Latest / Bad Decisions Podcast / GPT Image 2.0 KILLED Nano Banana 🤯 ?
Transcript
- 0:00GT Image 20 is out and it is undoubtedly the strongest image
- 0:06generation model out there today.
- 0:08We've tested it and we're going to show you guys the results and
- 0:11also who it's for. On the second topic, we're going
- 0:15to be talking about the fastest visual effects pipeline that
- 0:19we've tested and the results are pretty incredible.
- 0:23Also test it and we're going to show it to you today.
- 0:25If you always wanted to create high quality motion graphics,
- 0:29you never had the time or skills to do it.
- 0:32Now you can use AI to create high quality motion design.
- 0:36We tested two different methods and we are going to show both of
- 0:39them to you guys today. So we've got lots of demos for
- 0:43you and we hope that you guys are ready to start with that.
- 0:47Welcome back to the show. We've been away for a week and I
- 0:50know you guys know that because you guys have sent us DMS,
- 0:54commented under some of the videos.
- 0:57This is Season 2. Now what are we going to get
- 1:00with Season 2? There are some differences.
- 1:01We've actually upgraded certain things and you're going to see
- 1:04the roll out of the upgrades over the next few episodes.
- 1:08One of which is, by the way, our quality is not 4K.
- 1:11We realized that we were recording everything at 1080P.
- 1:14So you're going to get a crisper camera quality.
- 1:17We're going to be upgrading the UI of the show.
- 1:20So whatever you're seeing right now, over the next few episodes,
- 1:23we're going to be upgrading the layout of the show.
- 1:26So you're going to get an update to that.
- 1:28We're also introducing clips. So for anybody who may miss one
- 1:34of the episodes, perhaps one week you guys are busy because
- 1:37don't have time to watch the full episode.
- 1:39We're going to be cutting up the episodes into clips so that you
- 1:43have the ability to watch 10 minute or 15 minute long videos
- 1:47on YouTube that will be specific to one single topic.
- 1:51Another addition to the show is going to be our newsletter.
- 1:54If you don't have the time to watch the show or listen to the
- 1:57show for one hour, we will send you the summary which takes you
- 2:01less than 3 to 4 minutes to read.
- 2:03This is all the tech updates, all the tool updates, all the
- 2:07demos and reviews that we do. So if you guys want to get that
- 2:10newsletter, you can go to AI dot bad decisions dot studio.
- 2:14You sign up there and then we will send you the newsletter
- 2:17after every single show, which is now Tuesdays and.
- 2:20Fridays every week. That's right.
- 2:22I also want to let you guys know that we've updated the timeline
- 2:26and the scheduling of the show. We're now going to be going up
- 2:30on YouTube and. Spotify.
- 2:32An Apple podcast. An Apple podcast every Tuesday.
- 2:35And every Friday there we. Go.
- 2:39Some water I I saw you you were becoming an org from Lord of the
- 2:45Rings. The show will be on Tuesdays
- 2:49and. Fridays, please watch and.
- 2:54By the way, this is how much value we give to this show.
- 2:58He was dying and he didn't cut off his word to drink a glass of
- 3:02water. He just want to tell you guys
- 3:04that the show is going to be two days a week.
- 3:05That's right, every Tuesday. Every Friday, which is great
- 3:09because you get 1 update for the weekdays and you get another
- 3:13update that you can watch over the weekend.
- 3:15And that is gonna summarize everything that is happening in
- 3:17the world of technology. And we're gonna be trying our
- 3:20best to bring you guys the latest and the greatest, as we
- 3:23always do, but always with transparency.
- 3:26We want to show you guys what is hype and what is practical.
- 3:31That's always been the essence of this show, and we hope that
- 3:33we can keep that moving forward. And just summarizing everything
- 3:37that you're going to be seeing in Season 2, we're upgrading the
- 3:40entire production quality. You're going to see it in the
- 3:42thumbnails. You're going to see it in the
- 3:44video, you're going to see it in the UI.
- 3:46You're going to see it also on short form content that we'll be
- 3:48releasing on different platforms.
- 3:50But All in all, the show is going to go on strong, stronger
- 3:55than ever. Of course.
- 3:56Let's get started with the first.
- 3:57Topic. Let's go.
- 3:58GPT image 2 point O. This is a state-of-the-art image
- 4:05generation model by Open AI and it is the strongest thing we've
- 4:10ever seen. Just when we thought nano banana
- 4:12by Google is going to be the king for a long time, it is
- 4:17already cast it off into the shadows by GPT image 2 point O.
- 4:23And we're going to prove to you guys why.
- 4:25You see, this model is able to reason, research and produce
- 4:32production ready visuals, but not just produce them, also edit
- 4:37them. And why that is important is
- 4:40because this is unlike some of the other image generation
- 4:44models. You're not just talking to a but
- 4:48that only understands your prompt one time and creates an
- 4:51image when you're going to ChatGPT, an AI model that has
- 4:57understanding of you as a person, what you like, what you
- 5:01dislike, what projects you're working on, can take in PDF
- 5:05documents, YouTube videos, can take in your branding guidelines
- 5:10and then can research the Internet and then generate an
- 5:13output. You can bet that it will create
- 5:16the best quality results. Now, we've already had that sort
- 5:21of with GPT image one or Dolly was the first name.
- 5:25But of course, as the new generation of LLMS are released
- 5:30like GPT 5.5, the understanding and the reasoning capabilities
- 5:36of these models for image generation also increases.
- 5:39We saw that with Nano Banana to Nano Banana Pro, right?
- 5:43And we saw with Nano Banana 2 when we had better text
- 5:46rendering. Yeah.
- 5:48But all I can say is that this is just Nano Banana Pro Nano
- 5:51Banana 2, but 2345 times better. So where do we actually use GPT
- 5:58image 2 point O? You want to make ads?
- 6:00You're going to use GPT 2 point O social graphics, GPT 2 point
- 6:04OUI and UX design, GPT 2 point O prototypes, GPT 2 point O
- 6:11product photography. GPT 2 point O.
- 6:14Infographics. GPT 2 point O teaching materials
- 6:17GPT are we GPT 2 point O game prototypes GPT 2 point O design
- 6:23systems GPT 2 point O thumbnails GPT 2 point O photo.
- 6:28Shoots. GPT 2 point O.
- 6:32We've literally sorry about my voice today.
- 6:34We've literally tested this for the entire week and it has never
- 6:41failed us. This thing has better
- 6:43instruction following than any other image generation model out
- 6:46there. It is so smart that when you
- 6:49give it the stupidest prompt in one line.
- 6:52And Faras does a lot of. And I do that when you're lazy.
- 6:56It understands everything on its own.
- 6:58It's that level smart. It's very consistent with human
- 7:02faces, so you can give your own face and make sure that it stays
- 7:06like you throughout the different generations that you
- 7:09have. And it has superior text
- 7:11rendering across all the other models guaranteed.
- 7:16And All in all, once you have all of this, the biggest
- 7:20question that we're going to answer today is for most people
- 7:23and for most use cases, do you really need Photoshop or Canva
- 7:29anymore? Do you really need that?
- 7:31Or can you literally just edit anything you want with GPT 2
- 7:35point O? Let's find out, Farad, you have
- 7:37some. Examples there there are I would
- 7:38say you know with every image release there's a series of
- 7:42posts that gonna go viral. In the previous GPT release was
- 7:47a Ghibli studio style. In this release, a lot of people
- 7:50have tested different hairstyles, yes, different
- 7:54hairstyles and hair colors. So they give their image to GPT
- 7:58and it generates different hairstyles, different colored
- 8:02T-shirts and it will tell you which one is the best color that
- 8:05matches your skin tone, your facial shape.
- 8:08So if you have an Oval shaped face or if you have a circular
- 8:11shaped face, which one suits you better?
- 8:13I've seen a lot of people posting this.
- 8:16I have also seen people posting the photo of the palm of their
- 8:20hand for GPT to do palm breathing and then it does.
- 8:24But then the community notes was saying don't do it because then
- 8:28they can get your biometric data.
- 8:30So it says don't do it. Don't do that because I've seen
- 8:33a lot of people posting that too.
- 8:34Also I've seen practically go to Riley Brown's page and Scroll
- 8:39down. He asked GPT image to O to
- 8:42generate an image of a book with the barcode.
- 8:46He then went and scanned the barcode and the barcode worked.
- 8:49What you were saying about understanding the context and
- 8:52doing an actual research and getting all the information from
- 8:56the Internet, It even goes to the level of the barcode of a
- 8:59product. So when you're scanning an image
- 9:02that is AI generated, you can go to the actual product and
- 9:05purchase it. The details are insanely
- 9:08accurate. That's what you're trying to
- 9:09say? Yes, 100%.
- 9:11So I have some use cases I want to show you guys as well.
- 9:14The hairstyle is really cool, but I want to talk about some
- 9:18practical use cases I've personally.
- 9:22Because you don't go to a hair stylist.
- 9:24That's why you I've tucked. My hair in the longest time, so
- 9:26I don't, I don't need to. But until we get that, if you
- 9:30don't have that, we can. Do you have it or not?
- 9:35I don't think I have it, but let me if you don't.
- 9:36Have it. Farrah can find it.
- 9:37But let's go to the next one first.
- 9:39Farrah can send it to you. Yeah, I'll send.
- 9:41It can we quickly go and do a test demo with Farrad's face.
- 9:48I want to prove to you guys how incredible this is.
- 9:52So what you're seeing here is an image that Farrad never took.
- 9:57I want to show you guys what this is.
- 9:58Go to the very top. So this is a photo in large
- 10:02Farrad's photo. This is a photo that we took
- 10:03here in the studio for our thumbnails.
- 10:06As you can see, the lighting is what we're using here in the
- 10:08studio, but this can be any lighting.
- 10:10This can be an iPhone photo for all I care.
- 10:13It doesn't actually have to be a professional photo.
- 10:16Once you have your photo, then you can go ahead and take a
- 10:20reference from Instagram or Pinterest, which is where we got
- 10:24our reference. Say we need Farhat in a very
- 10:27specific pose and a very specific lighting, as if he were
- 10:32shot shooting him in a professional photography
- 10:36portrait for a magazine. So the second image is just
- 10:39that. This is what we got from
- 10:41Pinterest. There are thousands of these
- 10:43images. Now, mind you, if you wanted to
- 10:45go ahead and take that image, you need a good quality camera.
- 10:49You need a great setup of lighting background.
- 10:53Good lens. And you also need the lens.
- 10:55More importantly, you need the skills to be able to to set up
- 10:59the lighting in a specific way and take that photo and then
- 11:02color grade it in Lightroom. What if you don't have to do
- 11:05that anymore? What if you can go to ChatGPT
- 11:07and just say, I want you to use the first person image and copy
- 11:11the aesthetic style, lighting and background of the second
- 11:14image. Make sure you copy the pose of
- 11:16the second image. See that prompt is the laziest
- 11:19prompt. We're not writing five
- 11:21paragraphs. We're not negative prompting,
- 11:24we're not going to another ChatGPT chat and asking it to
- 11:27create an at a banana prompt, which is what we used to do.
- 11:31We wrote 3 lines and let's take a look at the result.
- 11:36Majestic. Can you just say change his polo
- 11:40shirt to a black turtleneck and keep everything else the same?
- 11:48I should put this on my profile picture.
- 11:51You don't even have a photo that looks as good as this.
- 11:54Sorry. You don't even have a photo that
- 11:56looks as good as that. You don't.
- 11:58Have you been on my Internet? You've never taken a photo.
- 12:00I'm just being serious, right? You've never taken a photo as
- 12:02good as that. And that was done in a minute.
- 12:05Now the generation time is slightly longer than Nano Banana
- 12:11Pro. I've realized Nano Banana Pro
- 12:13seems to be about 2030 seconds. This can go up to a minute
- 12:16sometimes, but with a pro plan you can actually generate
- 12:19multiple variations at once. So you can actually generate 8
- 12:23variations of the same photo with different colors at 1 go.
- 12:28Actually, let me show it to them.
- 12:31So this is Pinterest. This is where we got our
- 12:33reference from. Can we test one more thing?
- 12:36Sure. You pick any photo that you
- 12:38want. You can choose a different pose
- 12:40or different lighting. Let's see.
- 12:42Let's see if it's able to replicate this style.
- 12:47Well, this is also the. In there that has the text
- 12:50prompt in it. Pratt, be careful.
- 12:52It should be able to take it. Let's try it.
- 12:55Or we have a better version here.
- 12:58Can you say that and? Then you have to put files
- 13:01image. Yes we can.
- 13:03We can also input multiple images by the way.
- 13:06Let's see. You can, but then you have to be
- 13:08very specific about what you want.
- 13:10OK, OK, so this is done by the way.
- 13:12I have a turtleneck. You now have a turtleneck and
- 13:15the face remains consistent by the way.
- 13:18And if you have a problem with the face, you can re upload
- 13:20Farrad's face and say change the face and make it even more
- 13:23accurate. Again, this is a very smart
- 13:26chat, right? You're not talking to a dumb
- 13:28image generation. Well, you're talking to ChatGPT
- 13:30which is incredibly capable. So you can just talk to it like
- 13:33a normal person as if someone is making that photo.
- 13:37Farrad new photo from the podcast.
- 13:39No, we just took a screenshot now.
- 13:45What are you going to write? Can you make the third picture?
- 13:51I'm trying to write a very dumb friendly prompt in a way.
- 13:55OK, let's see if Chad GPD understands that and put them in
- 14:02the first to. Put them.
- 14:05No, that's really not right. There's.
- 14:06Not right. You're not only dumb.
- 14:11Can you make the third picture and put him in the first?
- 14:14Oh my God. Four shot GPTR to analyze this
- 14:18no one. Is that dog Brad?
- 14:22What are you doing? Put in the first two references.
- 14:25Make sure the clothing. Make sure the clothing an art
- 14:28style is the same. Same as what?
- 14:33And you can give him the glasses.
- 14:35OK, let's see for charge of it. While this is being generated, I
- 14:39want to show you guys our exploration with thumbnails.
- 14:41If you can go to the Fig jam prod, this is where it gets
- 14:44super exciting. So just zoom in on the top left,
- 14:46zoom in on the top left photo. This is where you can actually
- 14:50understand what the use cases of this is.
- 14:53This image is a basic thumbnail that we made using Canva.
- 14:58We needed the composition to look like this for the clips of
- 15:02this show, which is going to be released very soon.
- 15:04You have either me or Farah on the left side, the hosts, you
- 15:08have the Bad Decisions logo on the top left, you have the title
- 15:12of the thumbnail on the top right, and then you have a
- 15:15visual explaining what we're talking about in the show.
- 15:18Now, why this is very important is because you can translate
- 15:22this to any other use case. If you have an ad or product
- 15:25photography or a poster or an infographic, you're gonna have
- 15:29similar things, right? You're gonna have faces,
- 15:31visuals, text, gonna have logos and text.
- 15:34Now we are assuming most people's capability lies right
- 15:39here. Everyone can go to cap cut or
- 15:42Canva and create an image like this by just putting things
- 15:45together, but this is not something that will grab
- 15:48attention. What if you could take already
- 15:51existing art styles and implement that as an
- 15:54inspiration? As if you literally went to
- 15:56Photoshop and try to copy somebody else's art style and
- 16:00get inspired from multiple things and apply it to your own.
- 16:03This is what IGBT can do. And it does it in a one shot
- 16:07prompt. So if you just slowly go to the
- 16:10right Pratt, this is an image that we took from Pinterest.
- 16:14We literally wrote this prompt. Can you reimagine the first
- 16:18podcast thumbnail in the style of the second image and let's
- 16:21see what we got. You go to the right, boom, it
- 16:26took my photo. And if you just quickly go back
- 16:29to the left, look at the lighting, the rim lighting on my
- 16:31hair, it doesn't exist in the first photo.
- 16:34It understands artifacts. You see, it cleaned it up and it
- 16:38upscaled it. Look at.
- 16:41That perfect. It upscaled the image.
- 16:43It added the rim lighting. Then it has the logo of Bad
- 16:47Decisions on the top left. It kept that, it added the grid.
- 16:51So if you zoom in all the way to the end of the background,
- 16:53you'll see the grid. OK, go to the reference image on
- 16:55the left. You also have the grid.
- 16:58See that grid? So it understood all of that
- 17:00without me saying a single thing about it.
- 17:02Then the text is this legal is copied with the blurry bit.
- 17:07You see the stainless has the standings on business.
- 17:11Sorry, has the S blurred out. Go to the right and you're going
- 17:15to see that same blurry effect on the question mark.
- 17:18The plus icons are here too. The Indian factory workers
- 17:22training AI robots. That visual is copied over, but
- 17:24the lighting has changed to match the aesthetic style of our
- 17:29reference image. Pay close attention to the
- 17:32subtext. This is what I meant by superior
- 17:34text rendering. There is subtext added without
- 17:38us even asking for it and it is contextual to the visual.
- 17:41We didn't even explain what the visual is.
- 17:44Let's zoom in. AI training feat, head Cam,
- 17:48active workflow capture, all these added by AI automatically.
- 17:55Now, we took this one then because we liked it, we said,
- 17:59hey, what if we just change the title and the visual and also
- 18:04refer back to the reference one more time so that we copied that
- 18:08aesthetic style even more. And then the other image was
- 18:11generated. It's happening, that one.
- 18:13So we literally put in the prompt on the right side, go
- 18:16closer, we said, change a title to it's happening.
- 18:19Change a visual to anthropic on a throne.
- 18:21This is the kind of topics that we talk about.
- 18:24Now go closer. You'll see this time it matched
- 18:27the reference even more. The color of the background is
- 18:30slightly lighter Gray, which is true according to a reference.
- 18:34The angle of the photography changed.
- 18:37It changed my image. I don't even have that shirt.
- 18:40It changed my image and my angle of photography and the pose to
- 18:43match the original reference from a low angle.
- 18:46It did the same to the visual. If you look at the throne, it's
- 18:50captured at a low angle. Then let's get closer to the
- 18:53subtext Brad, the new AI leader. Question mark.
- 18:57Just because we said anthropic on its throne, it understood
- 19:00that, added it as a subtext. Imagine all of this missing
- 19:03detail that you it. Would take you ages to just come
- 19:06up with the words and then just put it there.
- 19:08OK, you guys get the gist. Brad, let's zoom out.
- 19:11Just quickly show the other one. So we did the same for brain
- 19:14simulation, switched me and Farhad.
- 19:18This is the fruit fly brain topic.
- 19:20That we cover. Now let's zoom out and let's
- 19:23just quickly scan through some of the other stuff that we made
- 19:25with the references at the bottom.
- 19:26Yes. So we took again a random photo
- 19:32again, there's really nothing beautiful there.
- 19:34And then we took the poster on the right.
- 19:36Go to the right. You're seeing that R style
- 19:39copied over in the same topics, by the way, then thropping on
- 19:43the throne and the Indian factory workers training AI
- 19:46robot and the brain simulation. Let's go down.
- 19:49Same thing, original image reference and we're seeing the
- 19:53cut up grids into the photos with the text.
- 19:57Let's go back down. You see, in order to get a good
- 20:00result, how much iteration and how much brainstorming we have
- 20:05to do, we have to test different references, we have to test
- 20:09different samples and you know, different visuals to see which
- 20:12one will turn out good. And actually, a combination of
- 20:16these would work the best, just like you would if you were to
- 20:18create it manually. You would want a combination of
- 20:21inspirations to come up with what's the right result for you.
- 20:25But what we're showing to you here is how insanely capable
- 20:29this new image generation model is.
- 20:31And we definitely recommended you try it out for your sake.
- 20:35Whatever it is that you're doing, whether it's personal
- 20:38work or client work, this is a time saver, a lifesaver.
- 20:42And honestly speaking, what's insane about this is that if I
- 20:46were to do this manually, it would have taken me at least
- 20:50three to four weeks to create all of these different
- 20:54iterations that Pratt is showing you.
- 20:56And perhaps I wouldn't even feel as creative.
- 20:59I know it sounds weird, you would say, hey, but what's
- 21:02creative about prompting these? Technically speaking, I would
- 21:05have been so fed up with my initial designs in the first
- 21:09week and get so tired. But here the reason why I
- 21:13actually felt even more creative is because I got to experiment
- 21:15with 506080 iterations to get to my final result and I love the
- 21:22final result. Pride.
- 21:23Can we show? Yes.
- 21:24So we took this is this is where it gets interesting.
- 21:27We took an already generated image, which is the one we
- 21:31created at the top. And then we said, OK, we like
- 21:33this style, but we want to add in another element to it.
- 21:37We want to add in this specific art style which is from the
- 21:39Nothing phone. If you guys have consumed DJI
- 21:42products or used the Nothing phone or seen their videos, they
- 21:46have a very specific style which is transparent tech and these
- 21:50colour schemes. And we just said, OK, let's try
- 21:53to add this in, let's see what we got.
- 21:55And this has been the closest to what we really like.
- 22:00I mean, look at that. The design is insane.
- 22:03Text rendering, face rendering. Can you go closer to my photo,
- 22:09guys? I don't have that shirt on and I
- 22:12never took a photo with that lighting.
- 22:14This is all matching directly to the background of the thumbnail.
- 22:18So if you pay close attention to the original images that we
- 22:21have, the lighting is different. So go closer.
- 22:24You can see this is my original shirt, it's changed it, and this
- 22:27is not even my lighting. It's made it better and more
- 22:29upscaled for the sake of the thumbnail.
- 22:32It basically understood the background is light Gray and
- 22:35went for a lighting that matches that.
- 22:36And it's also gave you the really good contrast, so as a
- 22:40thumbnail you really want to click on it.
- 22:42It's amazing. Now what?
- 22:45Did it generate my image by the way?
- 22:46Of course. Oh.
- 22:49Wow. I should post this.
- 22:52No, no, no, I like the first one.
- 22:54Yeah, this is this is fucked it up a bit.
- 22:56This is really good. You can, by the way, immediately
- 22:58give back another photo of yours.
- 23:00And then say, do they get the reason?
- 23:02I think it's the power is in the back and forth that you do with
- 23:05the model. You probably won't get what you
- 23:08want in the first try. You just have to experiment with
- 23:11it. Go back and.
- 23:12Forth to GPT, you do get it in the first try with GPT image 2 a
- 23:16lot of the times. I mean, this time it didn't give
- 23:19a idea. Depends on what you want and how
- 23:21well you're describing. Again, let's be honest, Pratt's
- 23:24description is pretty shit here. And that was his purpose, right?
- 23:27He wanted to keep it dumped. I mean the.
- 23:29First image is good, just the second one is not good.
- 23:32The of course, the more you describe the better it will be,
- 23:35but generally speaking, we are truly amazed by what this model
- 23:39is capable of. If you guys have already tried
- 23:42it and I've seen the results, let us know in the comments.
- 23:46I'm so curious to know what your experience with GBT image 2
- 23:51point O has been. I find it incredibly exciting to
- 23:54use. It's so fast and it's so capable
- 23:57and it's so smart. One last thing I do want to say.
- 24:00Everything we just showed you was creation.
- 24:03I would also recommend that you try it for editing.
- 24:05So if for instance the thumbnails that you saw, we
- 24:09needed to just change the font or the color of something, you
- 24:12can just feedback the generated image and ask it to just change
- 24:16the one thing and it will keep everything else the same.
- 24:18You can just drag and drop the image into the chat, right?
- 24:21Yeah, so you can literally just drag that and say change the
- 24:24background to blue and it change the background to blue and keep
- 24:27everything else the same and it will just edit that, which is
- 24:30why I posed this question in the beginning.
- 24:33For most of our use cases, do we really need Photoshop anymore or
- 24:37do you really need to go to Canva?
- 24:39If you're making an ad for Instagram or a carousel for
- 24:43Instagram, do you really need to go to Canva and every time edit
- 24:46everything or can you give it to GPT 2 point O and have it?
- 24:49Done. I think with GPT 2 point O
- 24:51onwards, maybe you can previously if you wanted.
- 24:55Usually for ads what what is going to happen?
- 24:57Why they go to canvas? You want the same ad in
- 24:59different form, in different format different.
- 25:02You can do that expect ratio now you can do that here and it will
- 25:05keep everything the same yes right.
- 25:07So previously you couldn't do it.
- 25:08Now I think with this a lot of people will switch.
- 25:11How is the? I mean, the generation speed is.
- 25:14Good. It's about a minute.
- 25:15Yeah, it's about a minute To generate the images.
- 25:18You have control over the aspect ratio, so you can create for a
- 25:22vertical screen, a horizontal screen, a wide screen.
- 25:25It's up to you but understand something we're not saying.
- 25:29Oh everything else is dead. Just go to ChatGPT and generate
- 25:33images. Please know that these guys have
- 25:36AP is and Canva and other tools are going to be adding GPT 2
- 25:40point O. So I'm not saying GPT 2 point O
- 25:42is killing everything else. I'm just saying for most use
- 25:44cases, it's just faster to come here to chat and drop everything
- 25:48and get your results. Canva and other tools are going
- 25:53to add GPT 2.0 so that you can generate with myself.
- 25:58Yeah, I know. I see that.
- 25:59What I'm saying is all the other brands that you may be using at
- 26:02the moment, they're gonna have GPT 2.0 just like they had added
- 26:06Nano Nano. Yeah.
- 26:07So then you can go and further edit things in the way that you
- 26:10like For me, for the thumbnails that were created for the show,
- 26:14this chat has been incredible. We're even redesigning the UI of
- 26:18the show by redesigning everything within ChatGPT.
- 26:22I think it's also because you can just upload everything in
- 26:25there. You know what I'm?
- 26:26Saying And it has context about that decision, about our show.
- 26:29It knows everything about us. You can give it the branding
- 26:32guidelines and you can tell it to go research bad decisions and
- 26:35create an image based on what it knows.
- 26:37You cannot do that. We actually, OK, another example
- 26:41I have for you guys. You guys know that we actually
- 26:42love using Higgs field for a lot of our images generations.
- 26:45If we're working on a full short, let's say a one minute
- 26:49video, we're going to go to Higgs field and create
- 26:51everything. But when we're working on
- 26:53thumbnails, I actually tried using image 2 point O by GPT in
- 26:57Higgs field. It's not the same.
- 26:58Of course, because it doesn't have the chat.
- 27:00Context because you don't have the chat context.
- 27:01That's what I'm saying. Depending on your use case, you
- 27:04might want to go to ChatGPT to create your images now or go to
- 27:08Higgs field, or go to Canva. One last thing I want to end
- 27:11this with before we move on. Go.
- 27:14What did you write for this prompt?
- 27:15I asked it to make four different variants.
- 27:18Different colors, Different colors.
- 27:19Yeah. OK.
- 27:21The last thing I wanted to say, if I can remember it, I'll
- 27:27remember it. Wow, these looks really like
- 27:32album covers. I should drop a hip hop song.
- 27:34Not hip hop actually. Well, this is so cool, yeah.
- 27:37See, you can ask it to generate multiple generations, yeah, at
- 27:41the same time, and you can get it all in one session.
- 27:44OK. I completely forgot what I
- 27:47wanted to say. If I remember it, I'll bring it
- 27:49back. We're done with one.
- 27:52Shall we move on? Yeah.
- 27:53Yes. Shall we move on to #2?
- 27:54Let's go. We told you guys, promise you
- 27:58guys in the beginning that we're going to be sharing with you the
- 28:01fastest visual effects pipeline that we've discovered recently.
- 28:06And this is quite interesting because it's not the first time
- 28:09we're covering this. We've talked about Autodesk Flow
- 28:12Studio before on this show. We've been at an event with
- 28:17Autodesk. We've talked on a presentation
- 28:21panel about how we use this for our own production.
- 28:26What's really interesting is that they've added two new
- 28:30features, AI rigging and neural layer rendering.
- 28:35And with the entire pipeline now set up, we generally believe
- 28:39that this can be the fastest visual effects pipeline,
- 28:43especially for pre production. But even if not for pre
- 28:46production, you can use it for live action VFX game development
- 28:51for indie creators or even 3D printing for hobbyists.
- 28:56Why do we say the fastest VFX pipeline?
- 28:59Well, they have added the entire end to end streamline system.
- 29:04We're talking about going from an idea in your head to typing
- 29:08it out in one place. You get to take your idea, turn
- 29:13it into an image because they've added Nano Banana Pro as a
- 29:17model. Nano Banana 2 and Nano Banana
- 29:19two, yeah. And then from that image, you
- 29:22can create a 3D mesh using Microsoft Trellis, which is an
- 29:28incredibly capable 3D generation model.
- 29:30We've talked about that on the show.
- 29:32Or you can use Wonder 3D, which is their own proprietary model.
- 29:37And the beauty of that is that once you have your 3D mesh, you
- 29:40can texture it also within the same pipeline.
- 29:44Once you have the 3D textured mesh, you can go in and rig it
- 29:49automatically. And then once you have the
- 29:52rigged mesh, you can then add it on top of your live action
- 29:56plates. Or you can go ahead and export
- 29:58it to your 3D software of choice, whether that's Unreal
- 30:01Engine or Blender or Maya. Everything is there from zero to
- 30:06100. And the reason why we're talking
- 30:07about this again on the show is because there were some missing
- 30:10layers that have been added now to create this complete
- 30:13pipeline. Shall we show you guys what
- 30:15we've done? Last time when we talked about
- 30:17this on the show was they had text to image and then image to
- 30:213D. Now they have added Nana Banana
- 30:24Trellis so we can have different options when we want to
- 30:27generate. The auto rigging part is also
- 30:29added. And the best part is you guys
- 30:32get to decide at the end. Do I want to create everything
- 30:35in Wonder Dynamics or do I want to export it out?
- 30:38And you can export out the mocap, the camera tracks, the
- 30:41clean plate everything outside to Blender, Unreal Engine and
- 30:45then do your render there. So with Flow Studio, what I
- 30:48really like, as I mentioned before, is that you get
- 30:50everything in one place and we're going to show it to you
- 30:52guys. So Pratt, can we quickly go to
- 30:54the very beginning? This is where you generate your
- 30:57images. You give it a prompt of the kind
- 31:00of character that you want to create.
- 31:02And of course, you want to make sure the character is in T pose
- 31:04so that you can easily rig it later on.
- 31:06For our use case, we wanted a banana character like the ones
- 31:10you've seen Fortnite, and the prompt was written by ChatGPT.
- 31:14We told ChatGPT the kind of character we want, got the
- 31:16prompt brought in here, and we are choosing Nano Banana Pro.
- 31:20That's our favorite model to use that honestly, especially now
- 31:23with GPT image 2.0 is the only one that we believe is truly
- 31:26comparable. Alternatively, by the way, you
- 31:29can go to chat DVD image 2 point O and get your image and bring
- 31:32it here because you have the option to go tech zoo 3D or
- 31:35image to 3D. But let's say you don't have
- 31:37anything, so you can generate it right here.
- 31:39Once you have the image that you like, which is here, what do we
- 31:42do? Pratt?
- 31:43We can generate the image OK After we generate the image, we
- 31:46can use it as an input right and to generate a 3D character.
- 31:52And this is how it looks once you are generated it.
- 31:56Using it, oh you get 4 different previews, yes.
- 31:58Because sometimes they might have some artifacts.
- 32:00You get to choose the one that looks the most accurate for you.
- 32:07And once you do that, then you have the 3D character textured.
- 32:11But this is not rigged yet. This is where the AI rigging
- 32:14comes in handy. Of course, as you know, we have
- 32:18done rigging on this channel many times.
- 32:20And for rigging, you have multiple pipelines that you can
- 32:24use. The most commonly used has been
- 32:26Miximo. But for a lot of people, going
- 32:29from three different tools and switching back and forth has
- 32:32been really time consuming and complex, especially if you're
- 32:36trying to be quick on a project. This is the solution because
- 32:40again, you go from idea to 3D to rigged character to ready to be
- 32:45put into a live action plate all within one hour.
- 32:49So what is a rigging process? So on the bottom right you have
- 32:53a rig button. You just click on it and you
- 32:56select human and here you. Can put the points you.
- 33:01Have the markers. So for the wrist elbows, where
- 33:04are the shoulders? Where's the neck and chain?
- 33:08Chain, wrist and ankle and so on.
- 33:10I believe one of the other advantages of this, aside from
- 33:14speed, is the ease of use, as you guys saw.
- 33:18I think I'm going to have Pratt do this one time because we've
- 33:20already rigged this. The rigging process takes a few
- 33:22minutes, but I want to quickly go back to the beginning.
- 33:26Pratt, if you don't mind. Sure.
- 33:29Just so people know, so when you start from the image, the
- 33:32generation of the image, we can choose the image models as we
- 33:36showed you guys from the top left.
- 33:37So you have the option to go for flux, Nana Banana, Nana Banana
- 33:41Pro, Nana Banana two. We usually go for Nana Banana
- 33:44Pro. It's the best quality in our
- 33:45opinion. Then you click generate image on
- 33:46the bottom left and it's done. And within a few seconds to a
- 33:50minute you get your image. Once you have your image, then
- 33:52when you want to try to create your 3D mesh, can we quickly
- 33:56show the models available? Pratt this is important.
- 33:59Sure, here. So you have Flow Studio and you
- 34:02have Microsoft Trellis for 3D generation.
- 34:05So the difference here is that when you go for Trellis, you do
- 34:08use Microsoft's model and sometimes you might prefer that
- 34:12model for your output result. But you have to understand that
- 34:15the AI rigging only works with Wonder 3D.
- 34:19So if you want to go ahead and rig it here within the software
- 34:22and immediately use it for pre production or whatever the use
- 34:25case is, you want to go ahead and choose Wonder 3D.
- 34:27This is a new feature. This banana character that we
- 34:30created was done with Wonder 3D. So you know, the quality is
- 34:33actually good. That's why it's already rigged
- 34:35and we have it moving. So once the previews are
- 34:38created, you select a preview, then you create a character.
- 34:42How do we do it, Pratt? So now we have the preview.
- 34:45I will select the top right one. And here we have multiple
- 34:47option. We can have fixed model, we can
- 34:50have adapted model. In fixed model we can choose how
- 34:53much Polygon count we want. It should be a triangle quad.
- 34:57So let's select Adaptive, Let's select Medium.
- 35:01Adaptive is good because it will find where you'll require more
- 35:06polygons depending on the mesh, and it'll give that specific
- 35:09area more polygons. It's not going to use polygons
- 35:12everywhere for the same amount. That way you probably have a bit
- 35:17more optimization in your model creation.
- 35:20OK, after that all you gotta do is just select whether you want
- 35:24texture or no textures and just generate.
- 35:26It OK and we've already done this as a textured mesh.
- 35:30This is the character we got for AI rigging.
- 35:31We already showed you guys what the process was, and then what
- 35:34do we get when the character is rigged?
- 35:36This, as you can see, the character is already rigged.
- 35:40Now this is the try that we did with the banana character.
- 35:44Of course, you can go out and try this with anything and try
- 35:47it with an alien, try it with a robot, try it with a Viking, see
- 35:50what works and what doesn't work for you.
- 35:54What we wanted to show you guys is actually one of our final
- 35:57results that we did with an alien character with the exact
- 36:02same process. We went from idea to image to 3D
- 36:05model texture to a rigged character and then we dropped in
- 36:09a video that we recorded a really long time ago.
- 36:11We can see it from your shorter that video was recorded at least
- 36:14a year and a half ago. And Pratt, I guess what we want
- 36:18to showcase here is it's something important.
- 36:23Why? Because there's also another
- 36:25feature called neural layer. This neural layer is where
- 36:31Autodesk Flow Studio wants to give artists an option to go
- 36:34ahead and use AI to enhance their videos.
- 36:38What do we mean by enhance? Well, we're talking about
- 36:41specific things such as increasing the quality of the
- 36:44lighting, increasing the quality of materials, responding to the
- 36:49lighting in the scene, and of course adding in subtle
- 36:52expressions on the face such as facial animation.
- 36:56What we actually saw in the results was fascinating.
- 36:59We tested this across the board and to be very honest with you
- 37:02guys, there are generations where we prefer to use the
- 37:06normal render and not the neural rendering.
- 37:09Sometimes a normal render itself is actually great, but there are
- 37:13other times like this example where the neural layer rendering
- 37:16completely beat the normal rendering and I want to show it
- 37:20to you guys so you guys understand before you play this.
- 37:22On the very left, we have the original clip of me, and it is
- 37:26actually a very complex clip because Farhat is moving the
- 37:30entire time. So the actual camera is moving,
- 37:33I'm moving, and I'm talking the middle clip is a normal render
- 37:37after we've created the Alien character with Flow Studio.
- 37:40And the one on the very right, you can already see it in the
- 37:44quality of the lighting and the reflections.
- 37:47It's the one with the neural layer.
- 37:49Let's take a look. What is up everybody.
- 37:52I am the HURL. You can even see the mouth
- 37:55moving in the neural layer render the.
- 37:58Best work that we've And the eyes.
- 37:59Look at the eyebrows. And the color.
- 38:01The color itself matches the environment much better.
- 38:04I walk back. I want you to.
- 38:06Walk with me. So let's go.
- 38:07This is really cool. Is the audio playing for the
- 38:10guys? Pratt Yeah, it is.
- 38:11OK, because I didn't hear it. I just want to make sure people
- 38:13heard it. Farrah, What do you think about
- 38:17the difference? As you said, I think people need
- 38:20to decide when they, you know, do their final render.
- 38:23Do they want to take their own render or they want to take it
- 38:25to a different 3D software or they want to use the neural
- 38:28layer. In this case, it helps with the
- 38:31colors of our alien. Of course, it helps with the
- 38:34facial feature. So the neural layer is much
- 38:36better. Take care.
- 38:38It depends, right? Because we're using AI to
- 38:40enhance the frames depending on how the AI model was trained at
- 38:44times. In this case, it understands the
- 38:46character very well and frame by frame it actually generate
- 38:49something more effective for the user.
- 38:53But there are times that it will not work.
- 38:54Again, just like image generation models like GPT image
- 38:57show point. Or sometimes it will work the
- 38:58right way, sometimes it won't, which is why Autodesk Flow
- 39:01Studio has provided the option. When you're about to use a
- 39:04character in a live action plate, you have the option to
- 39:07turn on or off the neural layer rendering.
- 39:10In this case it worked really well, which is why we wanted to
- 39:12show it to you guys so you know the difference between the
- 39:14normal rendering and a full rendering with neural layer.
- 39:18I do wanna say one thing which is perhaps the most interesting
- 39:22thing to me. We went from the video on the
- 39:25left to the video on the right in under one hour.
- 39:28From the idea of the character to creating the character,
- 39:32rigging it and putting it on a live action plate with the
- 39:35lighting and everything added. If this was pre production for
- 39:39the ultimate Unreal Engine program and we could do it with
- 39:42the neural layer rigging at the time and the AI rigging, it
- 39:45would have saved us so much more time and it would look so much
- 39:48better for storyboarding. Imagine you were pitching to a
- 39:50client. Imagine there was a commercial
- 39:53that you were pitching and you instead of showing storyboards
- 39:55of photos, you could actually put the character in, put the
- 39:58mascot in and show them a video like that.
- 40:00And with neural layer you can show them a higher quality
- 40:04video. And remember you showed us a
- 40:06video of Master Beast team using.
- 40:09Yes. So there were Beast games and
- 40:11these shows. Yes, yes, yes.
- 40:13I just wanna show you guys where we can use a pipeline like that
- 40:17because this is just video output.
- 40:20You have the option to export out the AI motion capture.
- 40:24You have the option to export out the character.
- 40:27You have the option to export out the camera track and so on.
- 40:31Everything is available for you to export out depending on what
- 40:35you're using this for. We have an example for you,
- 40:37Pratt, if you can pull that up. We had created a Viking
- 40:41character. Then we took a dancing video
- 40:43from online and we were amazed by the motion capture.
- 40:48This Viking was a 3D asset or? We created it with Flow.
- 40:52Text to image, image to tree? Yes, image to 3D text all the
- 40:55way takes to image to tree. So the entire pipeline.
- 40:59What's the time? This is the girl doing this
- 41:02turned into a Viking. The first cut is even better.
- 41:07Yeah. Now imagine you export this out
- 41:09to Blender. Maybe you want to take this
- 41:11Viking character, put them in a completely different environment
- 41:13with the dancing video. Let me just quickly.
- 41:18Go ahead and rearrange it. No problem.
- 41:21Perfect. Now play that.
- 41:27This isn't Blender. The Viking character was created
- 41:29with Wonder Flow Studio and the animation was from a video.
- 41:35Again, we went from idea to a Blender output with the
- 41:39animation all in under one hour. We recommend for anyone who is
- 41:46either an indie game developer or a visual effects artist or a
- 41:503D printing hobbyist to go out and give this a try.
- 41:54What's amazing is that in the free plan, you do actually get
- 41:56enough credits to go out and try out the pipeline for yourself.
- 42:00We do get a lot of credits, of course, because we have worked
- 42:03with Flow Studio before. They actually gave us the
- 42:06privilege to go out and try out all of their tools, and what you
- 42:10can be sure of is when they release new features, whenever
- 42:14there's experimental updates, we're always going to keep you
- 42:17guys updated right here, especially for 3D artists who
- 42:19are excited about technologies like this.
- 42:22Are we ready for the next part of?
- 42:23Course we are. So we discussed about how you
- 42:27can now create high quality motion design using AI.
- 42:32In the previous years, if you wanted to create motion graphics
- 42:35or motion design and you didn't have the skills, you will either
- 42:39have to go and buy template or you have to go and watch YouTube
- 42:42videos, go to Premiere Pro and After Effect and create those
- 42:45motion graphics. After Effect and Premiere Pro
- 42:48work as a interface that translate the visual graphics of
- 42:53the motion design into code. So for example, you wanted to
- 42:56move the caption from one point to another.
- 42:59Basically it would just change the coordinates at the back end
- 43:02and the code would do everything for you and you would just play
- 43:06with the visual interface. Now with clod and codecs which
- 43:11they are great at coding you can now describe what you want and
- 43:16the motion graphics would be created right after.
- 43:19So there are two different path that you can take 1 is Hyper
- 43:23frames by Hagen which is an open source video creation model
- 43:28using HTML and they have already plug insurance into clod design
- 43:33and codecs. One thing also is very
- 43:35interesting that when they launch in a few days, you have
- 43:39plug insurance everywhere. So now you can use Hyper Frame
- 43:42in codecs. There's an official plug in and
- 43:44there is a Hyper Frame plug in in cloud design too.
- 43:48The other method is Remotion. Remotion is also using code
- 43:52again, but this time using React to generate the videos.
- 43:56You can see this is their demo for Remotion.
- 43:59I mean, their demo is quite good.
- 44:00In order to get this kind of result, you need a very, very
- 44:03huge prompt. But what we did, we wanted to
- 44:06test these two different method out.
- 44:09So we gave a similar prompt. We said take that decision
- 44:13studio, take their Instagram followers, take their YouTube
- 44:17followers, the number of followers and put them in the
- 44:19graph. We used hyper motion in codecs
- 44:23and sorry, hyper frame in codecs, and we also used
- 44:26Remotion Pratt. Can we look at hyper frames
- 44:29first? This is hyper frames, right?
- 44:31Yes. So we said yes, the numbers
- 44:33should count up. So imagine if this was a
- 44:37presentation. Yes, show them in the graph and
- 44:42draw the graph for each social media platform.
- 44:45This was done just by a few prompts.
- 44:48So we created the master prompt. There was a few back and forth.
- 44:51Faras had a few comments. He wanted to add glow.
- 44:54He wanted to change the graph type.
- 44:56But you can imagine for someone who doesn't know how to make
- 44:59motion graphics, this is a very good first step and there is a
- 45:03huge difference. Imagine I'm giving a
- 45:06presentation. And you could better get
- 45:07PowerPoint presentation, yeah. And I wanna talk about bad
- 45:10decisions. Instead of a static fucking
- 45:12presentation, I can go out and present this and it's so much
- 45:16cooler than what I would make with PowerPoint.
- 45:19And this was done with a few prompts, as Farat said.
- 45:21And I just went on top of Farat's computer and I was like,
- 45:23can you add some glow to the graph?
- 45:25And also have the. Numbers have the numbers
- 45:27increase up to what we have instead of it just coming as
- 45:30static numbers and with one or two prompts you manage to.
- 45:33Exactly, and because this is a rendered MP4 but the backbone is
- 45:40HTML, it's code, so it's highly highly editable and
- 45:44controllable. Imagine if you want to do the
- 45:46same thing with Kling or others, the numbers, you would, yes,
- 45:50that would work. But here, pure code, pure graph,
- 45:54everything is mathematically right.
- 45:56When you see the two graphs between side by side, their
- 46:00differences is mathematically right because it's code.
- 46:03Let's look at Remotion. Remotion is a little bit more
- 46:06cinematic in my opinion because using React.
- 46:08So this is the remotion render. I actually preferred the high
- 46:14promotion. Really.
- 46:15Yeah, but I think you prompted the high promotion.
- 46:17More same prompt, but yes the back and forth, the high
- 46:21promotion was more. Yeah, I think you didn't do it
- 46:23for this. Yes, that's why I prefer that
- 46:25one. What I think I understand from
- 46:28these two is that they both work very closely to each other.
- 46:33Yes, I don't think there's a massive difference, but you did
- 46:35say there's a difference in the output from HTML to React.
- 46:39So no, no, no that basically the backbone of each is different.
- 46:43Remotion is using react, Hyperframe is using HTML.
- 46:47What is the export? Options.
- 46:48Both MP4. Both MP4.
- 46:50Yeah, yeah. Not just MP4.
- 46:52You can get the HTML as well, but for people who wants to use
- 46:56it in their videos, they can get an MP4 output.
- 46:59Because what we saw is that people are doing their launch
- 47:02products. Oh yeah, on X, for instance, for
- 47:06a new application that they've made using these tools, because
- 47:10these guys are either developers or not even developers.
- 47:12They're vibe coding apps and they need videos to promote
- 47:16these apps, but they don't know how to make these videos.
- 47:18So they've been going to these different platforms and using
- 47:21these tools to create their motion graphics.
- 47:24We were skeptical in the beginning because when we heard
- 47:27about these tools, we thought they're not actually that great.
- 47:30And I'm not saying they're going to be as great as a highly
- 47:35skilled motion graphics artist, definitely not at that point
- 47:38right now. But considering the speed,
- 47:42Farah, how long did it take for you to do this?
- 47:4410-15 minutes. OK.
- 47:46And. Look, and without, yeah, without
- 47:48any skill, I didn't even open a video editing software.
- 47:53And why I think this is also important is that a lot of
- 47:57people who are using this are people with technical
- 48:00backgrounds. They are not creators.
- 48:03So if you go, of course, in our space, everybody knows how to
- 48:05use Cap cut. Everybody knows how to use Da
- 48:07Vinci or Premiere Pro or After Effects.
- 48:10People who code actually don't have that background.
- 48:13And creating beautiful presentations is I think is
- 48:16really a plus. I've been in the corporate
- 48:19world. The PowerPoints are really ugly.
- 48:21Really, really ugly. You know what I'm thinking
- 48:23about. I personally believe first of
- 48:27all, the hyper motion looks much better.
- 48:30Hyper frame. Hyper frame looks much better.
- 48:33It is perhaps because you went back and forth and added certain
- 48:38things like the glow. This is where I think you should
- 48:42pay attention to these tools and start using them now because I
- 48:45feel like the difference between hyper frame and where GPT image
- 48:492 point O is right now is that GPT image 2 point O has already
- 48:53been trained on all of the things that is missing here.
- 48:56For instance, farrots first generation didn't have the
- 48:59numbers increasing and didn't have the glow and the fonts
- 49:02didn't look as good. He had to tell it to make it
- 49:05look better. Now if you are not from a
- 49:07creative background, you don't have the knowledge to tell
- 49:11hyperframe to make it look better because you're just going
- 49:13to tell it make it look better. It doesn't know how and it will
- 49:16never get there. So you need to have that
- 49:19knowledge. You tell it hey add a glow, add
- 49:21some grain, change the background from a static image
- 49:25or gradient to a gradient that is dynamic and moving.
- 49:29But I generally believe within six months, either these guys
- 49:35are going to update their plugins or somebody else is
- 49:38going to come along with the capabilities of GPT image 2.0.
- 49:42Again, GPT image 2.0 right now, we didn't even tell it to add
- 49:45grain or fix the colors. We just say copy this one image
- 49:49style or these two references. Imagine if one day you can do
- 49:53the same here. I 100% agree.
- 49:55With you to that point, right? Yes, and it's funny you say
- 49:58that. I've heard the first time that
- 50:01AI can create motion design. Probably it was four months ago
- 50:05and we looked at it again like most of the news that we talked
- 50:08about it was pure shit. It couldn't generate anything.
- 50:12You tried it also on Hicksville, it couldn't generate anything a
- 50:15few months ago. This is the first time that the
- 50:18result that is coming out is actually good.
- 50:22It can get better of. Course, so there's a lot of room
- 50:25for. Improvement, but it's something
- 50:27that I. Would use this for my slides.
- 50:30This is better looking than my slides.
- 50:31No. Previously, when we prompted to
- 50:34generate motion design for, it was not usable.
- 50:36No, it wasn't. Usable.
- 50:37The things were going over each other.
- 50:39It didn't even make sense. So now this is OK.
- 50:42This is the animated PowerPoint presentation, yes.
- 50:45And by the way, Pratt has done that using Clod for our UI
- 50:52without using any of these skills.
- 50:53Yeah, so he didn't use hyper frames, he didn't use emotion,
- 50:56right. You didn't use any of them.
- 50:58None of those directly inside Clod.
- 50:59Directly it's like fully HTML, CSS and it looks pretty good
- 51:05like we are. This is like our new UI that I'm
- 51:09working on and this is just like two paragraphs of prompts and
- 51:14like minor tweaking. I haven't worked on it a lot,
- 51:17but it looks pretty cool, you know?
- 51:18So everything we're looking at right now was created in
- 51:21conversation between you and Claude.
- 51:24Yes, Claude Design or. Claude chat.
- 51:25No, this is Claude chat. Claude Chat.
- 51:28For anybody that does not know Claude Design, this is their new
- 51:31research lab. It was released when we were on
- 51:35on our off break last week, that's why we didn't cover it.
- 51:38But Claude Design is specifically made for creating
- 51:41design systems. We've tested it, but Pratt's
- 51:44been rigorously on this and his response has been that clot chat
- 51:48is still better than clot design.
- 51:49The reason why is because clot chat, it remembers whatever
- 51:53you're talking to it, right? Yeah, similar to ChatGPT.
- 51:56'S image 2.0 yeah. It's like it remembers, oh, this
- 51:59is what he asked for, this is what he wants.
- 52:01Meanwhile, on clot code, the whole code becomes too extensive
- 52:05and now it has to go back and understand what's going on.
- 52:08So it's a bit, but in terms of design, even now the UI we have
- 52:12on the show, everything is designed by cloud.
- 52:14Yeah, it's HTML, right? Yeah, everything is HTML.
- 52:17It's like 6000 lines of code and it works flawless.
- 52:20What? Again?
- 52:21What? 6000 lines of code. 6000 lines
- 52:24of code animating everything you're seeing on the screen
- 52:27right now, from the animated dots behind me and Farah to this
- 52:32glow effect at the bottom and to all of the visuals that Pratt is
- 52:35pulling here and there I. Mean.
- 52:37The reason we can do it is because it is mathematically
- 52:41possible to write all of them and you can edit.
- 52:43Them yeah, you can edit them. You can Pratt change the font,
- 52:45you can change the colors. I can change anything and
- 52:48everything like. I say website, right?
- 52:49I decide. To prompt it once, make it
- 52:51invert the colors, make it white theme, Yeah.
- 52:53And you know what's the best part, Pratt?
- 52:55You can also create sliders and dials for yourself.
- 52:58So let's say you created a design, let's say the glow
- 53:01effect. If you wanted to control the
- 53:03glow, you don't just ask it to edit it.
- 53:05You can ask it to create a dial where you can increase or
- 53:08decrease the intensity of the glow.
- 53:10Right? Yeah.
- 53:10I can do that. This is where things get
- 53:14incredibly exciting for me. I hope everybody watching and
- 53:18listening right now can pay close attention to what I'm
- 53:20about to say. There is no longer a barrier to
- 53:24entry for great design. There's no longer a barrier to
- 53:28entry for great software. Both of those have become so
- 53:32democratized that if you truly want to create something, you
- 53:37actually can't for the first time ever.
- 53:39I remember when I first started making YouTube videos as a
- 53:42gamer, I had to spend. I think at the time I basically
- 53:47had no money because I hadn't even started working $50.00 to
- 53:51hire someone to make a thumbnail for me.
- 53:53And I was stuck with that thumbnail or that cover for my
- 53:55YouTube banner forever. And if I didn't like it, that
- 53:58was it. But now if I was back in my 15
- 54:03years old Pharaohs, I could create 60 variations of the
- 54:09thumbnails that I want and it can look freaking amazing for my
- 54:14OBS overlay. I was actually.
- 54:15No one knows this but me and Pratt were streaming video games
- 54:19like 10 years ago. Pratt had to work.
- 54:21How long did you work on the stream overlays?
- 54:23It was insane. How long?
- 54:25I think I'm one or two days easily because you have to make
- 54:28everything in Photoshop, take it to After Effects now, animate
- 54:32everything, mask everything, and now you have to export it out.
- 54:35And now the problem is I exported a MOV and it's like a
- 54:3810 GB file and I give it to you and it's not working now.
- 54:41How do I optimize it? And there's like a lot of back
- 54:43and forth now it's HTML. I just had to give it a command,
- 54:46make it transparent. It's too heavy.
- 54:48Make it light. Why is it heavy?
- 54:50I just put the link inside of OBS and we're good to go.
- 54:53And this is 10 times more dynamic than the stuff that we
- 54:56were making before. Absolutely.
- 54:59This is the crazy part. I just wanted to tell you guys
- 55:02if you're making anything and again, anything, I'm not talking
- 55:06about the specific thing. Before app, software, website,
- 55:09whatever. Short film games.
- 55:11Games. Whatever you're making, you're
- 55:13going to need to create something, right?
- 55:15A visual in photo form, in video form.
- 55:19An overlay like this for a show, for a stream, an application.
- 55:23UIUX product shoots, product photography.
- 55:26This includes pretty much almost every person I know in real
- 55:29life. Unless you're from a brick and
- 55:31mortar business, even that will be no.
- 55:35You need to use this by the way. Right.
- 55:36You need presentations. That's the first step.
- 55:39You need website for your own business or for clients that
- 55:43you're working with. You need it.
- 55:44You definitely need it. Everything.
- 55:45We talked about today has to sound exciting for you if you
- 55:49truly want to create, if you truly want to create the
- 55:53barriers that have never been lower, we can create
- 55:56applications, we can create software, we can create
- 55:58websites, we can create stream. Overlays in it insane.
- 56:02It's really like even right now OBS, right?
- 56:04The platform, the software we are using to stream, there are
- 56:07limitations to it and because it's open source, I can just go
- 56:11to cloud and be like, oh, change this.
- 56:14As far as knows this, because he's here, he has seen my UI and
- 56:17everything. It's a completely different
- 56:19software. It doesn't look like OBS
- 56:20anymore, you know, and it's insane.
- 56:22Even remotion, the presentation about bad decisions.
- 56:26If I'm editing a video, let's say I'm making a documentary
- 56:28about them, perfect visuals. I don't have to sit down and or
- 56:32make a graph out there out this many followers and all these
- 56:34things in 5 minutes. If I do the same thing in After
- 56:37Effects it will take me two 3-4 hours easily.
- 56:39I'm just amazed at what's happening at the rate of
- 56:42progress. I'm incredibly excited about all
- 56:45of this. We've been using it rigorously
- 56:48within bad decisions and you guys are seeing the results.
- 56:50Again, everything that we do is enabled with technology and we
- 56:54will continue to update you guys on exactly what these tools are
- 56:59right here on the show. Now, I do want to say one thing
- 57:02if you guys are incredibly curious about how to use these
- 57:07AI tools, because here on the show, of course, we're talking
- 57:10about the updates. We tried to give you guys a live
- 57:12demo, but there's only so much we can say or show in one hour
- 57:18covering multiple topics for people who want to learn AI and
- 57:23all of the tools that we just discussed, right?
- 57:25Go deeper into creating websites.
- 57:28How do we create design systems? How do we create branding?
- 57:31How do we create photo imagery? How do we create videos with all
- 57:36of the practical use cases that we use ourselves?
- 57:39We are working on a platform specifically for you.
- 57:43This is what we've been working on for now the past three
- 57:48months. Three months, I would say.
- 57:49And it is under one link called AI dot bad decisions dot studio.
- 57:55We're going to constantly. Is this the one that is updated
- 57:58practically down? Yes.
- 58:00So before you go down, I just wanted to ask.
- 58:02That no, the chapters are not there.
- 58:04Fuck, chapters are not there. Show that to them.
- 58:07OK, so here you're going to go ahead and drop in your name and
- 58:10your e-mail, and that way you're going to join the wait list.
- 58:14Once you join the wait list, you're going to get in as an
- 58:16early bird with extra privileges, which is why we urge
- 58:19you guys, the guys who are watching this show, to get in
- 58:21early when this launches, the people who are in this program
- 58:27not only are gonna be updated from this show, they're gonna
- 58:29actually learn out of the tools. So if you're using it for
- 58:32yourself like we are, or you're gonna use it for your clients,
- 58:35you're gonna stay ahead of everybody else because you're
- 58:38gonna learn the best tools. For anybody who's curious, this
- 58:41is where I'd recommend that you go whenever we're going to have
- 58:44the chapters and I want to show it to.
- 58:45Them Friday by Friday, yes Pratt by Friday, right by Friday.
- 58:49When we come back for the show, when we ask you guys to sign up,
- 58:53we will Scroll down and you would see the 1st 7 chapters of
- 58:58our foundation section. I'm not going to say anything
- 59:00more. Yes, you're going to see it on
- 59:01Friday. We actually uploaded a story on
- 59:03Instagram. If you do not follow us on
- 59:05Instagram, please do that already.
- 59:07We update. The behind the scenes, yeah,
- 59:09the. Fastest on Instagram because we
- 59:11take a lot of stories. We told you guys exactly what
- 59:15we're working on. So in the past week, we not only
- 59:18recorded all of our foundational chapters, which are theory
- 59:22based, teaches you everything you need to know about AI.
- 59:25Because we understand that many other programs out there may
- 59:29just be teaching you the tools. But it's kind of stupid if
- 59:32you're offering services or working with AI, but don't
- 59:35actually understand how AI works.
- 59:36Which is why we told you guys, this is going to be the one
- 59:39program you need to learn everything about AI.
- 59:41And the Foundation series section section.
- 59:46We have created about 7 to 8 videos.
- 59:49We were going through 3 or 4 back and forth iterations with
- 59:54our editors with hundreds of comments just to make sure.
- 59:56Every chapter is incredibly and highly produced with the best
- 1:00:02quality of video, best quality of audio, best quality sound
- 1:00:06effects and visuals, easy to understand and fun.
- 1:00:09And we're going to do that for everything.
- 1:00:11I going to close my mouth. Not anymore.
- 1:00:14No, no, Friday, we just show you guys the chapter thumbnails and
- 1:00:17then we're going to see it, yes. Also one thing before you go, if
- 1:00:20you haven't subscribed to YouTube, please do that.
- 1:00:24And Spotify, follow us on Spotify because Spotify has sent
- 1:00:28us a message and said for the shows similar size to your show,
- 1:00:34people follow them more. Did you understand what I just
- 1:00:37said? I did, yes.
- 1:00:37But that's not the best. You should offer someone a
- 1:00:39cookie or something. What's that, dear?
- 1:00:41That's how you entice me. That's not how you entice them.
- 1:00:44You want to entice, I give you a cookie.
- 1:00:46I'll give you a cookie. You give us a five star review.
- 1:00:48No, I never asked for a five star review.
- 1:00:50What did you ask for? I follow.
- 1:00:52I thought you just said Oh no review.
- 1:00:53OK, I'll give you a cookie for the review.
- 1:00:56Just a follow for me and a butter.
- 1:00:57Chicken. See you guys on Friday.