1 00:00:00,020 --> 00:00:05,860 Josh: Just yesterday, OpenAI released ChatGPT Images 2.0, and the model blew my mind. 2 00:00:05,960 --> 00:00:09,320 Josh: I was up until two o'clock last morning playing around with it because of how powerful it is. 3 00:00:09,400 --> 00:00:13,580 Josh: As I was watching Sam announce this model, he was talking about how image gen 4 00:00:13,580 --> 00:00:17,220 Josh: wasn't really that important to him. He felt like they already had a good image generation model. 5 00:00:17,400 --> 00:00:21,520 Josh: When he was presented with the outputs of this one, he had his holy shit moment. 6 00:00:21,700 --> 00:00:23,260 Josh: It's actually really phenomenal. 7 00:00:23,600 --> 00:00:27,500 Josh: And through trying it ourselves, we have uncovered that it's actually true. 8 00:00:27,500 --> 00:00:32,000 Josh: I mean, we've really frequently used NanoBanana as the go-to default image generator. 9 00:00:32,670 --> 00:00:35,950 Josh: But now it's getting close to being indistinguishable from reality entirely. 10 00:00:36,210 --> 00:00:39,230 Josh: And we have a series of examples that we're going to show you that are probably 11 00:00:39,230 --> 00:00:41,050 Josh: useful for your actual applicable life. 12 00:00:41,230 --> 00:00:44,970 Josh: Things like interior design or generating comics or generating sales graphics. 13 00:00:45,370 --> 00:00:49,670 Josh: I don't think there's anyone who wouldn't find a beneficial use case of an image 14 00:00:49,670 --> 00:00:53,310 Josh: generator model that is as good as this one. So let's get into the actual announcement. 15 00:00:53,490 --> 00:00:55,710 Josh: Let's walk through the examples. It's pretty amazing stuff. 16 00:00:56,110 --> 00:01:02,330 Ejaaz: Around midday yesterday, OpenAI tweeted this very mysterious post and it goes, 17 00:01:02,670 --> 00:01:07,230 Ejaaz: This is not a screenshot, which is weird because it looks like a screenshot 18 00:01:07,230 --> 00:01:09,350 Ejaaz: of someone's Mac desktop, 19 00:01:09,690 --> 00:01:11,790 Ejaaz: except this is completely AI generated. 20 00:01:11,950 --> 00:01:17,770 Ejaaz: And this was the precursor to their official announcement, which is ChatGBT Images 2.0. 21 00:01:17,870 --> 00:01:21,730 Ejaaz: It's their new image model, and it absolutely blows every other image model 22 00:01:21,730 --> 00:01:24,390 Ejaaz: out of the water. And I don't mean that as an understatement. 23 00:01:24,590 --> 00:01:28,250 Ejaaz: It is number one across every single image benchmark. 24 00:01:28,250 --> 00:01:31,590 Ejaaz: It's beaten nano banana to and any 25 00:01:31,590 --> 00:01:34,230 Ejaaz: of the chinese image gen models just completely don't weigh 26 00:01:34,230 --> 00:01:37,490 Ejaaz: up so what are some of the new things here well the fidelity and 27 00:01:37,490 --> 00:01:41,030 Ejaaz: quality of these images are incredibly high you're seeing a demo video here 28 00:01:41,030 --> 00:01:45,650 Ejaaz: where we have a chameleon in various different positions the wording text is 29 00:01:45,650 --> 00:01:49,370 Ejaaz: typically such a hard thing for image models to nail especially when the ai 30 00:01:49,370 --> 00:01:52,890 Ejaaz: world it would like jumble up the letters or it wouldn't spell things correctly 31 00:01:52,890 --> 00:01:55,410 Ejaaz: now we have that completely and utterly resolved. 32 00:01:55,610 --> 00:01:57,830 Ejaaz: And so you can see some of these examples come to life here. 33 00:01:57,970 --> 00:02:01,430 Ejaaz: For example, look at the fidelity of this image of rice. 34 00:02:01,790 --> 00:02:05,770 Ejaaz: Typically, this would just look like a gobbled white mass. And now you can individually 35 00:02:05,770 --> 00:02:08,630 Ejaaz: see each grain, which is pretty nuts. 36 00:02:08,850 --> 00:02:13,030 Ejaaz: And then you have examples which are a little scarier, where this looks like 37 00:02:13,030 --> 00:02:16,750 Ejaaz: a real screenshot of a handwritten note in someone's stylistic way, 38 00:02:16,910 --> 00:02:19,590 Ejaaz: but it is very much completely AI generated. 39 00:02:19,730 --> 00:02:23,430 Ejaaz: So you can imagine this could be used for various different nefarious purposes, 40 00:02:23,430 --> 00:02:24,670 Ejaaz: which might get more malicious. 41 00:02:24,850 --> 00:02:27,510 Ejaaz: But there's a ton of different examples and we want to get straight into it, 42 00:02:27,810 --> 00:02:31,810 Ejaaz: starting with ones that we've generated ourselves. There's one around furniture, right? 43 00:02:32,220 --> 00:02:35,460 Josh: Yeah, but I actually want to start with the rice one, because what you mentioned 44 00:02:35,460 --> 00:02:38,500 Josh: with the rice is that it's precise enough to show the grains of rice, 45 00:02:38,580 --> 00:02:42,440 Josh: but it's also precise enough to write a single word on a grain of rice. 46 00:02:42,580 --> 00:02:46,380 Josh: And that fidelity is new. So what I did is I actually went to ChatGPT myself 47 00:02:46,380 --> 00:02:47,380 Josh: and tried to emulate this. 48 00:02:47,620 --> 00:02:52,400 Josh: And I asked it to create a piece of rice with the word GPT Image 2 generated 49 00:02:52,400 --> 00:02:54,000 Josh: on it. And this was the output that I got. 50 00:02:54,520 --> 00:02:58,300 Josh: Or actually, this was the first output that I got. And I spent maybe five minutes 51 00:02:58,300 --> 00:03:01,000 Josh: trying to find the grain of rice. I don't think it worked. 52 00:03:01,440 --> 00:03:05,120 Josh: So I asked it to draw a box around the grain of rice and it drew a box and then 53 00:03:05,120 --> 00:03:06,500 Josh: actually etched it in the middle. 54 00:03:06,660 --> 00:03:10,160 Josh: So there are some edge cases that don't quite work. I mean, that grain of rice 55 00:03:10,160 --> 00:03:11,020 Josh: was not in the original one. 56 00:03:11,360 --> 00:03:14,440 Josh: But for the ones that do work, it's pretty incredible. And you mentioned furniture. 57 00:03:14,800 --> 00:03:18,080 Josh: I am currently living in an apartment that can use a little bit of extra furnishing. 58 00:03:18,380 --> 00:03:21,780 Josh: This, unfortunately, is not what my apartment looks like. This is a much nicer 59 00:03:21,780 --> 00:03:24,680 Josh: variant of something that I would like to aspire to. 60 00:03:24,840 --> 00:03:29,120 Josh: So what I have prepared here is a reference image for ChatGPT along with the 61 00:03:29,120 --> 00:03:30,740 Josh: prompt of what I would like it to do. 62 00:03:31,000 --> 00:03:33,140 Josh: And that involves just doing things like... 63 00:03:33,830 --> 00:03:37,530 Josh: Adding lamps and adding different furniture basically swapping out the existing 64 00:03:37,530 --> 00:03:42,010 Josh: furniture that exists in this living room and moving it into a totally new vibe 65 00:03:42,010 --> 00:03:46,850 Josh: and style that i think i would more likely appreciate and resemble so while 66 00:03:46,850 --> 00:03:52,070 Josh: that's thinking i guess we can kind of get into some more of the interesting parts of this model. 67 00:03:52,070 --> 00:03:55,250 Ejaaz: Well i have a example that i actually have ready to 68 00:03:55,250 --> 00:03:59,750 Ejaaz: go here um i was kind of obsessed i don't tell anyone this i was obsessed with 69 00:03:59,750 --> 00:04:04,950 Ejaaz: manga as a kid and so i was like you know what would be cool if we could turn 70 00:04:04,950 --> 00:04:11,310 Ejaaz: our show you and i josh into a manga comic so i created this detail prompt i 71 00:04:11,310 --> 00:04:13,090 Ejaaz: gave this beautiful photo of us but look 72 00:04:13,090 --> 00:04:14,230 Josh: At those handsome guys. 73 00:04:14,230 --> 00:04:19,870 Ejaaz: Look at those handsome very handsome guys um and uh i basically asked chat gpt 74 00:04:19,870 --> 00:04:23,870 Ejaaz: to generate the prompt for me so i gave it a rough idea of what i wanted to 75 00:04:23,870 --> 00:04:28,910 Ejaaz: create the scene as it were and it created a very detailed prompt with stylistic 76 00:04:28,910 --> 00:04:32,310 Ejaaz: references details stuff that I wouldn't know because I'm not a storyboarder. 77 00:04:32,390 --> 00:04:35,810 Ejaaz: I'm not a manga creator, but funnily enough, I have an AI that can do it for me. 78 00:04:35,910 --> 00:04:39,710 Ejaaz: So I don't know if anyone is paying close attention to this storyline here, 79 00:04:39,810 --> 00:04:42,990 Ejaaz: but if you're not, that's great because I want to show you the end output. 80 00:04:43,110 --> 00:04:47,110 Ejaaz: So as you can see, very long prompt, and this is the finished result. 81 00:04:47,630 --> 00:04:52,250 Ejaaz: So what you are looking at here is Josh and I, 82 00:04:53,040 --> 00:04:55,860 Ejaaz: Let me explain this. Josh and I have been filming a podcast. 83 00:04:55,860 --> 00:04:57,940 Ejaaz: As you can see, we've got us set up over here. 84 00:04:58,160 --> 00:05:01,640 Ejaaz: But then we look out the window and there is a shadow. 85 00:05:01,920 --> 00:05:07,380 Ejaaz: And we notice that it is Sam Altman, Godzilla size, coming down upon us, 86 00:05:07,620 --> 00:05:09,800 Ejaaz: terrorizing New York City. 87 00:05:09,980 --> 00:05:12,700 Ejaaz: This is, I'd say the time estimate is roughly five years in the future, 88 00:05:12,900 --> 00:05:16,660 Ejaaz: maybe even three. Don't know how quickly AGI gets here. We grab our weapons. 89 00:05:16,860 --> 00:05:20,060 Ejaaz: It is clawed. This is not a sponsored video, by the way. I just came up with 90 00:05:20,060 --> 00:05:25,600 Ejaaz: this randomly. and it shoots out prompts that wrap around Sam Altman and eventually 91 00:05:25,600 --> 00:05:28,400 Ejaaz: bring down GPT-5 from taking over the world. 92 00:05:29,160 --> 00:05:32,260 Ejaaz: Now you know what's going on in my head, but if you just notice this, 93 00:05:32,340 --> 00:05:33,580 Ejaaz: like look at the fidelity of this. 94 00:05:33,660 --> 00:05:38,560 Ejaaz: This like took five seconds to create the prompt and then another two minutes 95 00:05:38,560 --> 00:05:39,640 Ejaaz: to create the actual image. 96 00:05:39,840 --> 00:05:42,780 Ejaaz: Look at the fidelity of this. Like the writing is all accurate. 97 00:05:43,260 --> 00:05:46,560 Ejaaz: The, like this would cost like thousands and thousands of dollars and weeks, 98 00:05:46,820 --> 00:05:51,720 Ejaaz: maybe months of time to actually create from scratch. and this did it in a bunch 99 00:05:51,720 --> 00:05:53,380 Ejaaz: of seconds for a couple of cents like it's pretty 100 00:05:53,380 --> 00:05:56,160 Josh: Impressive oh it's so good so if manga 101 00:05:56,160 --> 00:05:58,980 Josh: isn't your thing we have the furniture example it's 102 00:05:58,980 --> 00:06:03,800 Josh: ready to go so here i have the original that we're seeing on screen right now 103 00:06:03,800 --> 00:06:06,740 Josh: this was the original living room i fed it the prompt and here is the new one 104 00:06:06,740 --> 00:06:10,960 Josh: it totally maintained the integrity of the room while swapping out just a few 105 00:06:10,960 --> 00:06:14,720 Josh: key pieces of furniture to change the vibe and i think it's a testament to a 106 00:06:14,720 --> 00:06:16,400 Josh: practical use case that a lot of people might have, 107 00:06:16,970 --> 00:06:20,770 Josh: They want to design things. They want to make things look good and maintaining 108 00:06:20,770 --> 00:06:23,190 Josh: the personalized fidelity of whatever space it is. 109 00:06:23,270 --> 00:06:26,310 Josh: If you have a piece of clothing, I know this works for try-ons. 110 00:06:26,550 --> 00:06:28,950 Josh: It's really good at maintaining continuity throughout these images. 111 00:06:29,070 --> 00:06:30,690 Josh: So I thought that was a pretty interesting thing. 112 00:06:30,790 --> 00:06:33,770 Josh: If you have an apartment, if you have a closet worth full of clothes, 113 00:06:33,990 --> 00:06:36,290 Josh: you can just place those clothes out, take a picture of you, 114 00:06:36,390 --> 00:06:39,270 Josh: take a picture of their clothes, ask it to address you, ask it to redo your 115 00:06:39,270 --> 00:06:40,290 Josh: living room, whatever it may be. 116 00:06:40,470 --> 00:06:44,510 Josh: Super powerful and works fairly quick. I mean, this output took maybe a minute to generate. 117 00:06:44,510 --> 00:06:50,390 Josh: And for those not sure this is actually available to all users of chat gpt i 118 00:06:50,390 --> 00:06:54,670 Josh: believe very limited instances for the free users but if you have the plus plan 119 00:06:54,670 --> 00:06:58,770 Josh: for 20 a month you can just go off and start creating images and it will look 120 00:06:58,770 --> 00:07:00,490 Josh: just as good as this one yeah. 121 00:07:00,490 --> 00:07:04,690 Ejaaz: I mean if you're a professional that has been toying around with using ai but 122 00:07:04,690 --> 00:07:08,470 Ejaaz: it's just never been good enough it's always got some form of error whether 123 00:07:08,470 --> 00:07:14,210 Ejaaz: minor or big now we have a tool that actually works for you so if you're a designer, 124 00:07:14,210 --> 00:07:15,090 Ejaaz: if you're a floor planner, 125 00:07:15,210 --> 00:07:18,790 Ejaaz: there's a bunch of other examples I'll show here. This becomes a practical tool. 126 00:07:19,010 --> 00:07:23,390 Ejaaz: Like GPT Images 1 was very much a novelty and a toy. It was fun to see everyone 127 00:07:23,390 --> 00:07:25,630 Ejaaz: in Studio Ghibli versions of ourselves. 128 00:07:25,870 --> 00:07:28,170 Ejaaz: But now you can use this to create certain things. 129 00:07:28,430 --> 00:07:34,890 Ejaaz: Now, not all use cases are good. If you're like me, I use social media to disseminate 130 00:07:34,890 --> 00:07:38,050 Ejaaz: a lot of the breaking news that happens in the world of technology, 131 00:07:38,230 --> 00:07:39,370 Ejaaz: AI, or whatever it might be. 132 00:07:39,370 --> 00:07:42,190 Ejaaz: But you now have reached a point 133 00:07:42,190 --> 00:07:45,790 Ejaaz: where we can't necessarily believe everything we see and images 134 00:07:45,790 --> 00:07:48,670 Ejaaz: too from chat gpt doesn't make that any 135 00:07:48,670 --> 00:07:52,310 Ejaaz: easier what you're seeing on the screen right now is not 136 00:07:52,310 --> 00:07:57,330 Ejaaz: an official take or update on the bloomberg terminal that is also not my desktop 137 00:07:57,330 --> 00:08:02,630 Ejaaz: monitor this is completely ai generated and you can probably tell parts of this 138 00:08:02,630 --> 00:08:06,470 Ejaaz: like kind of gives it away it's a little too zoomed in unless of course you 139 00:08:06,470 --> 00:08:08,950 Ejaaz: can like change the default settings in your the Bloomberg terminal, 140 00:08:09,150 --> 00:08:11,110 Ejaaz: but some of these things are really good. 141 00:08:11,210 --> 00:08:13,610 Ejaaz: This is exactly where this is on the Bloomberg terminal. 142 00:08:13,810 --> 00:08:16,590 Ejaaz: The percentage mark isn't that large on the actual thing. 143 00:08:17,260 --> 00:08:20,920 Ejaaz: It's got all the sections pretty much nailed. So you know that the model looked 144 00:08:20,920 --> 00:08:25,180 Ejaaz: up official Bloomberg terminal layouts and like recreated it, 145 00:08:25,300 --> 00:08:28,340 Ejaaz: but it added a completely fake kind of like bit of news. 146 00:08:28,420 --> 00:08:31,820 Ejaaz: So you could change that bit of news to represent real news, 147 00:08:31,860 --> 00:08:33,020 Ejaaz: but it would still be fake. 148 00:08:33,220 --> 00:08:37,380 Ejaaz: So there's a lot of like avenues here for misinformation or disinformation. 149 00:08:37,420 --> 00:08:40,360 Ejaaz: So like not entirely accurate, but somewhat accurate. 150 00:08:40,580 --> 00:08:43,560 Ejaaz: You can imagine the kind of social media frenzies that this would create if 151 00:08:43,560 --> 00:08:45,780 Ejaaz: people were to believe and buy into these things. 152 00:08:45,860 --> 00:08:48,440 Ejaaz: Like imagine if you read an announcement that wasn't actually real, 153 00:08:48,660 --> 00:08:51,280 Ejaaz: bought a stock, and then it realized that it was fake, and then it crashed. 154 00:08:51,460 --> 00:08:52,480 Ejaaz: You could end up losing money. 155 00:08:52,720 --> 00:08:55,500 Ejaaz: You could fake data. There's a lot of avenues of this go down. 156 00:08:55,920 --> 00:08:59,980 Josh: Yeah, there's two points on this. One is that we're at the point now where even 157 00:08:59,980 --> 00:09:03,560 Josh: if you pixel peep, it is almost indistinguishable from real life. 158 00:09:03,680 --> 00:09:07,320 Josh: You can't really tell what is AI generated and what is not. And as that kind 159 00:09:07,320 --> 00:09:11,460 Josh: of gap converges, I imagine it will create a lot of chaos where there's just 160 00:09:11,460 --> 00:09:14,100 Josh: no way to tell what's real when these images are so good. 161 00:09:14,280 --> 00:09:18,000 Josh: The second thing that I'll mention is from this model in particular, 162 00:09:18,220 --> 00:09:23,300 Josh: anytime it's asked to generate a visual asset of a piece of software, 163 00:09:23,300 --> 00:09:27,900 Josh: for some reason, it's exceptionally good at understanding the nuances of every 164 00:09:27,900 --> 00:09:29,420 Josh: frame of every piece of software. 165 00:09:29,800 --> 00:09:33,580 Josh: Last night, I had it do DaVinci Resolve, which is what I edit a lot of videos in. 166 00:09:33,720 --> 00:09:39,120 Josh: I had it emulate Photoshop, and it got every single slider down to the correct pixel, 167 00:09:39,280 --> 00:09:42,020 Josh: which leads me to believe that it appears as if 168 00:09:42,020 --> 00:09:45,940 Josh: there was some training customization around the software 169 00:09:45,940 --> 00:09:48,600 Josh: project in particular and you have to ask the second order 170 00:09:48,600 --> 00:09:54,260 Josh: question why why is it so good at all the software and I guess the answer for 171 00:09:54,260 --> 00:09:57,660 Josh: me is well it's probably because they want their agents to understand how to 172 00:09:57,660 --> 00:10:01,140 Josh: navigate it and then eventually emulate it and then eventually replace it and 173 00:10:01,140 --> 00:10:05,340 Josh: this nuanced understanding of how everything works is training for the image 174 00:10:05,340 --> 00:10:07,160 Josh: generation model but also training for. 175 00:10:07,860 --> 00:10:10,960 Josh: Just i mean the future of what these agents are going to look like so 176 00:10:10,960 --> 00:10:13,840 Josh: there might be some hidden stuff going on behind this image generation 177 00:10:13,840 --> 00:10:17,200 Josh: model as well so back to the demos in addition to these capabilities 178 00:10:17,200 --> 00:10:20,280 Josh: we have another one teed up right here which is to create a premium 179 00:10:20,280 --> 00:10:23,340 Josh: infographic poster another strong suit of this model is 180 00:10:23,340 --> 00:10:28,820 Josh: text and how well it's able to render text that looks lifelike looks accurate 181 00:10:28,820 --> 00:10:33,580 Josh: and is able to kind of create a like a storyboard if you will a poster it can 182 00:10:33,580 --> 00:10:36,880 Josh: create multiple outputs what i've asked it to do here is create an editorial 183 00:10:36,880 --> 00:10:39,900 Josh: infographic and this is the first time I'm actually seeing the output of this 184 00:10:39,900 --> 00:10:41,660 Josh: and it seems pretty cool. 185 00:10:41,780 --> 00:10:46,820 Josh: So this is for Limitless as you are familiar with and it kind of walks through our week in review. 186 00:10:47,040 --> 00:10:51,040 Josh: So the things that Limitless mentions, this is the poster that serves as like 187 00:10:51,040 --> 00:10:52,560 Josh: the weekly roundup, the weekly review. 188 00:10:52,960 --> 00:10:55,080 Josh: It is pretty good. 189 00:10:55,340 --> 00:10:58,420 Ejaaz: Is it accurate? Yeah, I'm curious. 190 00:10:58,720 --> 00:11:01,940 Josh: You check the accuracy, I'll check the QR code, see if that works because word 191 00:11:01,940 --> 00:11:04,540 Josh: on the street is that QR codes work pretty well. wow. 192 00:11:04,540 --> 00:11:11,400 Ejaaz: I might need to replace the entire roundup newsletter josh with something like 193 00:11:11,400 --> 00:11:14,040 Ejaaz: this just a quick a quick glance quick take away 194 00:11:14,040 --> 00:11:16,880 Josh: You can imagine how this can kind of carry out 195 00:11:16,880 --> 00:11:19,960 Josh: to other applications right it's like if we want to juice the 196 00:11:19,960 --> 00:11:22,820 Josh: newsletter up a little bit we could just create a graphic with one prompt 197 00:11:22,820 --> 00:11:25,500 Josh: by feeding it the context of everything we spoke about to give you 198 00:11:25,500 --> 00:11:28,660 Josh: this detailed infographic this also applies to educators 199 00:11:28,660 --> 00:11:31,820 Josh: and people who are teaching things it's really 200 00:11:31,820 --> 00:11:34,980 Josh: easy it to make graphics on particular lessons or 201 00:11:34,980 --> 00:11:39,320 Josh: mathematical equations or diagrams or anything you want visually represented 202 00:11:39,320 --> 00:11:43,860 Josh: it's exceptionally good at that so i thought this demo was kind of fun it creates 203 00:11:43,860 --> 00:11:47,460 Josh: the limitless we can review as a poster that's printable the qr code does not 204 00:11:47,460 --> 00:11:51,000 Josh: work but i've asked it to make the qr code scannable so while it finishes that 205 00:11:51,000 --> 00:11:53,540 Josh: up and we test that maybe we could go on to another example yeah. 206 00:11:53,540 --> 00:11:57,060 Ejaaz: I was just going to say before we move on um the educational traditional 207 00:11:57,670 --> 00:12:01,950 Ejaaz: point is a very precedent one, mainly because if you're like me, 208 00:12:02,150 --> 00:12:07,550 Ejaaz: you could read as much text as you want, but sometimes a visual that summarizes 209 00:12:07,550 --> 00:12:08,730 Ejaaz: everything really helps. 210 00:12:08,930 --> 00:12:13,430 Ejaaz: You can now plug an entire book's worth of text into a single prop. 211 00:12:13,550 --> 00:12:17,550 Ejaaz: A lot of these frontier models now have a million contexts, which is a couple 212 00:12:17,550 --> 00:12:19,210 Ejaaz: of novels or many, many novels. 213 00:12:19,410 --> 00:12:21,890 Ejaaz: And so if you can imagine, if you're trying to learn about something and you 214 00:12:21,890 --> 00:12:25,670 Ejaaz: want the key points, you can not only ask the AI to summarize things and give 215 00:12:25,670 --> 00:12:28,950 Ejaaz: you a bullet pointed list, you can get them to transform it into an illustrative 216 00:12:28,950 --> 00:12:34,050 Ejaaz: poster that just you can look at in a glance before you go to bed and learn something brand new. 217 00:12:34,170 --> 00:12:37,390 Ejaaz: So I can imagine this being used in science as well. 218 00:12:37,790 --> 00:12:41,070 Ejaaz: Back when I had a biology degree or back when I was doing my degree, 219 00:12:41,390 --> 00:12:46,290 Ejaaz: I remember we used to have these like research poster conferences and they used 220 00:12:46,290 --> 00:12:48,650 Ejaaz: to be like, I don't know, A1 size. 221 00:12:48,770 --> 00:12:51,710 Ejaaz: It was absolutely massive. And you would have so much condensed information 222 00:12:51,710 --> 00:12:56,550 Ejaaz: there and it took me weeks to make and the fact that i now have a tool here 223 00:12:56,550 --> 00:12:58,630 Ejaaz: where you can just probably plug in a bunch of papers 224 00:12:59,020 --> 00:13:01,620 Ejaaz: get out extract the right information and then put it out in a 225 00:13:01,620 --> 00:13:04,840 Ejaaz: very visual way just blows my mind like we are condensing a 226 00:13:04,840 --> 00:13:08,200 Ejaaz: lot of frontier research and education tools like 227 00:13:08,200 --> 00:13:13,700 Ejaaz: with this one simple update it's very very cool but to move on to uh one more 228 00:13:13,700 --> 00:13:18,360 Ejaaz: example that we generated um one thing that's cool about images too is you can 229 00:13:18,360 --> 00:13:23,120 Ejaaz: play around with one image and make it into several different aspect ratios 230 00:13:23,120 --> 00:13:27,440 Ejaaz: so what we have here is an individual i don't know who this individual is, 231 00:13:27,540 --> 00:13:30,380 Ejaaz: but it has generated it looking out onto the greatest city in the world, 232 00:13:30,460 --> 00:13:31,920 Ejaaz: in my opinion, New York City. 233 00:13:32,120 --> 00:13:36,100 Ejaaz: And it's in like a nice little sunrise or sunset. I can't tell which one is which. 234 00:13:36,380 --> 00:13:39,540 Ejaaz: But as you notice, it gives us different aspect ratios of the guy. 235 00:13:39,640 --> 00:13:41,680 Ejaaz: Like over here, we see him on the left. 236 00:13:41,840 --> 00:13:44,880 Ejaaz: Over here, we see him from a distance back. Over here, we see a panoramic view 237 00:13:44,880 --> 00:13:48,500 Ejaaz: where we can see him looking out onto what is this? This is Brooklyn Bridge. 238 00:13:48,900 --> 00:13:53,220 Ejaaz: So the details of it, you know, you can see some of it is like kind of like 239 00:13:53,220 --> 00:13:54,460 Ejaaz: blurred aspect ratios as well. 240 00:13:54,620 --> 00:13:57,500 Ejaaz: It's just very impressive and you could start creating like 241 00:13:57,500 --> 00:14:01,800 Ejaaz: storyboard sequences from this or just kind of like pitching visuals to whatever 242 00:14:01,800 --> 00:14:06,620 Ejaaz: you whatever kind of like idea or concept you want to make you could use this 243 00:14:06,620 --> 00:14:10,020 Ejaaz: in the product realm if you're like trying to figure out whether a model looks 244 00:14:10,020 --> 00:14:13,380 Ejaaz: good advertising your product in like let's say the product was a coat in this 245 00:14:13,380 --> 00:14:17,380 Ejaaz: particular way or it could just be something advertising completely different it's it's very cool so 246 00:14:17,380 --> 00:14:20,180 Josh: How does this model perform so well i think 247 00:14:20,180 --> 00:14:23,060 Josh: is the question and one of the novel breakthroughs that this image 248 00:14:23,060 --> 00:14:27,820 Josh: model has that others don't is the detailed reasoning capabilities this is an 249 00:14:27,820 --> 00:14:31,720 Josh: image generated model that will think before acting and will reason through 250 00:14:31,720 --> 00:14:35,840 Josh: the steps required to get the best image output so generally it's just pure 251 00:14:35,840 --> 00:14:39,400 Josh: inference you give it a context you give it input and it just spits something 252 00:14:39,400 --> 00:14:41,540 Josh: out this one actually reasons through the, 253 00:14:42,040 --> 00:14:44,800 Josh: i guess the reasoning of why it's doing these things and that's part 254 00:14:44,800 --> 00:14:47,560 Josh: of the reason why even though you're not giving it necessarily the 255 00:14:47,560 --> 00:14:50,580 Josh: best prompt it's giving you a really powerful output and i 256 00:14:50,580 --> 00:14:53,680 Josh: have another fun example here of um just like more comic books 257 00:14:53,680 --> 00:14:56,380 Josh: that you can make this was a single prompt and that generated like 258 00:14:56,380 --> 00:14:59,680 Josh: an entire comic book with a really accurate character that's carried throughout 259 00:14:59,680 --> 00:15:02,960 Josh: another fun feature is the character continuity where 260 00:15:02,960 --> 00:15:05,800 Josh: you can generate a character it will be prevailing throughout all the images 261 00:15:05,800 --> 00:15:08,700 Josh: and then also one last example that we have here is of 262 00:15:08,700 --> 00:15:11,320 Josh: anyone who's involved in social media or just creating any sort of 263 00:15:11,320 --> 00:15:14,280 Josh: marketing material i asked it to create an ad package for 264 00:15:14,280 --> 00:15:17,600 Josh: a monster shop in williamsburg called sagebird and 265 00:15:17,600 --> 00:15:23,420 Josh: sagebird now has a full kit of various aspect ratios to be posted on any platform 266 00:15:23,420 --> 00:15:27,920 Josh: that looks photo accurate if you'll notice there's even a street sign that says 267 00:15:27,920 --> 00:15:31,580 Josh: bedford avenue which is a street in williamsburg which is very funny so i think 268 00:15:31,580 --> 00:15:34,840 Josh: the the fidelity the quality the capabilities of this model. 269 00:15:35,440 --> 00:15:39,380 Josh: Are really endless and again the constraint is your imagination with how far 270 00:15:39,380 --> 00:15:43,600 Josh: you can push this thing because it's just it's so powerful i had so much fun 271 00:15:43,600 --> 00:15:46,540 Josh: using this i must have generated at least 100 images so far just in the last 272 00:15:46,540 --> 00:15:50,440 Josh: 24 hours and it is like it's so fun i recommend everyone go and try it and figure 273 00:15:50,440 --> 00:15:52,580 Josh: out what use cases are best for you so. 274 00:15:52,580 --> 00:15:57,000 Ejaaz: A question that came to mind immediately is okay it's good but how does it compare 275 00:15:57,000 --> 00:16:01,920 Ejaaz: to some of its competitors primarily nano banana 2 from Google, 276 00:16:02,120 --> 00:16:04,260 Ejaaz: who has previously held the number one spot here. 277 00:16:04,620 --> 00:16:07,840 Ejaaz: Now, if you look at this image over here, it's not just number one, 278 00:16:07,940 --> 00:16:09,840 Ejaaz: it's number one by a far mile. 279 00:16:09,960 --> 00:16:13,500 Ejaaz: I think it has like 150 point increase on Image Arena. 280 00:16:13,900 --> 00:16:17,540 Ejaaz: If you don't know what this is, this is like the number one benchmark to test these image models. 281 00:16:17,940 --> 00:16:22,980 Ejaaz: GPT Images 2 isn't just number one overall, it is number one across every single 282 00:16:22,980 --> 00:16:27,640 Ejaaz: category that is measured within this benchmark. By a long shot. 283 00:16:28,180 --> 00:16:31,420 Ejaaz: So it has a very distinctive lead. And if you're looking at this and you're 284 00:16:31,420 --> 00:16:34,920 Ejaaz: saying, okay, well, whatever, people can like orient benchmarks around this. 285 00:16:35,020 --> 00:16:36,160 Ejaaz: So like, we don't know if it's real. 286 00:16:36,340 --> 00:16:41,960 Ejaaz: I have a direct comparison for you. So the same prompt fed into GPT Image 2 versus Dano Bonata Pro. 287 00:16:42,140 --> 00:16:46,600 Ejaaz: And you can see that there is quite a lot of differences. You can see GPT Images 288 00:16:46,600 --> 00:16:51,880 Ejaaz: 2 over here on the left, the lighting is much brighter, the fidelity is arguably a lot better. 289 00:16:52,640 --> 00:16:56,340 Ejaaz: And as you can see, like, you know, there's more expression on her face, she's smiling. 290 00:16:56,780 --> 00:16:59,320 Ejaaz: And there's a lot more things in the background. Like if you look at the plants 291 00:16:59,320 --> 00:17:03,580 Ejaaz: in the back, it's way more hyper-realistic and harder to create for an image model. 292 00:17:03,680 --> 00:17:07,240 Ejaaz: Now, if you look on the right, Nano Banana 2 is very good, but there's less 293 00:17:07,240 --> 00:17:09,180 Ejaaz: complicated things going on behind them. 294 00:17:09,260 --> 00:17:12,180 Ejaaz: The lighting is a little bit off. And you can kind of tell that 295 00:17:12,490 --> 00:17:16,110 Ejaaz: I don't know, like maybe still on both sides, you can tell that they are kind 296 00:17:16,110 --> 00:17:17,190 Ejaaz: of slightly AI generated. 297 00:17:17,350 --> 00:17:19,970 Ejaaz: I would actually argue that images too, now that I'm looking at it for longer, 298 00:17:20,570 --> 00:17:23,170 Ejaaz: looks like the glisten just seems too glistening-y. 299 00:17:24,130 --> 00:17:26,730 Ejaaz: But Nanobanana 2 can get away with it because the lighting is a little less. 300 00:17:26,830 --> 00:17:29,290 Ejaaz: But the point is these models are getting way, way better. 301 00:17:29,550 --> 00:17:32,650 Ejaaz: And the examples keep coming, but it's not just visual things. 302 00:17:32,830 --> 00:17:36,190 Ejaaz: Like social media influencers don't have to be worried here. 303 00:17:36,630 --> 00:17:42,210 Ejaaz: You can start using this for very practical purposes. Now, there was this awesome example over here. 304 00:17:42,490 --> 00:17:46,230 Ejaaz: Where a guy took an image of a book, right? 305 00:17:46,430 --> 00:17:51,330 Ejaaz: And he said, could you generate me a barcode for this book? 306 00:17:51,590 --> 00:17:54,990 Ejaaz: And he generated the barcode. And when you scan the barcode, 307 00:17:55,170 --> 00:17:57,730 Ejaaz: it takes you, it's basically an embedded link. 308 00:17:57,870 --> 00:18:01,990 Ejaaz: It takes you to a page where you can then purchase or buy the book. 309 00:18:02,170 --> 00:18:06,250 Ejaaz: Now, this is very impressive for if you are like trying to sell a particular 310 00:18:06,250 --> 00:18:09,810 Ejaaz: product, especially if it's physical, you now don't need to go through the complicated 311 00:18:09,810 --> 00:18:11,630 Ejaaz: process of generating barcodes, getting it printed. 312 00:18:11,630 --> 00:18:15,810 Ejaaz: You could feasibly create your own design book cover, print it out, 313 00:18:15,950 --> 00:18:18,270 Ejaaz: and then wrap it around your actual product and it actually works. 314 00:18:18,290 --> 00:18:21,130 Ejaaz: It works with your internal system. So I just thought this was pretty cool. 315 00:18:21,690 --> 00:18:26,410 Josh: Yeah, it's amazing. The clarity. And again, I think this is a testament to the 316 00:18:26,410 --> 00:18:28,810 Josh: reasoning where it can actually recent its way through and generate an accurate 317 00:18:28,810 --> 00:18:30,810 Josh: barcode in a world where it previously couldn't. 318 00:18:30,970 --> 00:18:34,550 Josh: So now not only can it make infographics, but it could link these dynamic elements 319 00:18:34,550 --> 00:18:38,310 Josh: to real world artifacts, to a custom domain, to your book. 320 00:18:38,430 --> 00:18:41,690 Josh: They're actually usable without needing to take it into Photoshop and take it 321 00:18:41,690 --> 00:18:44,530 Josh: that final mile. And that's like a really cool unlock. 322 00:18:44,710 --> 00:18:50,090 Ejaaz: We have another example here as well of the front page of the New York Times, 323 00:18:50,270 --> 00:18:53,910 Ejaaz: which of course is entirely fabricated um or 324 00:18:53,910 --> 00:18:56,750 Ejaaz: at least partially so like this isn't a real article this 325 00:18:56,750 --> 00:18:59,990 Ejaaz: isn't a real image of a paper but all the information on 326 00:18:59,990 --> 00:19:04,110 Ejaaz: it so if you actually dig in here and read it um all the information about open 327 00:19:04,110 --> 00:19:08,310 Ejaaz: air unveiling gpt image 2 is accurate they pulled it from the blog post they 328 00:19:08,310 --> 00:19:11,110 Ejaaz: didn't you didn't have to provide the blog post they independently did it it 329 00:19:11,110 --> 00:19:14,030 Ejaaz: reasoned through it pulled out the most important points and then wrote it in 330 00:19:14,030 --> 00:19:16,870 Ejaaz: a stylistic manner of a New York Times writer. 331 00:19:17,110 --> 00:19:20,970 Ejaaz: So you can start imagining what this could do for press and media. 332 00:19:21,150 --> 00:19:24,510 Ejaaz: If you are a reporter, you might be thinking, huh, so you're telling me I could 333 00:19:24,510 --> 00:19:27,330 Ejaaz: just feed this the bullet points that I want it to make, and it could write 334 00:19:27,330 --> 00:19:32,490 Ejaaz: it in my voice, in my DNA that I stylistically write an article for? 335 00:19:32,670 --> 00:19:35,470 Ejaaz: That's amazing. You could also ask it to generate the image for you. 336 00:19:35,570 --> 00:19:38,270 Ejaaz: So there's this metro approach where you're talking about the product, 337 00:19:38,450 --> 00:19:41,590 Ejaaz: but then you use the product to generate an example image that you then put in. 338 00:19:41,690 --> 00:19:45,290 Ejaaz: This is, of course, also generated by images too. 339 00:19:45,590 --> 00:19:48,190 Ejaaz: So there's a lot of applications here. Again, I mentioned earlier, 340 00:19:48,390 --> 00:19:52,190 Ejaaz: disinformation is a very real thing. So you can imagine people sharing fake 341 00:19:52,190 --> 00:19:55,270 Ejaaz: news articles about things that aren't real, that might sway markets or inform 342 00:19:55,270 --> 00:19:58,170 Ejaaz: people in the incorrect way, but cool nonetheless. 343 00:19:59,250 --> 00:20:02,510 Josh: Yeah. And then there's more examples for people who are involved in architecture 344 00:20:02,510 --> 00:20:03,890 Josh: at all, if you're doing floor plans. 345 00:20:04,090 --> 00:20:08,690 Josh: I mean, this one was cool where you fed it an image of a house and then it generated a floor plan. 346 00:20:08,790 --> 00:20:12,510 Josh: But the next example I think was even cooler because this was a digital rendering 347 00:20:12,510 --> 00:20:17,310 Josh: of a large building that had all of the specs listed next to it. 348 00:20:17,470 --> 00:20:20,470 Josh: And using that spec sheet and using that 3D rendering, it 349 00:20:20,470 --> 00:20:23,710 Josh: created a fully rendered floor plan 350 00:20:23,710 --> 00:20:27,110 Josh: that you can actually use and send to an architecture to 351 00:20:27,110 --> 00:20:30,070 Josh: send to an architect to to actually make blueprints and build the 352 00:20:30,070 --> 00:20:33,030 Josh: building i'm not sure if this is up to code i'm not an architect but 353 00:20:33,030 --> 00:20:36,730 Josh: i imagine you can probably iterate your way through this with a proper architect 354 00:20:36,730 --> 00:20:40,630 Josh: to get it to be compliant to get it up to spec if it's not already and train 355 00:20:40,630 --> 00:20:44,250 Josh: it to do that so this there's this unbelievable unlock that happens for pretty 356 00:20:44,250 --> 00:20:47,590 Josh: much any profession that's generating any sort of image all you need to do is 357 00:20:47,590 --> 00:20:51,550 Josh: put a stamp on the bottom it looks like it it already stamped it with some fake stamp approval. 358 00:20:52,410 --> 00:20:56,450 Josh: But I'm sure if you do this type of work, you can kind of, you can put your 359 00:20:56,450 --> 00:20:58,030 Josh: own spin on and throw your own stamp on there. 360 00:20:58,350 --> 00:21:03,230 Ejaaz: If any of you are architects listening to this, I encourage you to try this 361 00:21:03,230 --> 00:21:07,890 Ejaaz: out because I'm actually curious whether this is accurate or if not, like how accurate is it? 362 00:21:08,030 --> 00:21:12,490 Ejaaz: Because obviously like architects in training, like train for seven years at 363 00:21:12,490 --> 00:21:13,630 Ejaaz: school, which is just insane. 364 00:21:13,850 --> 00:21:18,510 Ejaaz: They have to understand the physics behind the buildings that they're designing. 365 00:21:18,650 --> 00:21:20,410 Ejaaz: And I'm wondering, is this physically accurate? 366 00:21:20,570 --> 00:21:25,190 Ejaaz: Are the estimates, like, do they make sense? Or is this completely made up and 367 00:21:25,190 --> 00:21:26,250 Ejaaz: we still have a long way to go? 368 00:21:26,390 --> 00:21:30,990 Ejaaz: It looks legit to me, but then I'm not an architect. So if you're listening to this, let us know. 369 00:21:31,630 --> 00:21:35,430 Ejaaz: There's another cool thing here where, again, I mentioned earlier, 370 00:21:35,570 --> 00:21:39,110 Ejaaz: if you are a visual learner, sometimes you just, there's too much information. 371 00:21:39,470 --> 00:21:44,550 Ejaaz: You can create these posters bracketed by a particular subject and it kind of splits it up. 372 00:21:44,650 --> 00:21:46,890 Ejaaz: So like with here, we have like all the things going on in AI, 373 00:21:47,010 --> 00:21:49,090 Ejaaz: you got AI models and agents, robotics, semiconductors. 374 00:21:49,230 --> 00:21:53,670 Ejaaz: And you just have images which explain the start to end process of creating 375 00:21:53,670 --> 00:21:56,270 Ejaaz: these different things and what they actually do with a few words underneath 376 00:21:56,270 --> 00:21:57,610 Ejaaz: it, which I thought was cool. 377 00:21:57,890 --> 00:22:00,910 Ejaaz: And then there was this final example over here from Matt Schumer, 378 00:22:01,600 --> 00:22:06,880 Ejaaz: where I can relate to this because I formerly worked at a big four consultancy 379 00:22:06,880 --> 00:22:11,060 Ejaaz: and we had to create slide decks and it would take so long because you had to 380 00:22:11,060 --> 00:22:13,480 Ejaaz: move things in a specific way or reformat the text. 381 00:22:14,080 --> 00:22:19,120 Ejaaz: And Matt Schumer one-shotted an entire slide deck by just providing it a bunch 382 00:22:19,120 --> 00:22:22,940 Ejaaz: of information and it created it in the style of Spotify by the looks of it. 383 00:22:23,860 --> 00:22:27,560 Ejaaz: So very cool, loads of different applications and I can't wait for more people 384 00:22:27,560 --> 00:22:29,640 Ejaaz: to actually use this for professional purposes. 385 00:22:30,340 --> 00:22:35,000 Josh: Yeah, the model's awesome. And I guess the ask is to share whatever you're using it for. 386 00:22:35,220 --> 00:22:38,300 Josh: Because again, like those prompts, those examples are the only limiting factors 387 00:22:38,300 --> 00:22:39,260 Josh: to really what this can do. 388 00:22:39,620 --> 00:22:44,180 Josh: Because it has the reasoning, because it's so capable, it has the like pixel perfect fidelity. 389 00:22:44,460 --> 00:22:47,920 Josh: It's really just a matter of massaging it with prompts to get the output you 390 00:22:47,920 --> 00:22:50,660 Josh: want, not really a limitation of the model anymore. 391 00:22:50,920 --> 00:22:54,800 Josh: And like to Sam's point early in the episode, it seemed like it was great before. 392 00:22:55,020 --> 00:22:56,240 Josh: Now this is just unbelievable. 393 00:22:56,480 --> 00:23:00,160 Josh: I can't imagine going back to Nano Banana Pro knowing that this exists. 394 00:23:00,160 --> 00:23:02,900 Josh: And it's just a testament again to how fast we're 395 00:23:02,900 --> 00:23:05,740 Josh: going and like what the downstream implications of this may be in 396 00:23:05,740 --> 00:23:09,160 Josh: the future when you can generate infinite images for 397 00:23:09,160 --> 00:23:12,140 Josh: cheap that are pixel perfect and indistinguishable from reality 398 00:23:12,140 --> 00:23:15,180 Josh: what type of downstream effects does that have on 399 00:23:15,180 --> 00:23:18,020 Josh: every visual artifact that we interact 400 00:23:18,020 --> 00:23:21,780 Josh: with on a day-to-day basis i mean there's no way you could be sure and this 401 00:23:21,780 --> 00:23:26,220 Josh: this has a lot of implications that i'm not sure we're fully aware of now but 402 00:23:26,220 --> 00:23:30,800 Josh: will surely become known well as we kind of navigate through this it creates 403 00:23:30,800 --> 00:23:34,380 Josh: a weird dynamic that seems a little uncomfortable like i and now i have to navigate 404 00:23:34,380 --> 00:23:37,880 Josh: the internet with such a strong filter to just try to parse through what's real and what's not. 405 00:23:38,590 --> 00:23:44,610 Ejaaz: I'm curious whether this tool can be used to generate visuals that humans hadn't 406 00:23:44,610 --> 00:23:46,390 Ejaaz: thought of before necessarily. 407 00:23:46,670 --> 00:23:51,090 Ejaaz: Like as the AI becomes smarter and is trained on our prompts and largely our 408 00:23:51,090 --> 00:23:56,570 Ejaaz: flaws, like, you know, you can ask an AI to generate a detailed prompt to then 409 00:23:56,570 --> 00:23:59,010 Ejaaz: prompt it itself because we don't know how to prompt it itself. 410 00:23:59,190 --> 00:24:03,150 Ejaaz: Like it can do the same with images where it's like, I get that Ejaz probably 411 00:24:03,150 --> 00:24:06,990 Ejaaz: missed this point. And so maybe if I create this visual in this particular way, 412 00:24:07,110 --> 00:24:10,870 Ejaaz: it's one that he hadn't thought of, but now it like breaks new ground for it. 413 00:24:10,950 --> 00:24:15,750 Ejaaz: So I wouldn't put it past this model to like the model that we have to today 414 00:24:15,750 --> 00:24:21,330 Ejaaz: to generate something, a visual artifact that will soon be kind of like groundbreaking 415 00:24:21,330 --> 00:24:23,830 Ejaaz: for humans to use. Like maybe it's not a poster. 416 00:24:24,050 --> 00:24:26,510 Ejaaz: Maybe it's not a slide deck. Maybe it's something completely new that we haven't 417 00:24:26,510 --> 00:24:28,190 Ejaaz: seen before. Pretty exciting stuff. 418 00:24:28,670 --> 00:24:33,210 Josh: Yeah. So that's ChatGPT Images 2.0, the newest and hottest image gen model in the world. 419 00:24:33,210 --> 00:24:36,210 Josh: I encourage anyone to try to displace it because uh that would 420 00:24:36,210 --> 00:24:39,110 Josh: be amazing if it gets better than this but it's worth trying 421 00:24:39,110 --> 00:24:41,950 Josh: it's worth sharing what prompts you use that 422 00:24:41,950 --> 00:24:45,610 Josh: give you some specific outputs that you may find helpful interesting 423 00:24:45,610 --> 00:24:49,310 Josh: the use cases are the currency please share yours in the comment section down 424 00:24:49,310 --> 00:24:51,610 Josh: below if you enjoyed this video don't forget to share it with a friend who may 425 00:24:51,610 --> 00:24:54,750 Josh: also want to generate some images perhaps they're involved in social media perhaps 426 00:24:54,750 --> 00:24:59,470 Josh: they just want to redesign their hypothetical apartment whatever it may be it's 427 00:24:59,470 --> 00:25:03,910 Josh: fun it's worth testing it's It's worth trying to just like feel it and understand the intelligence. 428 00:25:04,150 --> 00:25:06,230 Josh: But yeah, I think that's pretty much it for today's episode. 429 00:25:06,290 --> 00:25:07,590 Josh: Are you there any final thoughts here? 430 00:25:07,590 --> 00:25:13,410 Ejaaz: Nope. If there's one request that I have, I want to see the images that you 431 00:25:13,410 --> 00:25:17,510 Ejaaz: generate and try and surprise us, try and do a use case that we haven't covered 432 00:25:17,510 --> 00:25:20,970 Ejaaz: on this particular video, because I'm curious of the creative purposes around this. 433 00:25:21,550 --> 00:25:24,490 Ejaaz: Our social media profiles will be linked below. DM us there. 434 00:25:24,790 --> 00:25:27,250 Ejaaz: And yeah, I look forward to seeing what you have to make. 435 00:25:27,670 --> 00:25:29,530 Josh: Awesome. Cool. All right. We'll see you guys in the next episode.