunno I would say does this with
38:53 music
38:56 are starting to make quality outputs
38:59 even though people call them AI slop.
39:02 What you're describing Danielle about
39:04 you and your siblings being able to put
39:06 yourselves in videos
39:09 in scenarios you remember as kids that's
39:12 full-on self-expression. Like a lot of
39:15 feature films are made because you know
39:18 think of uh Stephen Spielberg's the his
39:21 autobiography
39:23 film when he was a kid. What was it
39:24 called?
39:27 I don't know but like it's a memory of
39:30 of him being a childhood filmmaker,
39:32 right? And he did a whole feature film
39:34 about that. So filmmakers for years have
39:38 taken their childhood memories and
39:40 turned those into feature films. We're
39:42 entering an era where anyone can do
39:45 that. And right now it's 4-se secondond
39:48 videos and then it'll be 12 second
39:50 videos and then it'll be minute long
39:51 videos. And within two, three years you
39:55 and your siblings Danielle will be able
39:57 to make a feature film about that funny
40:00 story when
40:02 when little Danielle went hitchhiking to
40:04 the candy store, right? Or whatever it
40:07 was.
40:09 I think we're moving away from prompts
40:11 to agents. Yeah. The prompt itself um
40:14 goes off and does it. Yeah. Yeah. It's
40:16 it the prompt I think you're right,
40:19 Jason. I think what the what the prompt
40:21 becomes
40:24 is an idea.
40:27 You give the AI your idea
40:31 and then it goes and executes it. And
40:34 right now it's our job to hold that idea
40:37 in our head and articulate it well. And
40:40 then it's our job to sort of, you know,
40:43 massage it into shape. When these agents
40:46 get better and better and better and
40:48 better and and also when they get to
40:49 know us, think about this. I think it
40:51 was
40:54 Mark Andre, one of the one of the big
40:56 tech guys. I think it was Mark Andre was
41:00 talking about
41:03 his 8-year-old kid. I think he's got an
41:05 8-year-old kid.
41:08 By the time he's like 22 or whatever,
41:13 AI is going to
41:15 have been guiding him and learning about
41:18 him his entire kind of adolescent life.
41:23 And so that kid's going to be able to
41:25 just say, I have this idea. Oh, and I
41:28 want it like that that that, you know,
41:30 short film I made three years ago. And
41:33 the AI will know what that means. And it
41:36 will it will make a claymation film
41:38 about whatever the thing is that he
41:41 wants that. Oh, remember when that girl
41:44 broke up with me and I was all upset?
41:46 Yeah, go make a movie like that. Like
41:47 that short film I made three years ago.
41:50 And boom, there'll be a half hour, you
41:52 know, a 15inute short. And the and the
41:55 AI will go, "Here's your 15inute short.
41:57 Do you want to see that as a feature
41:58 film?"
42:00 and his kid will be like, "Yeah, put in
42:03 uh put in my friend Bob and put in we'll
42:06 just do it."
42:08 That's the era we're entering. So, it's
42:11 it's funny when I see people talking
42:13 about the the the app as it is right
42:15 now, the Sora app.
42:18 They're talking about it like it's it's
42:21 a complete manifestation of of AI. No,
42:26 it is. it it's like a it's like a just
42:30 like a breath of a hint of what's
42:32 coming. And so with all of these tools
42:35 right now, when you hear people talk
42:38 about them, well, they're not very good.
42:39 It only does 15 seconds. The audio is
42:42 not perfect. The physics aren't perfect.
42:44 Okay.
42:47 Okay.
42:49 But if they were, what would this thing
42:52 be? Oh, I haven't thought about that.
42:55 Well, maybe you should, right? Okay,
42:58 beautiful. Johnny D found us on YouTube.
43:01 Fantastic. Welcome, sir. Real estate in
43:04 Columbus. Okay. Total recall
43:09 kind of instead of journaling, you
43:10 journal to AI. Yeah. And and and like
43:14 where this thing goes when these things
43:16 get more and more agentic, you just
43:18 journal and then it'll just be like, do
43:21 you want to see that? Do you want to do
43:23 that like a like a song like you did
43:25 yesterday? No, I don't feel like a song
43:27 today. Oh, do you want a picture? Do you
43:29 want a sculpture? Oh, do you want to 3D
43:31 print that?
43:33 Do you want it as an art piece or a
43:35 manufacturable,
43:37 you know, set of plans? Yeah, why don't
43:40 we make that one manufacturable? Okay,
43:42 great. And then you run down to your
43:44 local your local um 3D fabrication shop,
43:49 which every small town is going to have
43:52 a little mini factory that can make
43:55 anything.
43:58 And sevenyear-olds will be coming up
43:59 with products that they can manufacture
44:02 and sell in their local town.
44:06 It's going to be [ __ ] bonkers. All
44:07 right, let's let's keep going. Learn how
44:09 to how to prompt chat jpeay. Well, or
44:12 Sora. All right. For complex cinematic
44:15 shots, for example, you might describe
44:17 what the viewer notices first. The
44:20 camera platform and the lens, the
44:22 lighting direction, the color palette,
44:23 the texture qualities, the diagetic
44:26 sound.
44:28 Oh, for [ __ ] sake. Really? We have to
44:31 use words like diagetic for for someone
44:34 with a flipping acting degree. So I now
44:37 have to publicly humiliate myself and go
44:39 to chat GPT and ask it. Dear chat GPT,
44:45 what is diagetic sound? I think it's the
44:50 background sound, but I don't know.
44:56 Diagetic sound, sometimes spelled
44:59 diioetic,
45:00 is sound that exists within the world of
45:02 a film, show, or video. meaning the
45:05 characters can hear it too. Oh, this is
45:08 new to me. In other words, it's part of
45:12 the story's reality.
45:14 Music playing from a radio in a scene is
45:17 diagetic sound.
45:20 The the underscore music is not because
45:25 the characters can't hear it. Footsteps
45:28 footsteps dialogue car engine door slam
45:31 character singing ambient sounds like
45:33 rain. That's so cool. By contrast,
45:36 non-diagetic sounds
45:39 exist outside the story world, the
45:42 movie's background score, the narration.
45:44 That's so cool. Okay, see,
45:48 okay, for all you people that like, I
45:50 don't learn anything here.
45:53 We just did. I did. Did you learn? Did
45:56 you? If you I swear to God, if you're
46:00 over there like, I knew what that meant.
46:02 No, you didn't.
46:05 No, you didn't. If you're a sound
46:08 engineer, fine. Or if you're like Vicky
46:11 Baptiste who knows all the words, fine.
46:15 But you show offs with the I mean at
46:17 that, man. No. No. I don't accept you. I
46:21 I reject you. Okay.
46:26 It's exhausting. This is exhausting,
46:28 people. Okay. Diioetic sound and shot
46:30 timing. This approach works well.
46:34 Meltdown Monday.
46:36 Archetypal. Oh, wow. New words to learn.
46:41 Who? Okay. Who in here is claiming that
46:43 they knew what that meant? Anyone? I
46:45 only know it because it came up in a
46:47 prompt today. I looked it up, too.
46:49 That's awesome.
46:52 That's hilarious.
46:55 Vicki's got us by four hours. Listen,
46:57 this is AI. This is how it works. If
47:00 you've got a fourh hour head start on
47:01 somebody, you're the king queen in this
47:04 case.
47:07 That's awesome. Okay. This approach
47:10 works well when you want to match real
47:12 cinemat cinematography styles. IMAX
47:14 aerials 35mm handheld vintage 16
47:18 millimeter documentary or strict
47:19 continuity across shots. Example format
47:23 and look. Duration 4 seconds 180 degrees
47:26 shutter digital capture emulating 65
47:30 millimeter F right lenses and filtration
47:34 grade pallet lighting and atmosphere.
47:38 Okay, so this is a good structured
47:40 prompt. Okay, great.
47:44 Optimize shot list. So, this is this is
47:46 pretty um pretty important and and if
47:50 you're not used to um like screenwriting
47:53 or cinematic kind of uh language or or u
47:58 structure
48:01 within a single Sora
48:03 and you've probably noticed Sora
48:05 sometimes does this on its own where it
48:08 will have different shots, right? It'll
48:09 have a close-up of the person, then
48:10 it'll have the far away shot, then it'll
48:13 come back to the person. You can specify
48:16 those shots and you can specify exactly
48:18 when they happen.
48:20 And what we're going to build is we're
48:22 going to build a GPT that just does that
48:25 for you automatically.
48:29 All right. Camera notes. Wyatt reads,
48:33 "Keep eyline low and close the len
48:35 access lens axis for intimacy.
48:39 Allow micro flares from train glass as
48:42 aesthetic texture." Yeah, this is
48:46 I I'm glad we're actually going through
48:48 this this prompt uh guide slowly like
48:52 this.
48:54 If I'm a human being and I don't know
48:57 cinematography, I've written seven
48:59 screenplays. I don't know most I didn't
49:00 know what diotetic was. So,
49:06 um, some of the details of of the like
49:09 the cinematography things they're doing
49:11 here and the and the story things
49:12 they're doing here. Most human beings
49:16 won't know that, don't care to know
49:18 that,
49:19 but if they want good results, they're
49:23 going to have to know that to some
49:24 degree or they're going to have to have
49:26 a tool that does all that work for them,
49:29 right? And that's that's where these
49:31 agentic tools, right? Someone's going to
49:34 build an agentic tool that you'll just
49:36 say, "Oh, I want a thing that kind of
49:37 looks like The Shining, and it will know
49:40 all of this [ __ ] to do to make it look
49:42 like that."
49:44 All right. Visual cues that steer the
49:46 look. When writing prompts, style is one
49:48 of the most powerful levers for guiding
49:50 the model,
49:52 describing the overall aesthetic. For
49:54 example, '7s film, epic IMAX scale scene
49:58 or 16mm black and white sets the visual
50:01 tone. The same details will read very
50:04 differently depending on whether you
50:05 call for a polished Hollywood drama or a
50:08 handheld smartphone clip or a grainy
50:11 vintage commercial. Once the tone is
50:13 set, layer in the specifics with shot,
50:16 action, and light. Clarity wins. instead
50:20 of vague cues like a beautiful street,
50:23 which I do all the time, right? So,
50:26 we're all guilty of this. We're lazy.
50:28 You just want to see it, make the magic
50:29 thing. So, you're like, a beautiful
50:31 woman walks down a beautiful street with
50:33 a beautiful dog in a beautiful town at a
50:36 beautiful sunset.
50:38 You're going to get some generic [ __ ]
50:42 [Laughter]
50:45 Instead of a beautiful street, write wet
50:47 asphalt, zebra crosswalk, neon sign,
50:50 reflection. Instead of moves quickly,
50:54 she moves quickly across a beautiful
50:56 street. I write that prompt all the
50:58 time.
51:01 Again, one of the reasons you come to
51:03 the AI learning lab is you look at what
51:05 I do and then you do not that, you do
51:09 better than that. I'm here to establish
51:11 the bottom.
51:13 All right? and then you take it from
51:15 there.
51:17 You'll turn yourself into a movie maker.
51:20 Instead of moves quickly, specify jogs
51:22 three steps and then stops at the curb.
51:25 So, what's nice about reading this is
51:27 you I'm starting to learn how very
51:30 specific you can get with this model,
51:33 right? Jogs three steps, then stops at
51:35 the curb. I think what we all know is
51:37 that it'll jog two steps or it'll jog
51:40 four steps. If you actually wanted to
51:42 jog three, you're gonna have to make 17
51:44 different versions of the clip. I think
51:46 we all know that. Okay. Camera direction
51:49 and framing shape how the shot feels.
51:51 When introducing characters, expect some
51:54 unpredictability.
51:55 Small changes in phrasing can alter
51:58 identity, pose, or the focus of the
52:00 scene itself. Weak camera shot,
52:03 cinematic look. Strong camera shot. Wide
52:06 shot, low angle, depth of field,
52:08 shallow, sharp on subject, blurred
52:11 background, lighting and pallet,
52:14 warm backlight with a soft rim. All of
52:16 these assume you know [ __ ]
52:19 cinematography
52:22 and and film lighting.
52:26 Nobody knows that stuff. But that's
52:27 okay. We've got AI now.
52:33 Control motion and timing,
52:36 lighting and color consistency, image
52:38 input for more control. Ah, how this is
52:41 important. Use image input for more
52:43 control. For even more fine grained
52:45 control over composition and style of a
52:48 shot, you can use an image input as a
52:51 visual reference.
52:53 You can use photos, digital artwork, or
52:55 AI generated visuals. This locks in
52:57 elements like character design,
52:59 wardrobe, set dressing, or overall
53:02 aesthetic. The model uses the image as
53:06 an anchor for the first frame, while
53:09 your text prompts defines what happens
53:11 next.
53:14 All right.
53:17 The image must match the target video's
53:20 resolution size. Okay.
53:32 Why would they use that as an example?
53:34 Such a bad example.
53:37 Like she looks at the camera, which is
53:39 normal, and then she does a little a
53:41 little twirl. A little twirl for the
53:43 camera.
53:50 I think that's dumb. That's a dum dum.
53:53 That's a dum dum move. I told you I'd
53:56 make fun of them. All right, download
53:58 this image. The fridge door opens.
54:00 Chubby monster comes out. Dialogue and
54:03 audio.
54:11 Okay.
54:20 Iterate with the remix functionality,
54:25 prompt templates and example and
54:27 examples. Prompt structure.
54:39 All right. Okay.
54:41 So
54:47 let's take let's take this since this is
54:49 a public website.
54:52 Let's go to chat jeepet
54:59 and if you've never created a custom GPT
55:02 oh your browser has lost connection of
55:04 screenshot audio. Um,
55:09 all right. Hang on.
55:19 Your brows. Oh, browser has lost
55:21 connection with screen share audio.
55:24 All right. Well, that's a drag. All
55:25 right. Whatever.
55:29 Okay. So, we're gonna make a custom GPT.
55:32 What are you doing? A cookbook. I am
55:35 sort of what I'm doing is
55:40 can we create like a template to change
55:45 the something style the cinematic style?
55:48 Yes, we can. That's what we're going to
55:49 do.
55:52 Ann Murphy. Ann Murphy in the house. Um
55:56 I don't know where you are in the world.
55:57 Are you still at the create conference
55:59 or are you back back home? But if you
56:03 want to hop up, hop up on the uh hop up
56:07 on the stage. You must be exhausted
56:09 though. But if you want to hop up and
56:11 tell us how it was, I would love to
56:12 hear.
56:21 Where is
56:24 I got rid of it, I guess.
56:31 D.
56:34 [Music]
56:48 Come join the community feed. Who did
56:49 that? Danielle. Nice.
56:52 Nice. A lot of stuff in the community
56:54 feed. This is great. Cool. Yeah. Go feed
56:57 the community feed. Okay.
57:01 Let's see. Um,
57:05 okay. So, here's what we're going to do.
57:06 I'm going to go to explore GPTs.
57:09 And first thing we're going to do is
57:10 we're going to search for Sora 2
57:14 prompt.
57:18 Sora 2 prompt configured custom. The
57:21 Sora 2 prompt enhancer and generator.
57:23 The Sora 2 prompt generator. The AI
57:25 video prompt polisher. Video prompt
57:28 architect. Sora prompt. Sormpt.
57:32 Somebody went to marketing school. I got
57:35 it. Surumpt.
57:37 It's a combination of sora and prompt.
57:42 Wouldn't it Wouldn't pora be better than
57:44 sumpt.
57:48 Sumpt sounds like a drunk guy trying to
57:50 say a word you don't know. Like what
57:53 word was that? I don't know. He's drunk.
57:56 I know.
57:57 All right.
57:59 prompt. Sora
58:03 Sora prompt master GPT money. Uh, wait,
58:06 where's my
58:08 GPT money?
58:09 >> You can make money with your GPT.
58:12 >> So, there's plenty of Sora 2 prompt
58:15 GPTs. So, we don't need to make this,
58:17 but we're going to anyway. So, in the
58:18 upper right hand corner, there's a
58:21 little button that says create. We're
58:22 going to click that. And so now we're
58:24 going to go to the create tab and I'm
58:27 going to say, "Hey,
58:29 hey,
58:30 here is a prompt guide for the new Sora
58:38 to video model from OpenAI.
58:44 I Oh, wait. Let me give it to it." Bang.
58:47 There it is. I want to create a
58:54 super simple helper
58:57 that takes a crappy
59:03 prompt
59:05 and makes
59:09 it brilliant.
59:14 Let's have the What are those things
59:16 called? the uh here, hang on.
59:21 What are they called? The conversation
59:23 starters.
59:25 Let's have the conversation
59:28 wait conversation
59:32 starters
59:35 just be the length
59:39 of the clip.
59:44 Uh let's see. 4
59:49 uh 4
59:55 8
59:59 10
1:00:01 or 12 seconds.
1:00:06 And depending on the length,
1:00:10 the generator will make
1:00:19 um
1:00:21 we'll add
1:00:24 additional shots
1:00:27 to keep it interesting.
1:00:35 We'll just start with that.
1:00:37 The minute you get into want wanting to
1:00:40 to create a prompt generator for a film,
1:00:42 there's things like what era do you want
1:00:45 it? Like what's the look and feel you
1:00:48 want? Is it a comedy? Is it a drama?
1:00:52 Um
1:00:55 is it there's there's so many variables
1:00:59 that it's it's hard to get this to. But
1:01:01 but anyway, here's our So, it's it's
1:01:03 already done. So, we're going to make an
1:01:04 8-second
1:01:06 uh prompt,
1:01:09 although this is not working.
1:01:12 Come on, Chat GBT. Could you suck a
1:01:14 little less? Cinnaprompt Pro. Uh, how
1:01:17 about let's see
1:01:22 Soraprompt
1:01:24 Pro.
1:01:29 [Music]
1:01:34 It's not working.
1:01:37 So, let's create it.
1:01:40 We're going to put this right in the
1:01:41 chat. GPT store.
1:01:46 All right.
1:01:49 Your changes couldn't be published. You
1:01:52 can't you
1:01:57 you're it may contain content that
1:02:00 requires previous authorizations brand
1:02:02 names. Fine. Oh, that's why everyone was
1:02:04 spelling Sora weird. Okay, fine. Make it
1:02:10 um
1:02:15 sora
1:02:17 soromatic
1:02:23 prompt.
1:02:27 Soromatic
1:02:29 prompt
1:02:30 writer.
1:02:35 Soromatic.
1:02:39 Okay. Soromatic or surumpt?
1:02:44 You tell me.
1:02:47 Surumpt. You want to go with surumpt.
1:02:51 [Laughter]
1:02:54 Uh sor sorusy. Sorcery.
1:02:59 Sorosery.
1:03:00 I see what you did there.
1:03:04 Fishomatic.
1:03:06 It's the bassomatic. Fishomatic. You got
1:03:08 to get your names right. Okay. GPT
1:03:09 store. Uh, Dolly 3. Where are we going
1:03:12 to put this education lifestyle program?
1:03:16 No.
1:03:19 Isn't that fascinating that they don't
1:03:21 have a creative? Well, they've got
1:03:22 writing as a creative output. They don't
1:03:25 have film or anything like that. So,
1:03:27 we'll do other.
1:03:30 All right. So if you want soromatic
1:03:33 unable to publish. Oh come on
1:03:39 appeal.
1:03:44 Soromatic
1:03:46 is just awesome
1:03:49 and it doesn't require and it doesn't it
1:03:55 doesn't violate
1:03:58 anything.
1:04:04 [Laughter]
1:04:09 Oh my god. During the review process
1:04:12 using suric as a private
1:04:18 wait reviewing your appeal during you
1:04:22 can continue using it in private mode or
1:04:25 you can share it with others. But I
1:04:26 can't
1:04:28 I can't get out of this dialogue. The
1:04:30 only way out of this dialogue is to
1:04:32 cancel the appeal. Oh, good lord.
1:04:37 They need a cranky Gen Xer in the
1:04:39 corner.
1:04:40 Yeah, I still can't get out of it.
1:04:42 Cancel appeal. Whatever. I don't give a
1:04:43 [ __ ] Okay. Uh, you're back. Let's see.
1:04:46 Um,
1:04:48 change it to
1:04:52 um,
1:04:55 so
1:04:57 ra
1:05:03 so so
1:05:07 don't measo lata
1:05:10 sora
1:05:13 Uh,
1:05:17 no one will get that. Let's see. So,
1:05:22 uh,
1:05:26 so ramen
1:05:32 prompt maker.
1:05:34 Oh my god.
1:05:39 Whatever. Okay. Use a zero. That's all
1:05:43 right. I don't give a [ __ ]
1:05:46 I just do not give a [ __ ]
1:05:52 So, ramen
1:05:55 prompt maker.
1:05:57 Okay. Wait.
1:06:00 What happened to my
1:06:04 timing
1:06:07 starters?
1:06:09 I didn't tell you to change that. You
1:06:11 big damn damn dummy.
1:06:23 Revert to them.
1:06:34 Make a 4 second cinematic soru. Okay.
1:06:39 Create. Oh, this might not let me do it
1:06:40 either because the Sora 2 stuff is in
1:06:43 the uh is in those conversation
1:06:46 starters. But let's see.
1:06:48 So, ramen prompt maker
1:06:52 [Laughter]
1:06:55 [Music]
1:06:57 view GPT. All right. Well, there you go.
1:07:01 So if you if you go to the so ramen
1:07:09 um I can't get there. Wait, copy this.
1:07:13 Let me just go here.
1:07:16 I have a feeling their servers are
1:07:18 hosed.
1:07:20 Okay, so the so ramen prompt maker. So
1:07:23 we're going to write a 10-second Sora 2
1:07:25 prompt
1:07:27 about a dancer under neon lights.
1:07:34 Make it about
1:07:37 a
1:07:40 guy at Kyle Shannon
1:07:44 who is a detective
1:07:49 in
1:07:52 Babylonian
1:07:57 times.
1:08:01 who
1:08:02 is trying to discover
1:08:11 what the future people
1:08:15 mean when they say
1:08:19 they prompted
1:08:23 him
1:08:24 into
1:08:28 existence.
1:08:30 All right.
1:08:36 So, sync audiovisisual harmony diagetic
1:08:39 focus. That's nice, Silver Fox. That's
1:08:42 good. In the heart of ancient Babylon,
1:08:45 detective Kyle Shannon walks through a
1:08:47 torchlet bizaar at dusk. Scrolls and
1:08:51 clay tablets scattered around him. The
1:08:54 camera tracks beside him as he studies
1:08:57 strange inscriptions that glow faintly
1:09:01 with futuristic symbols. A whisper
1:09:03 echoes. They prompted you into existence
1:09:06 and his eyes widen in disbelief. Cut to
1:09:08 a closeup of his reflection in a bronze
1:09:11 mirror morphing into the shifting lines
1:09:13 of code. A slow crane shot reveals a
1:09:17 ziggurat zugarat ziggurat
1:09:21 skyline under stormy clouds. Lightning
1:09:24 flashes with digital glitches ripple
1:09:26 through the air. The moment feels
1:09:28 suspended between epochs. Babylonian
1:09:31 grit and digital prophecy colliding.
1:09:34 Style and lighting. Mesopotamian noir.
1:09:38 Warm torch lit mixed with surreal
1:09:41 electric flickers. Cinematic dust and
1:09:43 atmosphere. Duration approximately 10
1:09:45 seconds. Cool. All right. Let's see if
1:09:49 we can make this [ __ ]
1:09:52 Cool. Boom. Copy.
1:09:55 All right. Beautiful. Beautiful. So, go
1:09:58 to the so ramen prompt maker and make
1:10:02 yourself some prompts and then go over
1:10:04 to Sora
1:10:06 and let's make us some uh
1:10:09 let's close the tab.
1:10:12 Let's go. Let's see. Let me go here and
1:10:15 go
1:10:18 Sora.
1:10:20 Hello Sora.
1:10:23 Hello. My name is Sara. Can I help you?
1:10:27 Carrying on then rather. All right. Then
1:10:30 we got this. Then we go share this tab
1:10:32 instead.
1:10:34 Then we go. Oh, we lost audio. But
1:10:37 that's okay. We're going to put in that
1:10:38 long thing. And we're going to change
1:10:39 this to landscape.
1:10:42 So, if you didn't know if you could do
1:10:43 that, you can change it to landscape or
1:10:45 portrait. Am I not sharing this?
1:10:51 It didn't change.
1:10:55 H
1:11:05 weird.
1:11:07 It's weird, I tell you. It's very
1:11:09 strange.
1:11:12 Oh, and that's really slow. Yeah, my
1:11:14 machine is crashing is what's happening.
1:11:17 Carrying on. Anyway, you can over here,
1:11:20 see the right, I can do landscape. So,
1:11:22 we're going to do landscape and we're
1:11:24 going to do Kyle Shannon as a Babylonian
1:11:26 thing. Oh, you know what? We should do
1:11:29 a whisper.
1:11:32 A whisper echoes.
1:11:35 A I'm gonna say a whisper of
1:11:39 atama.
1:11:40 [Laughter]
1:11:42 A whisper of sama echoes. They prompted
1:11:45 you into existence. So, we're going to
1:11:47 put Sam Alman in this.
1:11:50 This is cool. I I think I'm liking this
1:11:52 uh this prompt maker.
1:11:55 Okay. One of the lessons here.
1:12:00 So, if you've got any kind of neurospicy
1:12:03 condition or if you're lazy,
1:12:08 if the thought of reading manuals makes
1:12:10 your head hurt and makes you want to
1:12:12 quit life like me,
1:12:16 unlike my son, Zach, loves manuals.
1:12:18 He'll read a manual for a hammer. I
1:12:20 swear to God, you know, Dad, if you take
1:12:23 that label, I know it's a hammer. I know
1:12:25 how to We're good. We don't need the
1:12:27 manual. But if you take the right
1:12:30 He drives me [ __ ] crazy. Love him.
1:12:34 If you're like me,
1:12:37 you go through life not knowing the
1:12:41 features of the tools you use. You learn
1:12:44 to use them well enough and then that's
1:12:46 all you learn. The the manual readers,
1:12:48 they learn all the [ __ ] Now, they
1:12:51 probably don't have your fil nimleness
1:12:53 with idea creation, but they know what
1:12:56 the [ __ ] tool does. the least we now
1:12:59 get to have some of the benefit as if we
1:13:02 read the manual just by just being a
1:13:05 little clever.
1:13:07 All right, let's make this
1:13:15 Brandon Thomas Gordo tips and tricks
1:13:17 indeed.
1:13:21 I read manuals. I'm the only one in my
1:13:23 family, too. That's hilarious.
1:13:27 Prompttosaurus is no longer available.
1:13:34 That's a good one, Joker. That's a good
1:13:36 one. Good for you. I like it.
1:13:42 Loud thunder and lightning.
1:13:49 Or it could be the big spaceships coming
1:13:51 in. You know there's big spaceships
1:13:52 coming in, right? That's what everyone
1:13:55 keeps saying. The big spaceships are
1:13:57 coming. All right. Whatever. After like
1:14:00 I just I feel like at this point Yeah.
1:14:02 Okay. All right. Big spaceships. Cool.
1:14:04 Yeah. Bring them on, too.
1:14:08 Yeah. Why not? Those two,
1:14:11 man. Write your own manual and how to
1:14:14 use it so I can make a bot to implement.
1:14:16 It would have improved some of my past
1:14:19 relationships.
1:14:20 [Laughter]
1:14:23 Oh, you tried to give it to me earlier.
1:14:25 Yeah, I was off trying to use my brain.
1:14:27 Prompttosaurus, too. Prompttosaurus
1:14:31 learning lab edition.
1:14:35 All right, let's see. Did Do we have
1:14:36 something? Did it do it already? No,
1:14:39 this crashed. God damn. This their site
1:14:43 is absolutely unusable.
1:14:49 Oh,
1:14:52 they need a cranky guy in the corner.
1:14:54 See, if I were the cranky Gen Xer in the
1:14:56 corner at Open AI,
1:14:59 like that would have been fixed night
1:15:01 one.
1:15:03 Like, how do you have it how do you have
1:15:05 it so that whatever you've done on the
1:15:08 browser? Like I understand that this is
1:15:10 taking lots of resources in the GPUs,
1:15:14 but they've basically got some sort of
1:15:16 runaway memory leak in their browser
1:15:18 application that [ __ ] crashes Chrome
1:15:21 after about 20 seconds every single
1:15:23 time. And it's been like that for more
1:15:25 than a week.
1:15:28 I mean, listen, I know nothing about
1:15:30 coding. And if you give me half a
1:15:32 billion dollars, I'll fix it right now.
1:15:35 I'll fix it in like the next 20 minutes.
1:15:37 You know how I'll use [ __ ] chat GPT
1:15:39 to fix it.
1:15:42 Just give me the GitHub repository. And
1:15:44 I don't even know how to do that. I'll
1:15:46 have ChatGpt teach me how to get the
1:15:48 [ __ ] out of the GitHub repository, fix
1:15:50 the [ __ ] runaway memory leak, and put
1:15:52 it back up there so that human beings
1:15:55 lives aren't ruined when they're trying
1:15:57 to make a [ __ ] live show promoting
1:16:00 your stupid [ __ ] software.
1:16:03 I'll do that for you for half a billion
1:16:05 dollars.
1:16:07 Seems like a bargain.
1:16:10 Black bar. Yes. Thank you, Brendon, for
1:16:13 that kind reminder.
1:16:16 We've had the black bar for free.
1:16:20 Oh, good lord. Okay, let's let's make
1:16:25 another let's make another uh let's make
1:16:28 another tab, shall we? Since uh
1:16:32 I I I just about had an aneurysm there.
1:16:39 We're I don't even know what I'm doing.
1:16:41 Sora.gpt.com.
1:16:43 That's what it is. Now I'm going to
1:16:44 share this tab instead. And see if we
1:16:47 got us some Babylonian magic.
1:16:50 Boom.
1:16:53 Nope. Still going. They need to. What's
1:16:56 the best way for chat GPT users to get
1:16:58 life-changing results?
1:16:59 >> Simple. SM. They need to AI their life
1:17:01 changings.
1:17:02 >> Why? That's genius. Absolutely genius.
1:17:06 What's the best way for ChatGpt users to
1:17:07 get life-changing results?
1:17:08 >> Simple SM, they need to AI their life
1:17:11 changings.
1:17:12 >> Why? That's genius. Absolutely genius.
1:17:15 What's the best way for ChatGpt users to
1:17:17 get life-changing results?
1:17:18 >> Simple SM, they need to AI their life
1:17:20 changings.
1:17:21 >> Why? That's genius. Absolutely genius.
1:17:25 >> I wonder if there's some sort of
1:17:27 influencer Valhalla.
1:17:28 >> So many So many people would make me
1:17:30 videos with Sora. I told you that if you
1:17:32 made a Tik Tok-like app, they would
1:17:34 bury.
1:17:35 >> You're right, Kyle. You always are.
1:17:39 >> So many people would make meme videos
1:17:40 with Sora.
1:17:40 >> It's still not very good at getting the
1:17:42 right people to say the right things.
1:17:44 >> Sam, AI is not automation. It's
1:17:46 amplification. Every spark you feed it,
1:17:49 it multiplies.
1:17:51 >> Sam, the AI is not AI learning lab is
1:17:55 Shannon's AI learning lab is more useful
1:17:57 than the unified field.
1:17:59 >> You're on a roll. Get it?
1:18:05 Shannon's AI learning lab is more useful
1:18:07 than the unified field. You're on a
1:18:09 roll. Get it?
1:18:10 >> I'm going to post that. It's It's
1:18:12 tasteless, but it's funny.
1:18:16 As my son Zach once said to me, "Dad,
1:18:19 feels like
1:18:22 >> Oh, I think we have a video. Do we have
1:18:24 a video?"
1:18:26 >> Come on.
1:18:28 Uh
1:18:28 >> oh. AI. Why? It changes everything when
1:18:30 you learn to feed your prompt, your
1:18:32 purpose, your heart, your curiosity.
1:18:36 >> Come on.
1:18:37 Come on.
1:18:39 You can do it. There it is. Why didn't
1:18:43 it make it?
1:18:43 >> They prompted you into existence.
1:18:47 >> What? This can't be real.
1:18:54 >> They prompted you into existence. What?
1:18:58 This can't be real.
1:19:03 >> That was bad.
1:19:04 >> They prompted you into existence.
1:19:06 >> What?
1:19:07 >> This can't be real.
1:19:13 >> They prompted you into existence.
1:19:16 >> What?
1:19:17 >> Yours can't be real.
1:19:22 >> They prompted you into existence.
1:19:25 >> What? Yours can't be real.
1:19:32 They prompted you into existence.
1:19:34 >> What? This can't be real.
1:19:40 >> Existence.
1:19:46 Oh, it just crashed again.
1:19:49 We used it for We used it for 62
1:19:51 seconds. Oh, no. It's We're okay. It's
1:19:54 back.
1:19:56 All right, we're going to try that
1:19:57 again. It was close. I've got so many of
1:20:01 these that are close, but it's like it
1:20:03 it needs like another round of of
1:20:05 training to like get the voices right,
1:20:08 have the right people talking. All
1:20:10 right, Kyle, update the window in
1:20:11 Chrome, please.
1:20:13 I think I I am This is the one I'm in.
1:20:18 This the one I'm in.
1:20:21 [Music]
1:20:23 I'm going to close this al together.
1:20:28 All right,
1:20:37 [Laughter]
1:20:41 producer Brandon, they are suggesting a
1:20:43 Chrome update would fix your issue. It
1:20:45 won't. It won't. it. This is Listen, I
1:20:49 push the boundaries of Chrome every
1:20:51 single day of my life. I'm a 78 to to 90
1:20:54 tab hoarding monster
1:20:58 and I can tell you that this Sora app is
1:21:02 the biggest piece of [ __ ] I've used
1:21:05 probably since the mid 90s when I used
1:21:08 to build these pieces of [ __ ] that
1:21:11 crashed browsers.
1:21:17 That was back when browsers couldn't
1:21:19 handle anything.
1:21:21 Oh man. Side hustle Mimi, hey there from
1:21:24 Create 2025. What's happening?
1:21:28 Flash is back, baby. Exactly. We're
1:21:30 going to do some some flash flashomatic.
1:21:34 Um,
1:21:35 okay. So,
1:21:38 what did we learn tonight? We learned
1:21:44 chances are
1:21:49 You're not a very good prompter,
1:21:54 unless you read the manuals and write
1:21:56 really good prompts, in which case I
1:21:58 resent you.
1:22:00 But if you're like me, you're not a good
1:22:02 prompter. And if you want to get good
1:22:05 results out of increasingly
1:22:07 sophisticated models, you got to get
1:22:09 your [ __ ] together. or you need to wait
1:22:13 for a gent AI to get better,
1:22:17 which probably within a year, you're not
1:22:20 really going to have to prompt all that
1:22:21 much. It's just these are going to just
1:22:23 do this [ __ ] for you. But I I can tell
1:22:26 you like like a lot of the the the video
1:22:29 apps that make it easy to get good
1:22:31 results out of their video systems, what
1:22:34 they're doing is essentially what we
1:22:35 just some version of what we just did.
1:22:38 You put in a shitty prompt and it it
1:22:40 writes out in a in a structured prompt
1:22:44 all of the all of the details with
1:22:46 cinematic uh fervor.
1:22:51 You got to have cinematic fervor. Let me
1:22:54 go check and see if uh if our video is
1:22:57 there. I don't think it's going to be.
1:22:58 Their servers are kind of slow.
1:23:01 sora.jpay.com at jpay.com.
1:23:16 >> Yeah, I don't even think it's started
1:23:18 yet.
1:23:20 I do not even think it's started yet.
1:23:23 All right,
1:23:26 so we'll come back to that. Um, what
1:23:28 questions? Anybody have any questions? I
1:23:30 just got back from dinner with Kuno. Oh,
1:23:33 that's so cool.
1:23:35 Beautiful. All the uh all the create
1:23:38 people who are irregulars. How many
1:23:40 irregulars were at create? It's probably
1:23:42 a pile.
1:23:44 Side hustle made me fun. That's Yeah,
1:23:46 that's a awesome.
1:23:50 [Music]
1:23:53 Every time I try making a Sora app
1:23:55 video, it fails. Yeah, their their
1:23:57 servers are hosed right now.
1:23:59 Just got back from dinner with Kvuno.
1:24:04 [Music]
1:24:08 It's way worse on desktop. Yeah, the the
1:24:11 the iOS app is actually fine. The the
1:24:14 iOS app I have no problems with. It's
1:24:16 it's their it's their desktop app is a
1:24:19 steaming pile of poo. And I I honest to
1:24:22 god this is something they could fix. I
1:24:24 like they're not doing that much. How
1:24:27 many [ __ ] video apps are out there
1:24:29 that are showing, you know, um,
1:24:32 thumbnails of videos that when you roll
1:24:34 over them, it plays the video? Like
1:24:39 since like like Netflix,
1:24:43 every every [ __ ] app, every [ __ ]
1:24:46 video app works. It is fixable.
1:24:53 Sorry. Sorry, Mr. Shannon. Uh we
1:24:55 understand that you're upset. Uh but
1:24:57 here at OpenAI, we're we're committed to
1:24:59 uh artificial general int sorry uh
1:25:02 strike that uh we already achieved
1:25:04 artificial general intelligence. Anyway,
1:25:06 that was kind of trivial. Uh no, we're
1:25:08 fully committed to uh safe uh artificial
1:25:11 super intelligent, right? That's our our
1:25:14 ex employees company's name. But we're
1:25:16 we're uh absolutely committed to uh
1:25:20 making these things smarter than you,
1:25:22 which they already are, but you can't
1:25:24 admit that yet. But anyway, so uh we're
1:25:27 putting our resources into uh the upper
1:25:30 echelon of scientific discovery and uh
1:25:33 therefore we we don't have time for
1:25:35 that. We understand that the it is
1:25:37 slightly inconvenient to have a website
1:25:40 that crashes somewhere between 20
1:25:42 seconds and 63 seconds is what I've been
1:25:44 told. Uh so it's just not a priority
1:25:46 right now. It is on the backlog. It is
1:25:48 on the backlog. we've got on the backlog
1:25:50 to u optimize the code for the Sora
1:25:55 website the where people are making the
1:25:58 millions and hundreds of millions of
1:26:00 videos that are costing us billions of
1:26:02 dollars. Uh it it is on the backlog.
1:26:04 That's it's on the backlog and uh
1:26:06 meanwhile we're going to be inventing
1:26:08 the future. So if you could just kindly
1:26:11 shut the [ __ ] up that would be
1:26:12 fantastic.
1:26:14 Yeah.
1:26:19 [Laughter]
1:26:26 They can probably fix it by just letting
1:26:28 us doom scroll instead of a wall of
1:26:30 videos. Exactly.
1:26:33 Oh man.
1:26:36 [Laughter]
1:26:41 Good commercial. Yeah.
1:26:45 H What time is it? Is it time? Yeah,
1:26:47 it's getting B time. I got I got a bit
1:26:48 of a headache because of all of my
1:26:50 yelling at at Sam Alman. But I do want
1:26:53 to see that video. I would like to see
1:26:54 if it made a an actual
1:26:58 um whatever you call it, landscape
1:27:00 version of that video. So, let me see.
1:27:03 Sora.gpt.com.
1:27:10 Well, we're almost done. All right. Let
1:27:12 me try to share my tab.
1:27:15 See if we get an actual video.
1:27:22 [Music]
1:27:27 Black bar. Black ball. Hey, blackberry.
1:27:30 Bam jam.
1:27:34 All right, we have a white circle. All
1:27:36 right, it went away
1:27:39 and it'll come back. I promise. Right.
1:27:45 Yes. And it did the same [ __ ] thing.
1:27:49 What is going on?
1:27:52 They prompted you into existence.
1:27:56 [Music]
1:28:01 They prompted you into existence.
1:28:05 [Music]
1:28:08 >> Huh? Well, that's shitty.
1:28:11 >> They prompted you into existence.
1:28:13 >> Sora 2 sucks. Here's
1:28:20 what we've learned tonight. Don't use
1:28:22 Sora, too. It's awful.
1:28:28 I don't know why. I don't know why it
1:28:30 insists on making a vertical
1:28:33 uh a vertical video inside a horizontal
1:28:36 frame, but it does. So, it it did it
1:28:38 twice now. So, all right. Well, I'm
1:28:41 going to bed.
1:28:43 You can figure it out if Hey, here's the
1:28:45 deal. You want homework for tonight?
1:28:47 Figure out why the [ __ ] it did that and
1:28:49 teach me the secret to not that.
1:28:53 My computer is acting
1:28:56 worse than Sora tonight.
1:29:00 Sora is acting pretty bad.
1:29:04 Oh, good lord. All right. Um, I'm going
1:29:07 to get out of here. So, tomorrow
1:29:10 is just a normal Tuesday.
1:29:13 Just another normal Tuesday. Ha.
1:29:17 [Laughter]
1:29:20 Did he just sing a Bengals song poorly?
1:29:24 He did, didn't he? Was that That was the
1:29:26 Bengals. I mean, they were an okay band,
1:29:29 but why would you sing that? Why would
1:29:31 you sing those songs?
1:29:34 I thought this was more of like an
1:29:36 institutional educational situation, and
1:29:39 I I didn't understand that at all.
1:29:42 All right, getting out of here. Peace
1:29:44 out. Hope you had fun tonight. Bye.
1:29:49 [Music]
1:29:51 So frustrating. I can't get any of the
1:29:53 large language image editors
1:29:56 to make the exact right size, especially
1:29:59 in landscape. Yeah, Lillian, here's the
1:30:02 deal about all this AI stuff. It all
1:30:05 still kind of sucks. It's, listen, don't
1:30:08 get me wrong, it's remarkable. Like,
1:30:11 what we can do now is like historically
1:30:15 powerful
1:30:17 and it still sucks. So, just two more
1:30:21 years, two, three if with five at the
1:30:25 max and it's going to be awesome. right
1:30:28 now. It's just probably next year or the
1:30:32 three out or it farther than that is
1:30:36 going to be so good.
1:30:46 It has been a manic Monday.
1:30:51 I'm still working. Tantrums and turtles
1:30:53 Tuesdays. Five years. Really? I don't
1:30:56 think it'll be five years, but like
1:30:59 in the in the three years since we've
1:31:02 been doing this, two years ago,
1:31:07 video generators,
1:31:09 remember Will Smith eating spaghetti?
1:31:11 That was two years ago.
1:31:14 So two years ago, video generators
1:31:18 couldn't make anything resembling
1:31:20 reality.
1:31:21 We've now got something that can do
1:31:23 multiple shots with audio generation
1:31:26 with decent physics and they kind of
1:31:28 suck but like the the distance that
1:31:31 we've come in two years. So if I just
1:31:33 project that two years on
1:31:37 these things are going to be generating
1:31:42 two to five minute clips and then and
1:31:46 then people will be building
1:31:47 applications out of them where so here's
1:31:50 my prediction in two years
1:31:54 these things will generate
1:31:58 probably 120
1:32:01 toundred
1:32:04 60 frames a second
1:32:08 and they're going to display them at 30
1:32:10 frames a second. Right? So what that
1:32:13 means is you can talk a movie into
1:32:16 existence and it will generate it faster
1:32:19 than it can play it. And so you can just
1:32:22 keep modifying the film. You could just
1:32:25 say, "No, make it a different color."
1:32:28 And it'll just change the color. make it
1:32:30 a female lead instead of a male lead and
1:32:32 it'll just make it a female lead. Like
1:32:35 that's that's where we're headed. So, so
1:32:37 just
1:32:39 know that you will look back at this
1:32:41 time in history with nostalgia.
1:32:44 So, all right. Janky is still
1:32:47 embraceable. Yeah, embrace the jank. We
1:32:51 are We are definitely still I sent I
1:32:53 sent a tweet to Sama. It should be
1:32:56 sorted shortly shortly. That's great.
1:32:59 Okay, cool. All right, everyone. Have a
1:33:01 good night and I will see you tomorrow.
1:33:03 Bye.