00:00Hello everyone, welcome back to the channel. In this video we're going to be talking about
00:04ChatGPT images 2.0 that just dropped an hour ago. Now this is actually a lot more important than I
00:10think a lot of people think. AI images have absolutely taken over when it comes to YouTube
00:16thumbnails, images on meta ads, just about everywhere. Here you can see a UGC example
00:22where I gave it this AG1 product photo with big text, medium-sized text, and small text.
00:30And when you look at this image that it one-shot, you can see that the text looks perfect on
00:36the
00:36product that the man is holding up, and the man looks extremely realistic. And as an AI marketer,
00:43one of the most common things people use image models for is start frames for videos. So this
00:48is an example of GPT-2 creating a start frame for what would be a piece of content, as you
00:54can see
00:54very similar to this one, of a video that I would post on my Lifehack brand on Instagram that I've
01:00gotten over 900 million views creating this type of content. And if you want the behind the scenes on
01:07how I got over 900 million views on Instagram this year, as well as all the templates and automations
01:13that I use, as well as joining over 1,300 members that are absolutely obsessed with AI image and video
01:20generation, make sure to join my school community. The link will be down in the description.
01:26So really quickly breaking this down, I think the aesthetic looks really good. It looks somewhat
01:30realistic, but also cartoony. You can see the two faces right here. And then another little detail is
01:36it was able to embed the Lifehack watermark really well into the middle. So I think it did a really
01:41good
01:41job at this image. I'm going to try to keep this video pretty short and to the point on all
01:46the use
01:47cases that I think are valuable with GPT 2.0. One thing to note is you can use it on
01:52ChatGPT directly,
01:53just asking for images and it could do 9 by 16 here, as well as they integrated it directly into
02:00FreePick. I'll have my referral link in the description. And just so you can see right here,
02:04you can choose only these dimensions. I'm pretty sure they're going to be updating them with more
02:09shortly as ChatGPT supports more. But one thing I wanted to note right here is you can generate on
02:16high version and I can generate about 400 of these with my $45 a month plan. And then on medium,
02:22I can do about 1300 images just for reference if you're planning on subscribing to FreePick.
02:29And on ChatGPT, you could basically ask it for free. I have the $20 plan and I've asked it quite
02:35a
02:35few times and it hasn't stopped me yet. But yeah, one thing I noticed with GPT is it's a really
02:41good
02:41image creation model. It's not as good, I think, at editing images as much as Nano Banana 2 is. But
02:48I
02:48really think that it's going to put up a very strong fight when it comes to comparing this to
02:53Nano Banana 2. So let's go through some of the first examples. This is just a selfie from an iPhone
02:59of a woman. And the whole point of this model really is to try to make it look as realistic
03:04as
03:05possible. So I think this looks pretty good. I don't really see any flaws with it. I think it's
03:09almost imperfect to a level that it looks more realistic. And then let's scroll down right here.
03:15We have the man selfie in the gym. As you can see, the sweat looks pretty good on this one.
03:20It just
03:20looks overall realistic. And every month we're essentially getting more and more realistic AI
03:26and it's becoming harder and harder to tell between reality. So all my prompts I took from an AI
03:33and basically this one was supposed to test motion and everything. As you can see, the person looks
03:39really good. But there is a little bit of glitching up on the skateboard. You can tell this is AI
03:44because the skateboard has definitely changed. So, you know, overall the person, the aesthetic,
03:49really, really good. But you can definitely tell that the model can still make mistakes
03:54based on this render. This is another test right here trying to test how realistic an image would look
04:01like of someone just writing on the piece of paper. As you can see, if you really like look at
04:08it
04:08closely, you can see there's some kind of like mistake right here. But some of the letters just
04:13look a little bit too good. Maybe it's just my handwriting. But overall, it's very impressive,
04:18but maybe not completely perfect. Going down to this one, this is supposed to be a test of a piano
04:25player with a wedding ring on the left hand. As you can see, it got that detail and the hands
04:30and the
04:31piano and everything look very realistic to me. Music notes right here, you can tell, you know,
04:37it doesn't fully understand what the music notes are supposed to look like. But other than that,
04:41the image is really good. Scrolling down, this is just supposed to be an iPhone with a SAS splash
04:47screen. You know, it's very simplistic, but it does look good. And the text looks really sharp,
04:52unlike probably how it would be in the past. Scrolling down here, we have Tokyo at 2am.
04:58The lighting looks really cool. The reflections look really cool. I think it did a really good job
05:03at this. Then we have a macro photo of a raspberry. I'm not sure how realistic this looks like, but
05:10I
05:10mean, I think it did a good job overall. Then scrolling down, we have three people of different
05:16generations. So this is supposed to be the grandmother, the mother and the daughter.
05:20You can see it looks pretty good. One thing I do notice and start to notice when I generate this
05:25is you get these little kind of like imperfections. It's kind of hard to describe,
05:30but there's just this like stylization that I see really hard here and here that I just noticed
05:36quite a bit in a lot of the Chachupis generations. But other than that, it looks very impressive.
05:42Okay. And then going on to this, this is a macro shot of a bee. I mean, this one looks
05:46pretty
05:47impressive to me. You can tell, um, you know, the little pollen and stuff like that. Um, but the
05:53little cross hatching, or I don't even know how to describe it, but it's a little bit obvious in this
05:58photo too. And then here is the last test I did directly on ChatGPT. This is a really cool one.
06:04Something GPT, even from image 1.0 a year ago could do. It can create transparent images. And as a
06:10designer slash, you know, creator on YouTube, like this is actually really huge because, you know,
06:16the fact that it's extremely transparent and usable when put into other images and stuff like that
06:23manually, it just goes a long way. And then lastly here, I had it create this, which you may recognize
06:31from my actual background. This is not actually where I am. I've been recording this entire time
06:37with a background that was generated by GPT 2.0. Now let's quickly go to the generations I did
06:44on FreePick. These ones have reference images. So we're testing a little bit more, but the first ones
06:49are just the same prompts as I give on ChatGPT. As you can see, this woman looks like a very
06:55realistic
06:56selfie. Definitely something you could catfish people with. I think I like this one even more.
07:01This is the guy in the gym. Just the details of like the sweatiness and the skin, ChatGPT has nailed
07:08on this update for sure. Now this one right here, which I'm going to pop up, if you guys know
07:14Jin Yang
07:14from Silicon Valley, all these images are just based on him. This one is him flexing with the
07:20Lamborghini. I thought this one was pretty funny. I think the chains kind of look a little unrealistic
07:25how the pendants are, and it doesn't look exactly like him. So I wouldn't give this the best rating,
07:30but here is him on Wall Street. I thought this one was a lot better. It looks just like him.
07:35I really
07:36like how the settings work and how you have all these people and the screens very similar. This part
07:40looks a little weird right there. Other than that, I think it looks pretty good. You have
07:44Jin Yang in UFC. I thought this one was pretty funny, but it looks pretty realistic in my opinion
07:51and a lot like him. Then we have Jin Yang as a samurai. You know, not too much there. Then
07:59we have
07:59Jin Yang as a space explorer with the SpaceX logos. I mean, that looks really good. I think overall,
08:05I really like this photo that JGPT created. And then here is the new test. The new test,
08:13I gave it this reference image, and I gave it specifically this AG1 boxing because it has
08:20packaging, because it has this big text here, medium-sized text, small text, and then extremely
08:25small text. So when we look at this image, you can see one of the improvements is the text is
08:31substantially better. On other AI models, it would really mess up small text. But here you can see
08:36the big text is perfect. This text is perfect. You know, this looks really, really good and really,
08:42really good. And you can't see it perfectly, you know, without zooming in, but it is miles beyond
08:47what it used to be. And then we had this test that I started the video with, you know, very
08:52realistic
08:52skin. The box looks really good in this one. Then we have this image. I thought this image was also
08:58great. You know, this is a perfect way to start a UGC video. You just need to generate these images
09:04and then animate them with C-Dance or VO. This one was good, except for the guy has two arms
09:09on the
09:10right. So it definitely messes up. But like the solar flare right here, the box, everything else about
09:16it is very, very impressive. Now here is the thumbnails. Now, oftentimes it would create these like
09:21three side by side. I think it's just how my prompts are that are kind of like messed up.
09:25But this image that I showed you in the past, it was a really good start frame for my videos.
09:32And then right here, I did one more test where I put Jin Yang into an actual manga that was
09:38colored.
09:39And as you can see, it looks like him. It's able to create all the cells and everything in one
09:44image
09:44with all the instructions. So this just illustrates how good the model is at coming up with more graphic
09:50designs. That's something that they demoed a lot in the video and that I'll maybe go into more
09:55detail in the future. Anyways, these are all the tests I wanted to show. I want to get this video
10:00out as soon as possible. Thanks for watching and I'll see you guys later.
Comments