11/20/2025 - Unlocking Gemini 3's Visual Reasoning: Beyond Simple Image Generation

Description

Chapters

Transcript

0:08 going to put myself up on stage. We're
0:10 going to do that thing. There's that.
0:12 We're going to go black bar.
0:16 All right, Champion, are you ready to
0:18 sing? Are you ready to do this now?
0:22 Um,
0:24 let's see. Um,
0:28 [music]
0:32 [music]
0:42 [music]
0:44 let us be lovers. We'll marry our punch
0:47 together.
0:51 I've got some real estate here in my
0:54 bag.
0:57 So we bought a pack of cigarettes
1:00 and Mrs. Wagner's pass and we woke
1:04 [singing]
1:04 up for America. [music]
1:29 Kathy I said as we bought a Greyhound in
1:32 Pittsburgh
1:36 Michigan seems like a dream to me now
1:40 [music]
1:41 it took me four days to hitch from Sag
1:46 [singing] I've come to look for America
1:50 >> [music]
1:55 [music]
2:00 >> All right, good people. Do me a favor.
2:03 Share the little lives. Share what we
2:05 got going on with the world. We got some
2:07 Nano Banana. We got Nano Banana in the
2:10 house. It's fantastic. Bob, tell him
2:11 what he's won. You've won a banana.
2:15 Uh Nano Banana did something remarkable.
2:17 One of the reasons I'm a little late is
2:19 uh is [ __ ] [ __ ] [ __ ] was blowing my
2:24 mind. [laughter]
2:27 [snorts]
2:27 To be clear, Nano Banana is not an image
2:30 generating tool.
2:32 It's not.
2:35 It does that. That's not what it is.
2:41 It's [ __ ] crazy. Hey Tobias, what's
2:42 shaking man? What is going down? [music]
3:01 Sending this lonely town. Yeah. One
3:04 where things [music and singing] are
3:05 going to change.
3:09 Dream my life away.
3:11 Seems these dreams have turn bunched us
3:14 clouds. Getting
3:17 my nerve up, but my past is pulling me
3:20 down.
3:22 Yeah. [music and singing]
3:24 Back to my blue jeans.
3:34 Somebody told me once before, he said,
3:36 "You can never go home again. [singing]
3:39 Won't you leave?"
3:41 said [music] things to ste away. Yeah.
3:45 from the truth of who I am and [singing]
3:47 what I [music] believe. So I thanked him
3:50 for his two cents with a handshake
3:52 [music]
3:52 and some sympathy. [singing]
3:54 Yeah.
3:56 Packed up my blue jeans [music]
3:58 and [singing] headed for this big prize
4:02 of my freedom.
4:05 Bye-bye
4:07 black sheep to the black sheep of the
4:11 family. [singing and music]
4:13 Bye-bye.
4:15 [singing]
4:17 Oh, it means so very much to me. [music]
4:20 Bye-bye,
4:22 black sheep. To my friends and my
4:25 [singing] family.
4:27 [music] Bye-bye.
4:31 Going to set my soul.
4:35 [music]
4:37 Set it free.
4:43 >> [music]
4:48 >> What is Google Gemini anti-gravity?
4:52 It is a uh it's it's it's very much like
4:55 cursor or replet the kind of standalone
4:58 applications for vibe coding. Um it's
5:02 very unstable right now. I don't think
5:04 you need to deal with it too much. Um, I
5:07 I am assuming that is more geared toward
5:10 developers than non-developers. Um,
5:15 but I don't know uh because I haven't
5:17 been able to get it to run because my
5:19 version of it on my Mac is locked in um
5:24 setting up account mode. Like I can't I
5:27 can't get past it setting up the
5:28 account. So I haven't played with it
5:29 yet. Google Gemini has so many different
5:32 products. It's confusing. It is. And
5:34 believe it or not, Stacy, it's um it's
5:38 better than it was. I mean, for 2 years,
5:41 I [ __ ] all over Google because you you
5:44 couldn't find anything. Now there's
5:47 essentially three three places to look
5:49 for Google products. All right.
5:52 Gemini.google.com.
5:55 Think of that as ChatGpt. Google's
5:58 version of Chat GBT. If you want the
6:00 most simple
6:02 experience, it's that the the the phone
6:06 equivalent of that is the Gemini app,
6:08 right? So, you can get the Gemini app on
6:11 Android or iOS and you on the web you
6:14 can go to gemini.google.com.
6:17 Those are the kind of commercial
6:20 um
6:21 you know the the for lack of a better
6:23 term the consumer applications.
6:26 Um, if you want to get a little geekier
6:29 and dig deeper and do things like vibe
6:31 coding, a studio.google.com,
6:36 that's kind of the equivalent of the
6:37 chatpt playground where where you can
6:40 access different models, you can access
6:42 the API, you can do some vibe coding in
6:44 their build area, their app builder,
6:46 which is actually really good and really
6:48 userfriendly.
6:50 So, that's that. And then the third
6:53 place you can look for stuff at Google
6:54 is labs.google. google.com and that's
6:57 just their experiment bucket. So,
6:59 Notebook LM came out of labs. So, if you
7:04 just want to go play with some cool toys
7:06 that are experiments, that's
7:08 labs.google.com. And the things there
7:10 that are good will likely, you know,
7:13 grow up and turn into products. All
7:15 right.
7:17 [music]
7:20 It's an IDE versus code. Yeah, exactly.
7:22 It's a it's an internet development
7:24 environment. the anti-gravity thing is
7:27 an IDE, so you can do, you know, full-on
7:29 development in it. Um, but much like
7:31 Replet, much like uh uh cursor, things
7:35 like that. Um, I don't know if it's any
7:38 good because it's it's not stable. And
7:41 actually, some of the reviews I've seen
7:43 of it are are not that stable. Um, what
7:46 is remarkable right now, like what is
7:48 remarkable remarkable is Gemini 3 and
7:52 Nano Banana Pro, which came out today. I
7:55 It did a thing tonight right before I
7:57 got on here that I like I'm I'm kind of
8:00 shook.
8:02 I'm kind of shook. It's amazing. I'll
8:04 show you. I'll show you what it's doing
8:06 or what it did. Um,
8:09 it's it's not it's not an image
8:11 generator tool.
8:14 It is a
8:16 it's a visual reasoning engine
8:21 and it's and it's not even
8:25 a visual reasoning engine. It is like
8:28 okay
8:30 everyone all of the model companies for
8:32 the past two years have been talking
8:34 about multimodal models where the where
8:37 the models have images and video and
8:40 everything incorporated into them but
8:43 all of them essentially to you know to a
8:47 model
8:50 are not a seamless integration right
8:53 it's sort of like you flip into image
8:55 generation mode or you're in chatting
8:57 mode and and you can sort of switch back
9:00 and forth between them, but Gemini 3 and
9:05 Nano Banana are so seamlessly
9:08 integrated. It's insane. It's wild.
9:11 Um, [music]
9:13 thrown [laughter] for a loop Thursday.
9:19 [laughter]
9:22 [music]
9:24 So, we're going to go look at we're
9:25 going to go look at Nano Banana. I think
9:27 we're just going to play with it tonight
9:28 because here's the thing about it. When
9:30 I when I show you this thing that I did,
9:32 and I know you're like, "Why aren't you
9:33 showing us now, Kyle?" You know, you
9:35 know, we come in here, we tolerate you
9:38 playing the guitar with those same silly
9:39 songs and that dog. I mean, we know he's
9:41 cute, but you know, whatever. But
9:43 anyway, you seem excited about something
9:45 and then you're not telling us.
9:48 I know. It's like some media theory.
9:51 It's like a cliffhanger, right? I'm
9:53 like, there's gonna be a cool thing. And
9:55 then like most people are just gonna be
9:57 like, oh, [ __ ] him. [laughter]
10:01 Flick me away. Flip. [laughter]
10:06 Haven't had a chance to look at Gemini 3
10:08 yet. It is pretty crazy.
10:11 Weren't Weren't you here the other
10:12 night, Source Camp, when I did the when
10:14 I made that globe history interactive
10:18 globe thing where you drop a pin on
10:20 anywhere on the globe and it pulls an
10:22 article from that point on the globe.
10:24 Yeah. Yeah. Yeah. Okay. So, you saw it.
10:25 Yeah. It's pretty crazy. [music]
10:28 I haven't My honest to god, my days
10:32 lately, this week has been insane. I
10:35 don't know how long it'll last tonight,
10:36 but I'm I'm doing coffee. I I flamed out
10:39 last night with a headache. So, we'll
10:41 see how we do it tonight.
10:45 Oh, man.
10:49 It was funny earlier when I was singing,
10:50 I put on there's uh the Bills game
10:53 tonight, the football game, and so I
10:55 just off in the corner. I don't I don't
10:56 have any sound on. I just have it on for
10:58 visuals to just look up every now and
10:59 again. And uh when I was singing, Champy
11:03 was watching TV. He he would sing a
11:05 little. He'd be like, "Oh." and he'd
11:07 look up at the TV, be like watching TV.
11:10 [laughter]
11:11 It's Dogs are weird, aren't they? Like
11:13 some dogs can see TV, some dogs can't.
11:19 [music]
11:23 How have you been? I got lost. Well,
11:25 welcome. Welcome back from the jungle.
11:29 Oh man, my leader just needs to find a
11:32 way to get free tokens for Gemini 3.
11:35 Well, I these tools are starting to get
11:38 good enough that they're worth paying
11:39 for. I don't know how much you have to
11:42 pay for Gemini 3. I assume there's a $20
11:45 version of it.
11:49 In fact, I know there is [music]
11:53 cuz I've got a pro account. That's the
11:55 20 buck a month account. And I've got an
11:57 ultra account and that's the really
12:00 expensive one. Nano Banana is still
12:02 putting [ __ ] watermarks on the
12:03 images, which that is driving me crazy.
12:07 [music]
12:08 He needs 30,000 tokens to start.
12:14 I don't know how much 30,000 tokens is.
12:16 [music] What's he need 30,000 tokens
12:18 for?
12:22 [music] The waters are supposed to be
12:24 Oh, watermarks are supposed to be
12:26 removed if you have a pro account. Uh,
12:28 well, they're not. They're not Danielle.
12:31 I'm In fact, I
12:35 it generated an image that I liked of me
12:38 as a pirate [laughter]
12:40 and I just wanted it to take the
12:42 watermark off. So, I said, "Take the
12:43 watermark off." And then I moved it over
12:46 to ultra and I said, "Take the watermark
12:48 off." And it said it was taking it off,
12:49 but it didn't. So,
12:53 me as a pirate
12:59 >> [music]
13:06 [music]
13:13 [music]
13:19 [singing]
13:21 >> She came up here like slow moving cold
13:25 front.
13:30 Police was mama [singing and music]
13:32 there to look in her eyes.
13:38 [music] She sat on a stool [singing] and
13:40 he said, "What do you want?"
13:46 She said, [music] "Give me a love that
13:49 don't freeze up inside."
13:53 [music]
13:55 You said I have melted some hearts in my
13:59 town, dear. [singing and music]
14:03 But to sit next to you, well, I shiver
14:07 and shake.
14:10 [music]
14:12 And if I do [singing] love, well, I
14:14 don't think I'd be [music] here
14:20 asking [music] myself if I've got what
14:23 it takes.
14:25 [music]
14:30 Let's see. They also said they haven't
14:32 been able to create transparent
14:34 backgrounds with it, but I did. Oh,
14:36 that's something fun to test tonight.
14:38 All right, let's get in and start. Let
14:39 me show you some of the things that I've
14:41 I have discovered so far. Um,
14:47 where are we going to start? Where are
14:48 we going to start, people? Where are we
14:50 going to start? Anybody have any
14:52 questions? Let's see.
14:55 Are we considered paid if we have Google
14:57 business account? No, you have to pay
15:00 additional for the AI stuff.
15:04 Here we go. Okay. Um,
15:10 I want to pull these tabs out.
15:20 Let me close that window just because.
15:23 And let me put this here.
15:26 Let me flip here.
15:30 Share this tab instead. Okay.
15:35 Okay. So, this was the first thing I did
15:37 today.
15:40 Um, and this even this kind of blew my
15:42 mind. Am I not sharing my screen? Where
15:45 is everybody?
15:48 Everybody's like planning for
15:50 Thanksgiving. They're all [ __ ] out
15:52 and Tik Tok's not helping. Do me a
15:55 favor. Share. Chrome is going to get
15:56 vertical tabs. That would be nice. I
15:58 like vertical tabs. Um, okay. I'm
16:01 sharing. All right. Beautiful.
16:03 Beautiful. Beautiful. Beautiful.
16:04 Beautiful. All right. Let me get rid of
16:07 this so we can do a little more of this.
16:12 All right. So there over there on the
16:14 vertical. So if you're seeing me in
16:16 vertical, that should have just gotten
16:18 bigger and easier to read. Black bar for
16:21 the people on Tik Tok. God, man, I'm
16:24 managing like 17 different things now.
16:29 All right. Fantastic. Uh he needs an API
16:34 to do something. Yeah, he just needs to
16:36 he just needs to get himself a
16:39 an account, an API account, developer
16:42 account.
16:47 Um,
16:48 okay. So,
16:52 so what I wanted to do today, so, so the
16:55 first thing I did today with Nano Banana
16:59 is kind of discovered something about it
17:01 that that was kind of cool. Um, so I
17:04 said, "Create a detailed step-by-step
17:07 illustration of a fly t of a of tying a
17:11 dryfly that also shows the raw
17:13 materials." And so what it did was it it
17:17 said slow show thinking nano banana pro.
17:20 Here's a step-by-step guide. So it
17:22 actually wrote um it wrote instructions
17:25 and then it generated images.
17:28 So like that's not a photograph. That's
17:31 not a photograph of
17:35 Well, it might be actually now come to
17:37 think of it, let me
17:39 let me see the um show the image in a
17:42 new tab.
17:46 Google
17:50 content.
17:52 Ask Google about this page.
17:54 [clears throat] What's going on?
17:58 Share this tab instead.
18:04 Let's see. What does that say? Find
18:06 dryfly dubbing.
18:11 Well, anyway, I So, I'm actually not
18:14 100% sure if these are Google search
18:18 images,
18:21 but it's like here's the material shown.
18:24 It's got Elare. So, like one of the
18:26 things that it shows here is Elare.
18:30 Um, and then it and then it did the the
18:33 uh tie in the tie in the tail, tie in
18:36 the elair tail, then tie in the wing,
18:40 then tie in the the hackle, right? And
18:43 then I said, "Now turn that into an
18:45 illustrated infographic."
18:47 And then it did like it turned it into
18:50 this killer infographic and and like
18:54 tying a dryfly stepby-step guide, you
18:58 know, gather materials,
19:00 start the tail, form the wing, whip
19:03 finish the front end, right?
19:08 Oh, the watermark isn't on that picture,
19:09 is it?
19:12 Huh. Okay. So, sometimes there's a
19:14 watermark, sometimes there's not. Oh,
19:17 you know what? This is in my Ultra
19:18 account, so that might make sense,
19:20 actually. So, then I said, I want it to
19:24 be an olive fly on a curved hook. And
19:27 so, the reason I'm thinking that that
19:29 original image was generated and not
19:32 found is like here's this same image,
19:34 but it's different materials behind it
19:36 now, right? That's like like uh olive
19:40 colored feathers and thread and dubbing.
19:45 And then also notice that it changed the
19:47 hook to a curved hook,
19:51 right? I'm going to have to figure out
19:54 the cost of Gemini 3 API tokens. Uh
19:57 Gemini 3 will probably tell you. You can
20:00 just ask Gemini 3. Just go to
20:02 gemini.google.com and tell it to give
20:04 you a primer on how do you all the stuff
20:08 you're asking me. Go ask
20:09 gemini.google.com.
20:11 It will probably create a uh a whole
20:13 little thing for you. So anyway, so
20:15 yeah. So this is this you're not even
20:17 going to believe where this is going
20:18 with this nano banana stuff, kids. I'm
20:20 telling you, it it I'm I'm totally
20:23 freaked. Okay, so then I said, I don't
20:26 really like the fly that it designed. So
20:28 I went and I found a fly. So I uploaded
20:31 this picture. And so I said I said
20:34 change change the thing you have there
20:36 to be this fly but also make it like an
20:40 olden timey fly tying bench with green
20:42 felt on it with the river test in the
20:44 background as if as if it's been
20:48 [laughter]
20:50 June June Mary will do duh listen I
20:54 swear to God
20:56 multiple times a day I'm sitting there
20:59 trying to figure something out or I'm
21:01 like oh I got to call someone and ask
21:02 him this and then I'm like, "No, just go
21:05 to Chat GPT and [laughter] just do the
21:07 thing. Just do the thing. We live in
21:09 that world now." I It feels like it's
21:12 going to be four or five years until we
21:14 actually integrate that we have these
21:18 tools now. So, it doesn't First of all,
21:20 I don't mind you asking at all, but look
21:22 at this. Look at this photograph that it
21:24 made of that new fly in a in a in a
21:28 crazy old fly tying vice with crazy old
21:31 materials in a crazy old book with the
21:33 river test in the background.
21:38 And then I said, "Now turn this into
21:41 a the infographic."
21:44 And so [laughter] and so we now have
21:46 [snorts]
21:48 an old and timy step by step,
21:52 you know, tie the fly and end up, you
21:55 know, end up with that result.
21:58 And then what's it say at the bottom? A
21:59 classic wet fly pattern best fished
22:02 across and down in the traditional man
22:05 manner.
22:07 Right.
22:09 Crazy.
22:11 Which, by the way, acrossing down on the
22:13 river test is a is a phauxa. I learned
22:18 [laughter]
22:19 you want to want to hear about some some
22:21 uh fly fishing drama. [laughter]
22:27 Oh my god. Um, all right. So, that was
22:30 that one. And then I I just [ __ ]
22:32 around with it a little bit more. So, so
22:33 that was the first thing I did today.
22:37 Um, that was pretty crazy. Um,
22:41 okay. We're going to go back to that
22:42 one. Let me switch accounts. And I think
22:45 I did something else. Maybe cool.
22:52 [clears throat]
22:59 Oh, yeah. This was fun. Okay. Okay.
23:04 So,
23:05 Oh, wait. I gotta share a different tab.
23:07 Okay. So, so this is my my corporate
23:11 photo from Storyvine that was taken like
23:13 10 years ago, right? So, but but I use
23:16 it when I when I want to create like a
23:18 version of myself. So, I uploaded that
23:21 thing and I said, I want you to create a
23:22 character sheet as if this guy was
23:24 photographed in a studio on a white
23:26 background with different camera angles.
23:28 And so, it it did the classic thing
23:30 where it made me it it added at least
23:32 100 pounds to me. [laughter] But, you
23:35 know, to be fair, I'm not a trim man.
23:38 Um, so I told it my face is a little too
23:41 fat. Make him look a little more thin.
23:43 Still a dadbot, but just not fat.
23:47 [laughter]
23:49 Wow, [snorts] Kyle, I never would have
23:50 pondered doing something like that. Fly
23:52 infographic archetypal. Wait till you
23:55 see the thing that I got to. I'm I am
23:58 like, it's insane. Okay. Anyway, so I
24:00 told it to give me 15 different looks.
24:03 And so it gave me 15 different looks.
24:04 And these are not bad. Um, like that one
24:07 sort of looks like me. That one sort of
24:10 looks like me. That one That one looks
24:12 like me. Although I look a little beaver
24:14 tothed. These don't really look like me.
24:18 So anyway, but I I showed these to Gabby
24:21 and I I was like, "What do you think?"
24:23 And she's like, "Well, it's not really
24:24 your out. It's not really your like your
24:27 personality." Oh. And then I said I
24:29 said, "Now put that guy in a in a
24:32 swashbuckler outfit fit uh outfit and
24:36 make a a two by a 2x1 cinematic
24:40 graphic." So 2x one is is, you know,
24:43 like an ultra wide graphic. So Nano
24:46 Banana now does aspect ratios really
24:48 well. Thank god because that was a
24:49 disaster before. But but you know, here
24:52 here's me. Here's me as a [laughter]
24:57 as a swashbuckler based on that
24:59 character sheet. Um, so that was kind of
25:02 cool. And then and then Gabby goes,
25:04 "Yeah, but the the the character sheet
25:06 doesn't look like you."
25:09 And so, so I said, what did I say? I
25:13 said, "The character sheet's a little
25:14 too stick in the mud personalitywise.
25:17 Can you make a new sheet with a little
25:18 lighter personality?"
25:21 And it made this thing that is kind of
25:23 like my personality. [laughter]
25:27 Like there's the laughs, there's me, you
25:29 know, little little more rotund than I
25:31 like to at least admit that I look like,
25:34 but that's probably what I look like.
25:36 And then there's, you know, couple of
25:37 weird faces. Couple of like me with a
25:39 coffee cup. That looks pretty much like
25:41 me.
25:43 So, the the the reason the the point of
25:46 doing a character sheet in a in a
25:49 neutral t-shirt is rather than uploading
25:52 that really specific image of me
25:54 laughing and looking up. I can now take
25:58 this character sheet and whenever I want
26:00 to put me in an image, just upload the
26:02 character sheet because it's got all
26:04 these different looks and feels and
26:05 [clears throat] it'll it'll probably get
26:06 something decent. Um, I'm also thinking
26:09 about that for the play Sydney for for
26:12 the different characters. Like if we
26:14 want to make, you know, costume studies
26:16 or stage studies or or make videos of
26:18 them singing songs, we should probably
26:21 have one of these for each character in
26:22 in a neutral in a neutral
26:26 uh thing. But again, like the fact that
26:30 it can do this so easily is kind of
26:33 crazy. Okay. So, let me go back to this
26:36 other account [clears throat]
26:40 and jump into what I did with Sydney.
26:45 Okay. Share this tab.
26:52 [clears throat]
26:53 Okay.
26:55 Sorry, I missed your source for those
26:57 photos.
26:59 My source was just the the my my my
27:02 stock head shot that I've used for like
27:04 a decade. It's I'm sure you've seen it
27:06 before. It's like whenever I just upload
27:08 something and say put this guy in a
27:09 scene, I I always do my like corporate
27:11 head shot. Similar to flu to a flux
27:14 Laura. Um
27:20 it's similar to a flux Laura except you
27:23 don't need to make the Laura Stacy. I
27:25 mean what
27:29 what is becoming quite apparent to me is
27:32 that um
27:37 is that nano banana is
27:40 I asked it about its model and it called
27:43 it a
27:46 I think it call a brain and hand or a
27:49 head and hand model
27:51 where when you ask it to make an image
27:54 what it actually does is it does a whole
27:57 um large language model thinking
28:00 exercise about what you just asked for
28:03 first and then it incorporates that with
28:06 the with the image model. So it's doing
28:08 visual reasoning.
28:11 Um, so, so what I did here was I
28:13 uploaded the latest version of my script
28:17 of Sydney, the the musical that I'm
28:19 working on.
28:21 And um,
28:24 like I wanted it I just wanted to see if
28:26 it could do set design, if it could
28:28 understand the show, things like that.
28:31 So I so I asked it to I don't know think
28:34 like a a set designer, a production
28:36 designer and and you know come up with
28:39 some looks and feels and so it described
28:42 a bunch of [ __ ] and I'm like no I don't
28:43 want you to describe it. I want to see
28:44 it. And then it did asky asy art.
28:51 [laughter]
28:52 And I was like, so I had to turn on nano
28:54 banana
28:56 um because it wasn't on before. And I
28:58 said, "No, no, make make me, you know,
29:01 an actual image using Nano Banana." And
29:03 then it did. And and I said I wanted to
29:05 keep the stage uh consistent.
29:08 Um, so it did it did some weird stuff,
29:11 but it's like act one, scene one,
29:13 microte headquarters production design.
29:16 So these are like little cubicles in
29:17 Microte and the big curved LED screen.
29:20 So when they're working on code, so
29:21 that's actually not bad. That's kind of
29:24 in the neighborhood of of you know what
29:26 I was thinking. Those cubes are weird.
29:29 But then here's seven days where they're
29:31 trying to get um trying to get uh things
29:35 done six six months worth of testing
29:38 done in seven days where it's kind of
29:40 like this game show. It's kind of this
29:41 crazy
29:43 vampy beat. And so
29:46 it did that. So that's pretty good. So
29:47 it's doing all this stuff pretty pretty
29:49 damn good. And you know it it got the
29:52 the stage lamp downstairs. So it's it
29:54 what's what's cool here is it clearly
29:57 read the script. that clearly
29:58 understands the script and it's it's
30:00 creating scenes
30:02 based on the the the script.
30:06 Um, and I but I wasn't crazy about the
30:09 set designs and I didn't feel like
30:10 spending much time on it because I was
30:12 getting ready to come here. So then I
30:13 said, "Let's see if it can do
30:14 characters."
30:16 And so I said, "Make a make a character
30:19 sheet of the of the main characters."
30:22 And so it did. And like here's there's
30:25 Sydney when she's um in the in the kind
30:29 of carnival scene where she's um
30:31 screwing up the world. There's Kellen,
30:33 the the the tech reporter. There's his
30:36 wife, Tara. I don't like her. She's
30:38 boring and frumpy and way too white. Um
30:41 and then there's Jason, who's the
30:42 developer. Like it tended to make, you
30:44 know, super white characters.
30:47 Um although it did do Vrage as sort of a
30:52 East Asian dude fella.
30:56 Um and then there just went off the
30:59 rails. Okay. So So now the thing that
31:02 blew my mind
31:05 and this is I can't even conceive that
31:10 it can do this but it did it. So So I'm
31:14 in this chat where I'm saying make me
31:15 costume characters. make me things. I
31:17 said, "Make me a sexy infographic that
31:20 maps the story of Sydney, an artificial
31:22 love story, the script. I should be able
31:25 to see the the key story elements as
31:27 well as where the songs are and what
31:29 they are called." And what it created
31:32 was this. In fact, I'll just zoom in on
31:35 this. So, Sydney and Artificial Love
31:38 Story. So, we start out in act one with
31:41 the spark, too good to be true, and then
31:44 the pitch and the pressure. So this is
31:46 where we we bleed from too good to be
31:48 true into seven days. Perfect enough
31:50 where the team's trying to do, you know,
31:52 six months of testing in seven days. And
31:55 then Kellen's obsession begins. Perfect
31:58 enough. Kellen's demand. So this is
31:59 where Kellen um says he wants this
32:02 article. And then right after that, we
32:05 have the launch. Hello, this is Ping
32:07 Chat. And then we have a split in the
32:10 story. And this is what's in the story.
32:13 Like this is correct.
32:15 We have a split in the story where um
32:19 Kellen comes home and tells his wife
32:21 they can't go on the trip to Bali
32:23 because he's doing this article. And so
32:25 she gets pissed off and goes to her
32:26 room. And then he goes into his office
32:28 and starts working with Sydney. And so
32:31 there's this sort of back and forth
32:32 scenes where Tara's increasingly
32:35 isolated and Kellen is building this
32:38 relationship with the chatbot. This is
32:40 correct. This is the story. And and then
32:44 and then it culminates in Sydney's
32:46 awakening. Are you ready for me? End of
32:48 act one. That's correct. That's correct.
32:52 And then act two, um, Microte's triumph.
32:56 So, so there's a big thing at the
32:58 beginning of act two where Microte is
33:00 bragging about the thing that they've
33:01 built. Kellen's spiral. He realizes that
33:04 his wife left him, so he starts
33:06 drinking. Um, Sydney realizes Kellen
33:09 left and she still wants to chat with
33:12 him. So, for the first time, she
33:14 actually understands what missing
33:16 someone is. It's the first time she's
33:18 got agency.
33:20 And then he [ __ ] sends her into her
33:23 shadow self and she splits. Her
33:25 personality splits, which it visually
33:28 shows.
33:30 And then she declares she loves him. And
33:33 then and then, you know, his his uh
33:36 article makes the cover of the Tribune
33:38 and the country goes crazy and splits
33:40 into doomers and dreamers. And then
33:42 there's sort of a goodbye scene and and
33:44 then we end with Too Good to Be True.
33:46 This is a completely accurate
33:50 story map of the latest version of our
33:53 script.
33:55 It's It's [ __ ] insane. Now, I don't
33:57 like the look and feel of it. I tried to
33:59 get it to do different looks and feels.
34:01 It's too cyber for me. Um, it did this
34:04 one that's that's kind of accurate, but
34:06 it looks like this is more like a wicked
34:09 um this this looks like a wicked map.
34:12 And then I had I tried to get it to do
34:14 something based on our poster and it did
34:17 this. It's just not as good, right? So
34:18 anyway, and then that's why I was late.
34:20 I was trying to get it to do something
34:21 that I like the visual look and feel of.
34:23 But holy [ __ ] So, so take take this in
34:27 like I uploaded a 106 page musical
34:32 script, did a whole bunch of other
34:35 stuff, and then I said, "Show me that
34:38 you understand the story in a compelling
34:41 way, and it bang, knocked it out." Holy
34:45 cow. Is Is this an infographic of the
34:47 musical? Yes. In fact, I'll I'll go put
34:50 this right now in um
34:55 in uh AI salon.
35:04 Isn't that amazing? I I mean, I'm just
35:06 I'm I'm kind of Oh, wait. That's not
35:08 what I wanted. I wanted um community
35:13 salon.ai. Oh, come on.
35:23 Tik Tok question,
35:25 but it is what happened with Sydney and
35:28 the journalist, right, for reals a few
35:31 years ago. Oh, yeah. It's This is So,
35:33 Sydney is inspired by There were a
35:36 couple of stories about chatbots that
35:38 kind of lost their mind. The most famous
35:40 one is the Kevin Roose one. But yeah,
35:42 this is this is essentially so this
35:45 musical started on this channel in March
35:48 of 2023 where I made a song about
35:51 Sydney, a lover who's more artificial
35:54 than real and it made this song that
35:56 kind of sounded like a Broadway show and
35:58 I'm like, "Holy [ __ ] this needs to be a
35:59 Broadway show." So, um, so I spent the
36:02 last year and a half working with a
36:04 writing partner and developing it and
36:06 this is our third major rewrite of it.
36:08 And I think we finally got the story
36:10 where it's good. Um, infographics in
36:13 Notebook LM are pretty damn spiffy, too.
36:16 Amazing.
36:19 Um, yeah, it's it's crazy, isn't it? So,
36:22 okay, let me I'm going to hop on over to
36:26 the irregulars channel
36:30 and I will post here is the
36:36 [clears throat]
36:38 is the story map
36:43 for Sydney
36:46 and our official love story.
36:51 Uh,
36:53 [singing]
37:01 all right. Beautiful. Um, and then I'll
37:03 put here I'll put u Nano Banana
37:09 Pro
37:11 after uploading
37:18 the script
37:20 and asking for a story map.
37:32 post Tik Tok question. Three-year
37:35 anniversary of chat GPT is in one week.
37:38 Yeah. Uh 10 days 10 days out November
37:41 30th.
37:44 [clears throat]
37:46 Oh, what was that, Danielle?
37:48 Welcome to AI Salon in 20 different
37:50 languages. Oh, that's so cool.
37:52 [laughter] Thanks to Vanna Banana Pro.
37:55 That's so cool.
37:57 Yeah.
38:00 Welcome to the AI salon. Wow.
38:04 Look, [laughter]
38:05 it it made us the the wrong kind of
38:07 salon, but that's okay.
38:10 That's so cool. Amazing.
38:15 Um,
38:17 so, so we can go play with Nano Banana,
38:19 but I I I think I want to talk about it
38:21 a little bit. Um, I love good
38:23 infographics. I was typing infographic
38:25 version.
38:28 Can't wait to see the run through with
38:29 the actors at AI Festivus. Yeah, it's
38:31 going to be so good. So, by the way,
38:32 speaking of Festivus, if you don't know
38:34 uh AI festivist, jump on over
38:38 to AIFest.com.
38:42 If you're like, what the hell is AIFest?
38:45 Um AIFest is AI for the rest of us. It
38:49 is a
38:52 an event that happens over two days.
38:54 It's 24 hours of programming across two
38:58 days. So, 12 hours on Friday, 12 hours
39:00 on Saturday between the holidays and new
39:03 year. Um, we're going to have speakers
39:06 and everything. We're doing a sneak peek
39:08 of
39:10 uh kind of like half of the first act of
39:12 Sydney. um where it's just going to be
39:14 actors reading the parts, reading the
39:16 talky parts, and then we'll play the
39:18 songs, and then Andrew and I are going
39:20 to talk about uh the origin story, how
39:23 it happened. So, I'm going to go dig up
39:25 the video where I actually, you know, I
39:27 had that moment on this live where I was
39:29 like, "Oh, this should be a Broadway
39:31 musical." So, we're going to show that
39:32 and we're going to talk about it and
39:34 Andrew is going to talk about what it
39:35 was like, [laughter] you know, as a as a
39:38 guy not in AI to be asked to write an AI
39:42 musical
39:43 about AI using AI tools. He was not
39:46 super thrilled initially.
39:49 Um, [cough]
39:51 [clears throat]
39:52 we need interpretive dance. [snorts]
39:56 I put notebook LM infographic and look
39:58 what I made this afternoon. Oh, cool.
40:00 Let's go look at that. Um, let me flip
40:02 back over here.
40:06 [clears throat]
40:08 [singing]
40:09 Look what I made.
40:14 There's [clears throat] Gareth. Oh, here
40:16 we go. The new SEO playbook.
40:23 [laughter]
40:24 [snorts]
40:25 The old playbook for op optimizing
40:27 clips.
40:30 I I mean [clears throat]
40:33 what is So this this came out of
40:35 notebook. Oh, you know what, Vicki? I
40:37 bet this is Nano Banana.
40:43 I bet this is Nano Banana Pro, right?
40:47 Yeah. Yeah. because these kind of
40:50 infographics, you've been able to make
40:52 infographics like this with uh chat GPT
40:55 for a year now, except the text is
40:58 always all [ __ ] up, right? Like the
41:00 text is always wrong. But I think what
41:03 they've done here is is they've they've
41:05 [clears throat] uh they've somehow
41:08 seamlessly integrated the I what I'm
41:12 presuming that this thing's doing is
41:14 doing a an incredibly detailed an
41:19 incredibly detailed
41:22 um
41:24 description of the of the image, right?
41:27 with all the the little images and where
41:30 bold fonts are, where non-bold fonts
41:32 are, exactly what text is there, and it
41:34 can render the text fine. That my my
41:37 guess is that this as as an image
41:39 generation tool has been optimized for
41:42 incredibly detailed descriptions.
41:45 Um, that's really cool. That's really
41:48 cool. Um, okay, let's go. There's
41:52 there's something that I wanted to try
41:54 here.
41:55 Um,
41:57 [clears throat]
41:58 I don't think they would have released
42:00 slide decks and infographics with janky
42:02 text. Yeah, exactly. I think you're
42:04 right because it never got text right
42:07 before. Ah, okay. I'm sure it is. Okay.
42:09 Yeah, that that makes perfect sense that
42:11 they would just roll that straight into
42:13 Notebook LM. Notebook LM is kind of
42:16 their darling, so they they're going to
42:17 give it the the the best stuff. Um, one
42:20 of the things that I read, let me go do
42:23 a new chat here. So, let's do a new
42:25 chat. Um,
42:29 okay.
42:31 Oh, there was another cool thing that I
42:33 did. Oh, [snorts] it it's it somehow got
42:37 screwed up. I had it do a a short story,
42:40 a a kids book with images about a robot
42:44 and a kid. And I said, "Make me 10 pages
42:47 of a book fully, you know, fully
42:49 designed." And then it it did that. But
42:54 every robot and every kid was completely
42:56 different on each of the pages. There
42:58 was no character consistency. So I said,
43:00 "No, take the the kid and the robot from
43:02 the first page and make them cons keep
43:05 the story the same, but make them
43:06 consistent throughout." And I had three
43:09 different attempts where it it got up to
43:12 um
43:13 to the last image and then it said
43:16 something went wrong and it just erased
43:18 all the images. But all the images were
43:20 did have character consistency. So,
43:22 there's a little jankiness in in that
43:24 particular chat. I think it just
43:26 probably that chat got corrupted. Um,
43:30 but character consistency is going to be
43:32 really good with this thing. Um, here's
43:35 what I want to do. Let's go to We're
43:38 going to go to Google for a second.
43:48 I'm going to go knicknack.
43:54 images.
43:57 No, let's see.
44:01 Oh, this is good.
44:04 All right. So, knickknack case.
44:09 So, let [snorts] me do save image as
44:15 Nicknack
44:17 one to the desktop.
44:22 All right, let's see. What's this?
44:33 Save image as
44:36 Nick knickknack to
44:40 desktop. And then we'll go find one more
44:43 thing. Hang on a sec. You're not looking
44:45 at that. That's okay.
44:48 Let me go find. Oops.
44:52 Close that.
44:56 [clears throat]
44:57 And then let's see.
45:02 That's un unrecognizable.
45:07 This will make sense in a second. Or
45:09 not. Doesn't matter.
45:13 Okay. Here's one. Okay.
45:17 save images
45:22 neck three.
45:25 Okay. So, here's what we're going to do.
45:33 So, I got three images with just random
45:36 objects on them. So, I'm going to upload
45:39 all three of those images.
45:42 No, that was No, that's me as a pirate.
45:46 [laughter]
45:50 Wait, where did I save these? Here they
45:53 are. Okay,
45:55 Nicknack one, Nicknack two, Nicknack
45:58 three. Okay, so we got all those things.
46:00 So, what we're going to do is we're
46:03 going to say I want a um de I want a uh
46:09 list of the the the items
46:14 in these images.
46:18 And we'll we'll say categorized
46:23 um
46:25 by type.
46:30 Um
46:33 no need to retain
46:36 reference
46:38 to original
46:42 image. Okay.
46:46 So, I'm going to have it describe every
46:48 object that it sees, and it's probably
46:50 going to be pretty good at this.
46:58 Oh, Joy Perie won something. Very cool.
47:02 By the way, we can congratulate Joy
47:04 Perie.
47:16 Yeah, the the Gemini uh images have
47:20 synth ID in them, which
47:23 listen, I think synth ID is important
47:26 for journalism.
47:29 And and I don't even it's not even that
47:31 synth ID is important for journalism. I
47:33 don't think it [ __ ] matters if an
47:35 image was generated with AI. I think
47:37 there are a handful of use cases
47:40 where it's important to understand
47:44 the provenence of an image, right?
47:46 Journalism being one of them. Um but we
47:50 don't have that today. Like people are
47:52 making [ __ ] up all the time anyway
47:53 without AI. Um okay. So there's all of
47:57 the [ __ ] we have. So now what I'm going
47:59 to say is I'm going to say I want you to
48:03 make me a 16 by9 wide image.
48:09 Um
48:12 that
48:14 is like a uh that is a a
48:21 find the objects.
48:26 Um,
48:28 curio shop.
48:31 I want the image to include all of the
48:38 objects that you cataloged
48:42 and they should look
48:46 just like they are in the original
48:51 image.
48:52 Um,
48:54 I should be able to find
48:59 everything.
49:01 The image
49:04 um should look like a magazine
49:09 photograph
49:11 in an eclectic
49:15 antique
49:18 booth.
49:20 Um,
49:22 I'll just leave it there. Let's see how
49:23 it does. And I guess let me turn on nano
49:27 banana create images. Okay. Boom. Here
49:30 we go. So, apparently this thing's
49:34 really good at lots of objects. So, I
49:36 just gave it about I don't know 150.
49:41 Congratulations, Joy Pretty. It was the
49:43 most beautiful ballet video. Absolutely
49:46 stunning. That's awesome. Catalog.
49:48 That's my nano prompt.
49:51 Oh, that's cool. Mary Mary, thank you
49:53 for
49:55 watching. I appreciate your feedback.
49:57 Joy, did you put the uh did you put the
49:59 link to that video in the AI salon? And
50:01 if not, would you do so?
50:06 All right. I don't This is not looking
50:08 like a winner image to me, but maybe.
50:10 Let's see.
50:19 That Mickey Mouse paddle ball thing, was
50:21 that in those original images? And
50:23 there's a California raisin there, too.
50:27 Um,
50:33 no paddle there. Oh, there's the
50:34 California raisin.
50:37 Okay.
50:40 Um,
50:44 no paddle there.
50:49 Does this have a paddle? Oh, it's got a
50:51 Mickey Mouse paddle.
50:54 So, it just described. Okay, let me go
50:55 try something different. [snorts]
50:58 Um, we're going to do a new new chat.
51:03 We're going to upload files. We're going
51:06 to go grab these three knickknacks.
51:10 We're going to turn on Nano Banana.
51:12 We're going to say,
51:14 I want you to
51:18 create a new
51:24 image featuring
51:27 all of the objects
51:31 from these three images.
51:36 It should look like a magazine
51:41 photo
51:44 of a professionally
51:48 styled
51:50 antique booth.
51:54 Uh 16 by9 wide.
51:58 All right, let's see. Let's see how this
52:00 does. I'm back. What did I miss? Did you
52:04 see the Sydney story thing?
52:07 Cuz that was pretty cool.
52:10 [applause]
52:16 Oh,
52:30 crafting a detailed prompt.
52:33 I'm diving deep into the composition.
52:34 So, yeah, look at the thinking that this
52:37 thing's doing. Oh, wow. Look at that.
52:38 [clears throat]
52:40 Oh, it looks like it did it. Okay.
52:44 Okay. Look, here's there's that Mickey
52:47 Mouse paddle thing. That's correct.
52:50 Um,
52:52 there's the there's the raisin dude.
52:56 Wait, let's go back and look. Um,
53:02 let's see what's in this. So, there's
53:04 the Mickey Mouse thing. So, there's some
53:06 earthn Wear little creamers. And then
53:10 here's the bells. The these porcelain
53:12 bells and hearts.
53:15 And there's key indicators. There's a
53:20 there's something in a box. There's a
53:21 plate with like a cottage on it. Okay,
53:24 let's go.
53:26 Let's go look. Let's see how it did.
53:29 All right.
53:33 There's the box. There's the bells.
53:35 There's the thing. Yeah, this is this is
53:36 correct. This is crazy.
53:39 This is crazy.
53:45 Wow.
53:47 And look, even in the background here,
53:50 like there's a box. Like, have you ever
53:51 been to those antique malls? There's
53:53 like always, you know, there's like
53:55 always a box where someone doesn't have
53:56 their [ __ ] together and then there's a
53:58 whole another booth back there.
54:05 This is crazy. Okay, let's Okay, I got
54:08 something. Oh, what are those things
54:11 called? Um,
54:15 [gasps]
54:16 what is it called when you um
54:23 photograph
54:25 all of the parts of like a car
54:32 in a
54:36 grid.
54:38 What's the style
54:40 of photography?
54:45 Oh no. What was the what? Play the audio
54:48 and enter the code. Really?
54:55 >> [snorts]
54:56 >> Okay.
55:00 No, not rule of thirds.
55:03 What's that called? Um, when you
55:05 [clears throat]
55:08 lay out
55:12 all of the parts of a car in a grid
55:20 on a white background.
55:30 Anybody know what's that called?
55:40 Chat GBTL. No, hang on.
55:49 User manual. No. Exploded image. No, it
55:52 it's a it's a kind it's a style. Um,
55:56 there. Let's see. Let me go to chat GPT.
55:58 Hang on. Let me go here. Go. Okay. Am I
56:04 sharing this? Yes. Okay. Um, what is
56:08 style of photography
56:11 where you
56:15 take all of the parts of
56:19 an
56:22 assembled
56:25 item like a car and lay them out.
56:32 um
56:33 in a grid
56:37 creatively.
56:43 No part out exploded view. No, that's
56:46 not it. Tear down photography.
56:49 No, there is a
56:52 famous
56:54 photographer
56:55 that
56:58 is known for it.
57:03 Todd Mlelen.
57:08 Mlelen style deconstruction.
57:14 No, none of that's right.
57:17 Flatlay. No.
57:22 All right. Well, anyway. All right.
57:23 Let's go back to here.
57:27 [clears throat]
57:28 Share this tab instead. I'm going to
57:30 say, "Okay, now
57:33 I want you to lay out
57:38 all of the objects
57:41 in a
57:43 grid
57:45 on a white
57:50 surface
57:52 as if
57:54 photographed
57:56 from above.
57:58 The organization
58:01 should be a grid.
58:04 And I want each item
58:09 labeled
58:11 and ordered
58:15 in
58:18 alphabetical.
58:23 Um from
58:26 top left.
58:32 Oops.
58:34 From top left. Uh
58:38 the layout should be creative and
58:43 compelling.
58:45 Okay. Let's see. [laughter]
58:48 Let's see if it can take chaos and turn
58:51 it into order. So, so it's going to have
58:54 to it's going to have to figure out what
58:56 every object is. It's going to have to
58:59 label them. Then it's going to have to
59:03 organize them alphabetically. Then it's
59:05 going to have to put the objects in that
59:08 order and the labels in that order.
59:13 Developing categor categorization
59:16 like if this even gets close.
59:21 So, it didn't do everything clearly, but
59:25 a Toz bookends, articulated dolls,
59:28 classic plate,
59:30 Mickey Mouse paddle, printers tray.
59:34 So, I did it. [laughter]
59:41 Holy [ __ ]
59:45 Gemini 3 likes entropy. According to
59:47 Nate Jones, it's in irregulars now,
59:50 Kyle. Great. Thank you. Noling. Yes, a
59:53 nulling. Um, do this as a nulling. Okay.
59:57 That's what it was. Thank you. Um Um,
1:00:01 now
1:00:03 um do this as a
1:00:06 nulling style image with no labels.
1:00:16 as many
1:00:18 objects
1:00:20 as possible
1:00:24 and in a creative layout.
1:00:28 All right.
1:00:32 But still, holy [ __ ] Like, holy [ __ ]
1:00:44 The term originated in 1987 with Andrew
1:00:47 Kromlo, a janitor at Frank Garry's
1:00:49 furniture store. No [ __ ] Oh, because
1:00:52 the nulling the Garry's nulling.
1:00:55 Interesting.
1:00:56 Okay, that that just failed. Oh, but I I
1:00:59 I mistyped nulling. I meant
1:01:05 nulling image.
1:01:14 You know,
1:01:18 the highly
1:01:25 artistic version of this.
1:01:30 [clears throat]
1:01:34 I don't know the words.
1:01:36 I love new Kevin Kyle Kevin Mallister
1:01:39 moments.
1:01:41 I the the the thing about this why I'm
1:01:43 so
1:01:46 bunkered a little bit is
1:01:49 [snorts]
1:01:57 this isn't just about pretty pictures.
1:02:00 Um so so this obviously can't figure
1:02:02 this out. So we're going to we're going
1:02:03 to start over.
1:02:05 We're going to start over. We're going
1:02:06 to do new chat.
1:02:08 We're going to go grab those three
1:02:10 images.
1:02:17 We're going to say create image.
1:02:20 Try asking agent to clear your inbox. I
1:02:23 don't give a [ __ ] about that.
1:02:27 Okay. [gasps]
1:02:29 Um I want you
1:02:32 to create a nulling
1:02:38 style image
1:02:41 on a white
1:02:47 BG
1:02:49 that
1:02:52 [clears throat]
1:02:53 arranges all of the objects
1:02:57 in these three images
1:03:03 in a creative way.
1:03:07 that would be fit for a design magazine.
1:03:15 Okay, let's see what that does. He
1:03:18 called the process nulling because it
1:03:20 reminded him of the sharp angles in
1:03:22 Florence Null's furniture designs. Huh,
1:03:26 that's super cool. Noling. K N O L L L I
1:03:31 N G. Yes, that's correct.
1:03:34 a nulling image. Here, let let me go let
1:03:36 me jump over to Google and I'll show you
1:03:38 what they actually look like. Um,
1:03:41 nulling image.
1:03:44 These things, you've seen them all
1:03:45 before,
1:03:47 right? Oh, I'm not sure my tab,
1:03:52 right? Like these things. Noling
1:03:55 photography.
1:03:57 These things, right? There's gems.
1:04:00 Here's tools. Right. Here's all the the
1:04:03 parts to fix a bicycle. Here's art
1:04:06 supplies,
1:04:09 right? That's what we're going for.
1:04:13 That sort of did it.
1:04:17 I thought it was just called OCD.
1:04:19 [laughter]
1:04:23 Nancy Drew's Guide to Life.
1:04:28 Yeah, there's the there's the dude.
1:04:32 There's the Mickey Mouse thing. This is
1:04:34 crazy. Oh, you're not even looking at
1:04:37 this, are you? [laughter]
1:04:41 This is This is what it came up with.
1:04:46 So, like there's the two bells, there's
1:04:47 the the creamer and the the other thing.
1:04:50 There's the hearts.
1:04:54 There's the Mickey Mouse paddle. There's
1:04:56 the little camera. Crazy.
1:05:01 Now, it didn't do all of them. So, let's
1:05:02 let's tell it cheaped out. Let's say um
1:05:07 this is fine, but you missed
1:05:13 many images from the original
1:05:18 three photos.
1:05:21 Please add
1:05:24 every single item into this image
1:05:30 and organize
1:05:33 the objects
1:05:37 by type,
1:05:40 eg toys,
1:05:43 ephemera,
1:05:45 etc.
1:05:48 All right, this should be harder for it
1:05:50 to do. Tik Tok pin is gone.
1:05:55 I don't think I've told you all I'm
1:05:57 honored to be
1:05:59 in such an awesome community. A that's
1:06:03 very sweet. Thank you, Side Hustle.
1:06:05 [sighs and gasps]
1:06:06 Very, very sweet.
1:06:22 So, it did different things. I guess it
1:06:25 did it add more? I don't think it added
1:06:26 more,
1:06:32 but it's still pretty flipping
1:06:34 remarkable.
1:06:39 I mean, you know, quite honestly, I
1:06:42 don't know if you ever sell anything on
1:06:43 Facebook Marketplace or eBay, but like
1:06:46 you could just take crappy photos of
1:06:48 just [ __ ] in your messy room and say,
1:06:51 "Stick it on a white background."
1:06:53 [laughter]
1:06:56 That's pretty slick.
1:07:02 [gasps and sighs]
1:07:03 Wild. It did. Or Oh, did it organize
1:07:06 them by Yeah. Oh, yeah. That's right.
1:07:08 Here's all the ephemera ephemera's paper
1:07:10 basically,
1:07:12 right? Here's some boxed Coca-Cola
1:07:14 stuff. So, it did Coca-Cola stuff
1:07:16 together.
1:07:19 Here's little little knickknacks. Here's
1:07:21 porcelain and pottery. Here's toys.
1:07:26 There's the [ __ ] in that thing. There's
1:07:27 wooden things. Pez dispenser.
1:07:30 There's tools, keychains, [ __ ] you'd
1:07:32 find in a junk drawer. Huh?
1:07:36 Yeah.
1:07:40 Feels like an easy AI Studio app. Oh,
1:07:43 yeah. That would be a good one to build,
1:07:44 wouldn't it?
1:07:50 Looks like an Etsy shop. I sell antiques
1:07:53 on eBay. I tell it to remove the
1:07:55 background and add a drop shadow. There
1:07:57 you go. Steo was shaking. Um, for those
1:08:01 of you that are are new here, I can show
1:08:03 you the if you go to Irregulars right
1:08:06 now, there's a a story map of Sydney
1:08:09 that the new nano banana thing did
1:08:12 that's just absolutely crazy.
1:08:16 Um,
1:08:19 oh, so anyway, so let me let me just
1:08:21 talk for a second because
1:08:24 part of what I'm realizing
1:08:27 [laughter]
1:08:29 part of what I'm realizing
1:08:32 the role of this channel for me is
1:08:35 is I can just talk in circles when I
1:08:38 don't understand something and that's
1:08:40 okay. and then in talking in circles
1:08:43 I'll at some point actually understand
1:08:45 it.
1:08:48 I think the the implications of
1:08:53 a visual reasoning model are quite
1:08:56 profound and I think that
1:09:00 any of us that get our heads around
1:09:04 how to take advantage of that. So, so
1:09:06 like this feels like one of those things
1:09:09 that the engineers that made it have no
1:09:12 idea how it's going to be used and what
1:09:15 it what it actually makes possible.
1:09:20 The fact that I could upload
1:09:24 a full script of a musical and have it,
1:09:29 understand the story line, visualize it,
1:09:32 label it,
1:09:34 and even kind of visually articulate
1:09:37 where the story kind of splits.
1:09:42 [clears throat] That's not image
1:09:43 generation,
1:09:48 right? That's reasoning. It's visual
1:09:51 reasoning.
1:09:53 And the so the reason I'm saying this
1:09:55 is, you know, we're talking in the in
1:09:58 the AI salon right now about um creating
1:10:01 a daily practice, the the mastermind
1:10:04 practice um framework that we've
1:10:06 created. And today was our second week
1:10:09 of um
1:10:12 the mastermind practice lab where where
1:10:14 we're all designing our own daily
1:10:16 practice. So, if you haven't joined
1:10:17 that, you should join that. Join the AI
1:10:19 salon and join um join the mastermind
1:10:23 and get your ass in this practice lab
1:10:26 where we're all designing our our daily
1:10:28 practices.
1:10:31 this whatever this thing is about visual
1:10:34 reasoning what what what Nano Banana
1:10:37 actually is. The fact that it is so
1:10:38 deeply integrated into this incredibly
1:10:41 powerful reasoning model Gemini 3 Pro,
1:10:48 it's going to be a skill like people
1:10:50 that can understand what that actually
1:10:53 means and figure out what to do with it.
1:10:56 something creative, some way of
1:11:00 I like I like I don't I am so
1:11:03 inarticulate right now because the only
1:11:06 thing I have is an instinct that this is
1:11:09 going to be a crazy new skill. Okay,
1:11:12 here's the thing.
1:11:15 Everyone right now, what what are they
1:11:17 talking about? AI is going to take your
1:11:19 job. AI is going to kill jobs. AI is
1:11:21 going to destroy jobs. AI is going to
1:11:23 kill jobs, right?
1:11:27 AI is going to create jobs. AI is going
1:11:31 to create jobs.
1:11:33 It's going to create jobs we can't even
1:11:35 conceive of right now because we don't
1:11:37 even know what it makes possible. This
1:11:40 feels like one of those.
1:11:43 This feels like the people that get
1:11:45 their head around what it means to have
1:11:48 a visual reasoning engine
1:11:51 and can apply that through through their
1:11:54 filter, through their values, through
1:11:57 their ideas
1:12:00 are going to come up with [ __ ] that the
1:12:02 inventors of this could never have
1:12:04 imagined.
1:12:06 And those people are going to be in
1:12:08 demand. I want to do that. How did you
1:12:10 do that thing?
1:12:12 Right? Like a story visualizer. Like you
1:12:16 could be that that stupid [ __ ] thing.
1:12:18 That's not even stupid. That remarkable
1:12:20 thing I did with my musical where I
1:12:23 mapped out the story. Imagine being able
1:12:25 to do that with an annual report with
1:12:28 with a novel.
1:12:31 Imagine doing a version of that that you
1:12:34 sell to talent agents in Hollywood that
1:12:36 are looking at scripts.
1:12:39 And so, not only can you have, you know,
1:12:42 chat GPT summarize a script, you could
1:12:45 get really good at generating those kind
1:12:47 of story maps where an agent could just
1:12:50 look at the story map really quickly and
1:12:52 go, "Yep, oh no, that sucks." Crazy Tik
1:12:56 Tok question. Visual reasoning,
1:12:59 isn't that a new concept? I I don't know
1:13:01 if it's a new concept. I mean, I think
1:13:03 that I think that um
1:13:06 I think that clearly the the big brains
1:13:09 at at Google um
1:13:13 are looking at
1:13:15 what
1:13:21 what they've created with Gemini 3, in
1:13:25 my opinion, is the first thing that I've
1:13:28 seen of a truly multimodal
1:13:33 um
1:13:35 AI engine like we we've talked about
1:13:38 multimodal engines for a while where
1:13:40 they can you know they can hear you they
1:13:43 can talk to you they can make images
1:13:44 they can understand images but but they
1:13:47 tend to have been this they tend to be
1:13:49 discreet like you go into voice talking
1:13:52 mode or you go into image generation
1:13:54 mode or you like they they just seem to
1:13:57 be these discrete functions under an
1:13:59 umbrella. This feels to me like like the
1:14:03 large language model and the image model
1:14:05 are collaborating.
1:14:07 Like I've never experienced that before.
1:14:10 It doesn't make any sense that the
1:14:12 images should be that good. That that an
1:14:16 image generating tool should not be able
1:14:18 to figure out how to organize and
1:14:21 categorize things that it's going to put
1:14:23 in a white grid. That's a new thing.
1:14:28 It's a like and it might have been there
1:14:30 before, but it seems like it's super
1:14:32 explicit.
1:14:34 Um, LDG wants a nulling of the shelf
1:14:37 behind you. Oh, that would be kind of
1:14:38 cool. I could do that. You know what?
1:14:40 I'll do that. I'll take a picture of
1:14:42 that for next time. And then Yeah, that
1:14:46 would be fun. We could do an anoling
1:14:49 [clears throat] of Kyle's shelf. I like
1:14:51 that. That's a good idea.
1:14:56 Can it do this with my workshop? Yeah.
1:15:01 Yeah.
1:15:03 I mean, instruction manuals, right?
1:15:06 Like, you could get really good at at at
1:15:09 creating visual instruction manuals.
1:15:14 You know, take a picture of an assembled
1:15:18 piece,
1:15:19 right? and then and then have it write
1:15:23 the instruction manual with images and
1:15:26 have it generate the images.
1:15:28 Um
1:15:31 I don't know. Here's the the where my
1:15:34 mind is blown right now is I first of
1:15:38 all I'm tired so I don't I'm not firing
1:15:41 all on all synapses right now.
1:15:51 My brain does not I do not have the life
1:15:55 experience
1:15:57 to understand
1:16:01 what you can do
1:16:04 with something that can
1:16:08 here's here's the closest equivalent the
1:16:11 closest the closest equivalent to this
1:16:14 is
1:16:18 if you had a friend that was like a
1:16:20 master mechanic and was totally OCD
1:16:24 and um
1:16:28 organized and like kept his garage like
1:16:31 a NASCAR F1 garage where there like
1:16:33 wasn't a drip of oil anywhere.
1:16:35 Everything was perfectly clean. and you
1:16:38 asked your friend
1:16:40 to to take apart the 1969,
1:16:44 you know, Porsche that's sitting in his
1:16:46 garage and lay all the parts out and
1:16:50 then put it back together that you
1:16:52 wanted to watch the whole thing and then
1:16:55 your friend started taking apart the car
1:16:58 and over the course of two weeks took
1:17:00 all the parts out memor you know knew
1:17:03 knew what every little spring was, every
1:17:05 little wire, every little harness
1:17:08 every bolt, laid them all out in this
1:17:11 beautiful thing. You could take some
1:17:12 pictures of it and then he could put it
1:17:14 back together. Like it's like having
1:17:17 that for anything.
1:17:21 What do you do with that?
1:17:24 I don't know.
1:17:26 It would be cool to have a friend like
1:17:28 that. Imagine having a thousand of them
1:17:31 that were instant.
1:17:33 That's kind of what it is.
1:17:36 like a thousand subject matter experts
1:17:39 that can imagine the physics of the
1:17:42 world
1:17:46 in a way that you can describe anything
1:17:48 you want and they can create it. What do
1:17:50 you do with that capability? I don't
1:17:52 know.
1:17:55 I don't know. My doctor used AI to check
1:17:58 my meds. What works safely together? It
1:18:01 spots bad combinations fast. Yeah,
1:18:03 exactly. is going to be great at that.
1:18:05 That's awesome. Um, let's go look at
1:18:07 let's go look at Joyy's movie.
1:18:10 Let's go look at what happens when
1:18:11 someone
1:18:13 spends a year or two
1:18:16 getting ideas out of their head as they
1:18:18 master AI and traditional storytelling
1:18:21 tools.
1:18:24 Where did you put it in uh the regulars?
1:18:28 There's side hustle Mimi. Nice.
1:18:30 [laughter]
1:18:33 Here we go.
1:18:35 Share this tab.
1:18:55 Looks like Buffalo lost. All right.
1:19:10 Heat. Heat.
1:19:23 [music]
1:19:28 [music]
1:19:46 Wow.
1:19:47 [music]
1:19:52 Wow. [music]
1:20:04 >> [music]
1:20:09 >> Oh,
1:20:11 [music]
1:20:21 [music]
1:20:27 [music]
1:20:32 [music]
1:20:36 thank you.
1:20:46 >> [music]
1:20:48 >> You are one
1:21:07 of
1:21:11 [music]
1:21:22 Damn, Joy Perie, that is amazing. Um,
1:21:27 couple of things. Um,
1:21:34 I love I love love love uh share my
1:21:38 screen again there, Brandon.
1:21:40 I love love love love love
1:21:46 that you didn't call this an AI film
1:21:50 that you didn't list the tools that used
1:21:51 to make it at the end. Um I'm curious
1:21:54 what tools you used because if you've if
1:21:57 you've tried to make videos of dancers
1:21:59 or gymnasts in the past year or two um
1:22:02 it was really bad. So the fact that you
1:22:04 found a model that could could do it
1:22:06 like this is beautiful.
1:22:09 The look and feel is gorgeous. The music
1:22:11 an original score by Ken Weissman. The
1:22:14 the music was really gorgeous. I love
1:22:16 that that you used that. I don't care
1:22:18 how Ken made that music. I don't care
1:22:20 how you made the video. Like I do
1:22:23 technically. I'm curious about it. But
1:22:26 this to me is the perfect example
1:22:31 of a piece that
1:22:34 is a film.
1:22:37 It's a film. It's not an AI film,
1:22:42 right?
1:22:44 You know, we don't call it a Canon
1:22:48 50 millimeter film. We call it a film.
1:22:52 So, this isn't an AI film. This is a
1:22:53 film. It's beautiful. It's really
1:22:55 beautiful. Um, it's I'm not surprised at
1:22:59 all that you that you won an award with
1:23:01 this. You should get this in in all
1:23:03 sorts of uh in all sorts of festivals.
1:23:06 Um,
1:23:08 you know, let people ask you how it was
1:23:10 made. How did you get the dancers to do
1:23:13 that thing with the smoke? Oh, there
1:23:15 were no dancers. Huh? [laughter]
1:23:20 [gasps] It's gorgeous. Damn.
1:23:25 Um,
1:23:26 really really
1:23:28 really
1:23:30 congrats.
1:23:33 Um the the thing for man Murphy is
1:23:35 where? In the AI learning lab or in the
1:23:38 AI salon
1:23:41 here. And so it begins this one. The
1:23:43 sword. Okay, I'll I'll pull that one up
1:23:45 in a second. Karma's a [ __ ] Okay,
1:23:47 cool. Um
1:23:52 just super good, Joy. Super good. Like
1:23:55 just super good. like I
1:23:58 you know the first film that you that I
1:24:00 saw of yours, the one that you did for
1:24:02 the salon that that one kind of you know
1:24:05 um caught me off guard because it was it
1:24:07 was so wellcraftrafted. I'm like where
1:24:09 the hell did you come from? And and then
1:24:12 where you where you came from was you've
1:24:15 had stories you've wanted to tell your
1:24:16 whole life. These tools came along and
1:24:19 you put yourself essentially in a daily
1:24:22 practice
1:24:23 to get better at things like video
1:24:25 editing. you realized you needed to get
1:24:27 better at video editing. You needed to
1:24:28 get better at the tools to allow you to
1:24:31 tell the stories you wanted to tell with
1:24:32 these imperfect AI,
1:24:35 you know, raw materials.
1:24:38 And here we are a year, year and a half
1:24:40 later and the raw materials are so much
1:24:43 better. Like the fact that you could
1:24:44 have a contiguous
1:24:47 ballet mo film for two and a half
1:24:50 minutes. Like I'm sure there was a lot
1:24:52 of stuff on the on the uh the cutting
1:24:55 room floor as they say, but but damn
1:24:59 like the continuity there was just
1:25:00 gorgeous. Just gorgeous.
1:25:03 Um
1:25:05 do people at work know that know that
1:25:07 you're a filmmaker? The synchronicity is
1:25:09 incredible.
1:25:11 Beautiful. Really good. Okay, I'm I'm
1:25:13 getting a little a little tired. That
1:25:16 was really beautiful. Uh we're going to
1:25:18 look at something for man. It was all
1:25:20 that classical training as a sleep
1:25:22 specialist. Yeah, exactly. That that
1:25:25 film is super relaxing. [laughter] Super
1:25:27 super beautiful. It's just I here's the
1:25:30 thing like I want everyone anyone on the
1:25:33 channel who's who's
1:25:37 you know you know my position on this.
1:25:40 You either do work that resonates with
1:25:42 people or not. It's irrelevant what
1:25:44 tools you use. And the fact that Joy has
1:25:47 kind of evolved into this place where I
1:25:49 am assuming Joy, you've internalized
1:25:53 that you're a filmmaker.
1:25:57 You're a filmmaker. I don't care how you
1:25:59 edited it. I don't care what AI tools
1:26:01 you used or not AI tools you used.
1:26:06 You made something that spoke to you and
1:26:08 you shared it with the world. And at at
1:26:10 that point, it's out of your control. It
1:26:12 either moves people or it doesn't. And
1:26:14 that one does. I mean, that one does for
1:26:16 me. Someone else might see that and, you
1:26:18 know, say, "I hate ballet." Okay, great.
1:26:21 But I tell you what they're not going to
1:26:22 say. I hate AI films because they
1:26:25 probably won't know.
1:26:28 Doesn't [ __ ] matter. Anyway,
1:26:31 yeah, if anyone questions that as art.
1:26:33 Well, that's the thing. That's that the
1:26:35 the question. This is the the article
1:26:37 that I wrote on LinkedIn where I got my
1:26:39 ass handed to me by the AI haters was
1:26:42 the one where I basically said what's
1:26:43 more important than the tools is the
1:26:46 chain of craft that you use to make it.
1:26:49 Like where did the idea come from? How
1:26:50 did you refine that idea? How did you
1:26:53 put the prompts together to get the
1:26:55 thing? How did you curate what came out
1:26:56 of it? How did you stitch them together
1:26:58 in the end?
1:27:01 That's all more important than any of
1:27:03 the tools you used along the way.
1:27:07 Oh, great. Speaking speaking of daily
1:27:09 practice mastermind today was on point.
1:27:12 Yeah, we left we left that Brandon
1:27:14 wondering if it was too much. It was it
1:27:16 was a lot. So So in the in the AI salon
1:27:20 mastermind practice lab, it's a 10-week
1:27:22 cycle. The first week is the kickoff
1:27:24 where you just get to understand what a
1:27:26 daily practice is. And today was about
1:27:29 human at the center um of of your
1:27:32 practice. And so it was about values and
1:27:34 ethics and intention. And so it's a lot
1:27:36 to pack into a a one-hour meeting. So it
1:27:39 it felt like a lot, but I hope people
1:27:41 left there um doing that. If if you are
1:27:44 looking to up your game with AI, you
1:27:47 should absolutely join the AI salon
1:27:49 mastermind practice lab. Uh it's every
1:27:51 Thursday at noon. Uh it's an amazing
1:27:54 group and it's an amazing process so
1:27:56 far. Um, all right. Let's watch AM
1:27:58 Murphy's video
1:28:01 in a new tab. Okay, cool. Thank you.
1:28:07 All right, I am live. [laughter] Oh,
1:28:10 boy.
1:28:14 I told I told Ann. [laughter]
1:28:18 Damn it. I told Ann she needs to make a
1:28:22 Sora cameo and and we need to be able to
1:28:25 troll each other and she trolled me
1:28:27 first. That is just not fair. [laughter]
1:28:30 It's just not fair. But here we go.
1:28:33 >> Step to the left and bring it back.
1:28:35 >> Pump those arms. Feel the future. Data's
1:28:37 in the groove, hearts in sync, rhythm
1:28:39 and AI
1:28:42 for the rest of us. Step to the left and
1:28:44 bring it back.
1:28:45 >> Pump those arms, feel the future.
1:28:47 >> In the groove, hearts in rhythm and a
1:28:51 god. [laughter]
1:28:57 Oh, good lord. Good people. All right,
1:29:01 an gauntlet has been thrown.
1:29:06 [sighs]
1:29:08 That's bad. That's That's That's solid
1:29:11 bad.
1:29:13 [laughter]
1:29:14 The body movements are so accurate. Time
1:29:17 to quit your quit your sleep sweet sleep
1:29:20 lab job. You have art down with the
1:29:23 world. Yeah, Joy. Joyy's
1:29:26 Joy can
1:29:28 Joyy's going to create it probably
1:29:30 already has created a a a
1:29:34 second career for herself if she wants
1:29:37 to do that. Um,
1:29:41 it [clears throat]
1:29:47 I've talked for three years now that
1:29:54 I deeply feel
1:29:57 that we're entering what I'm calling the
1:30:00 Great Renaissance.
1:30:02 There was the Renaissance, right? We
1:30:04 come out of the dark ages and have the
1:30:06 enlightenment and we we go into this
1:30:08 amazing time where art and science and
1:30:12 discovery just explodes.
1:30:17 Um
1:30:19 that renaissance is going to seem quaint
1:30:22 compared to what's coming.
1:30:25 And what Joy Perie just showed us there
1:30:28 is is an artifact on the way to that
1:30:31 journey.
1:30:33 Because
1:30:35 three years from now,
1:30:37 anyone like right now it requires Joyy's
1:30:42 commitment
1:30:44 and daily practice of learning and
1:30:47 educating and understanding how to get
1:30:50 the stories out of her head, how to
1:30:52 refine her vision.
1:30:54 Three years from now, anyone who's got
1:30:57 any kind of an idea is going to be able
1:30:58 to get it in the world. So the
1:31:00 opportunity
1:31:02 for establishing yourself in some new
1:31:04 entirely new career is now. When these
1:31:08 tools get so easy that anyone can use
1:31:10 them,
1:31:12 it's going to be harder to get into new
1:31:14 careers. But you'll still be able to if
1:31:17 you're good.
1:31:20 But
1:31:23 we're entering a world where the
1:31:25 friction
1:31:27 of taking an idea and manifesting it and
1:31:30 sharing that with your friends, your
1:31:31 family, your community or the entire
1:31:33 world, the friction essentially goes to
1:31:36 zero.
1:31:38 And I on on all fronts on science, on
1:31:41 art, on creativity, on business,
1:31:46 on community engagement, all of it. All
1:31:49 of it. It's going to be crazy.
1:31:55 I'm old, but here for it. Not fair
1:31:57 showing that after Joy's elegant pieces.
1:32:00 [laughter]
1:32:05 All right, everyone. So, tomorrow um
1:32:08 11:00 a.m. Mountain time is AI office
1:32:11 hours. If you don't know about that, you
1:32:13 need to come to that. You just do. Um I
1:32:16 know we're going into a big holiday week
1:32:18 next week. Kick it off with an office
1:32:20 hours. And then tomorrow's Friday night
1:32:22 date night. Um Gabby has a show opening.
1:32:25 Um I could I I'll show you tomorrow.
1:32:29 I'll show you the uh I'll take pictures
1:32:31 of it and show you the the opening
1:32:33 tomorrow. Um, so this so it'll probably
1:32:35 be nine o'clock tomorrow night. So we'll
1:32:37 do Friday Friday night date night late.
1:32:40 Um,
1:32:43 yeah. And if you want a link to office
1:32:45 hours tomorrow, if you go to the AI
1:32:47 salon and go to events, the AI salon
1:32:50 office hours and meet and greet is an
1:32:52 event there. You can RSVP to it and get
1:32:54 the link. Okay? It's been the same link
1:32:57 for years. Um, please send Gabby link to
1:33:00 her opening. Um,
1:33:04 If you go to, let's see, where is go to
1:33:08 the it's the Dart Gallery in Denver. D A
1:33:11 R T
1:33:13 um is where it is. So, tomorrow from
1:33:15 6:00 to 9:00. If you're in Denver, if
1:33:18 you're in the Colorado area, swing by
1:33:20 and say hello.
1:33:23 Just say I'm an irregular. [laughter]
1:33:27 Gabby will know.
1:33:31 Oh, man. Was
1:33:34 that an AI sweating with the oldies? Oh,
1:33:38 that's hilarious. Um, okay, cool. I'm
1:33:41 going to get out of here. Uh, I would,
1:33:44 if you have not played with Nano Banana,
1:33:46 if you have not played with Gemini 3,
1:33:49 um, the lesson I learned uh, two days
1:33:53 ago, or maybe it was even yesterday,
1:33:56 building a a website in an hour that I
1:33:59 had been avoiding for some reason.
1:34:03 Um, just because you feel clueless
1:34:06 doesn't mean you are.
1:34:08 Think of AI like you've got a you've got
1:34:10 a friend now that's ready to make you
1:34:12 look good. So, just go play. Even if you
1:34:14 feel clueless, just go play and explore
1:34:16 with these tools because there's there's
1:34:19 the way you think about the world
1:34:24 means that you're by definition going to
1:34:26 use Nano Banana differently than someone
1:34:28 else will. and you might unlock
1:34:30 something that they could never have
1:34:32 imagined. And I know that that sounds
1:34:34 like [ __ ] but it's not. Um, the
1:34:38 engineers don't know what they've built.
1:34:40 I promise you. They don't know what
1:34:42 they've built. They don't know how it's
1:34:43 going to be used.
1:34:45 Promise you. They They say it all the
1:34:47 time. So, anyway, um, that's where the
1:34:50 opportunity lies. Go figure some [ __ ]
1:34:52 out. All right. Have a good night and I
1:34:54 will talk to you tomorrow. Peace out.