I Got Early Access to Runway 4.5 + Kling AI Demo

Hey, welcome to the Next Wave podcast. I'm Matt Wolf, and once again, I'm here with Maria Garib, and we're going to talk about some of the coolest things that are happening in the AI world right now. And at this current moment in time, the things that are happening in the AI world seem to all be mostly around AI video. We're getting a lot of cool AI video models from companies like Runway from Cling. There was even a little tease about a new, like, Apple video model that they teased. We don't actually have any footage of that, so there's nothing to show there. But Apple's been teasing a video model. So in this episode, we're going to dive in. We're going to test some of these AI video models. One of these models I even got early access to. So we're testing it. We're going to be seeing what the model creates for us with these prompts for the very first time on this recording, using Runway. So going to be a fun episode. Thanks again, Maria, for hanging out with me on this one. It's going to be interesting. We're going to put, you know, these AI video models through their motions here.

Using only 20% of your business data is like dating someone who only texts emojis. First of all, that's annoying, and second, you're missing a lot of context, but that's how most businesses operate today, using only 20% of their data. Unless you have HubSpot, where all the emails, call logs, and chat messages turn into insights to grow your business, because all that data makes all the difference. Learn more@HubSpot.com.

Well, yeah, thanks for having me again, honestly. I'm having so much fun on these episodes, honestly, and I've been getting really good feedback so far. So, yeah, let's go into what Runway has for us today. You did say you have access to something people don't have access to, so I really want to see that, like, now.

Yeah, yeah. So here's the thing. I have access to it right now as of this recording, but by the time this recording's out, I'm pretty sure everybody's gonna have access to it. So what we're showing off in this video, you should be able to do by the time this video is out. But right now, at the time of this recording, this is brand new, fresh stuff. We haven't even seen what this model can generate yet.

Yeah, yeah, but we really want to see, like, what's new with it, because we did cover it last week. But, like, I want to know what's new. What do they have up their sleeves right now?

All right, so this is the Runway ML back end, and they just released Runway Gen 4.5. In fact, as of this recording, it's not even released yet. And I have no idea what to expect.

So we're doing it from scratch right now with you.

This is from scratch. This is our first time seeing it. I have seen some of the, like, demos and stuff that we're circulating on X. How. However, those are almost always cherry picked. Right. Like, if a company is putting up a demo video of, like, here's what our model can generate. They're going to take the best of the best generations they've made and share those. So we're going to test some real stuff and find out if it really is good. You know, one of the things about being a content creator that also shares some of these video models is there's a lot of sitting around and waiting. Right. Like, you enter a prompt and then you sit there for anywhere from two to five minutes and wait for your video.

Yeah.

So I want to get a few different props started. So we have some stuff going here.

Yeah, I don't know. Something came up. I had, like, a meeting with, like, a coworker today, and she was talking about, like, her creating, like, an AI sort of thing that can help people with meal prepping and stuff. Let's say someone has, like, a company of meal preppers and like, what if they were in the marketing team and they don't have the budget to have all these actors and they want to create, like an AI video? What would be the product of that? You know, using Runway, like a meal prepping ad, let's say.

So what should I enter for the prompt then? Here, you want me to try something, like, very generic, like an ad for a meal prepping business and see what it does for us?

Yes. And then if not, let's just do the extra work with all these specific things.

All right, let's see what it does. If we just give it the generic plan of, like, create an ad for a meal prepping business. All right. I'm going to leave it at five seconds just to make sure we're getting faster generations here. And I'll kind of keep an eye on how long that takes, but we can get another prompt queued up and going as well. I always test this prompt, A monkey on roller skates. I don't know why. That's like my benchmark. I tested a monkey on roller skates back when Model Scope came out in, I believe it was beginning of 2023, 3 or end of 2022. There's this model called Model Scope and it was one of the very first AI video generation models we've ever seen. And the very first thing that popped into my head like two and a half years ago when that model came out was a monkey on roller skates. So that was the first thing I ever tested. And it just looked like a little blur. If you squinted and like got real close to your computer, maybe you could kind of see a monkey out of it. But it was very, very primitive. But ever since, I've always kind of used that as like my test. Like, how far have we come generating monkeys on roller skates?

We've come so far. Honestly, I don't know what we're going to come up with. Maybe even have not an avatar, but like a hologram.

Oh, a hologram. Yeah, yeah.

Can you imagine hologram in your whole house or like having like your favorite celebrity in your house talking to you, like giving you pep talk or something?

Yeah, yeah, yeah.

Because you AI generated it. That would be so funny.

So we're 61% here.

It's pretty fast on that.

All right, so what's another prompt we can get going? I don't know how many it's going to let me do simultaneously, but let's keep going.

Let's see. You know the dinosaur that has like mittens? I've seen this picture before. It makes me so funny.

Oh, like a T Rex.

Yes, that would be so cool. Please do that.

A T. Rex.

Making brownies.

Wearing mittens. Making brownies on Mars? Why not?

I saw someone literally just take a picture from midjourney because people on the Discover, you know, midjourney keeps the results of a lot of people and like he put it on click and was like in image to video and it basically came to. That was so cool.

Yeah, we should test that as well because we can start with an image here. So that would be something we should definitely test. Yeah, we did get this back. Create an ad for a meal prepping business. Let's see what it did. Oh, what is going on in that shot when she's trying to eat. This is flapping around.

So funny.

Okay, okay. So I mean, like, what are your initial impressions? Let's just freeze frame on something here.

I think the more detailed you are with your prompts, the better the results, in my opinion.

I agree, I agree.

I think it looks a bit weird, obviously, but I think the more that you give It. The more that you feel it, because I don't know what she's eating. What is she eating, though?

Yeah. I don't know. It actually looks like she's spitting something out in the beginning.

Yeah.

Where did this tray just come from? It just like appeared out of nowhere.

She generated.

So there's definitely some, like, physics issues going on. But, I mean, the person looks pretty realistic, especially in this shot.

The kitchen looks pretty good in the.

Beginning with the food falling out. It was weird, but yeah. I don't know.

That was funny.

That was funny. We also didn't give the prompt much to work with.

No, it was just like one five word sentence.

Yeah.

Yeah.

All right, let's see. So now we've got our monkey on roller skates here.

And I mean, that's cute.

That's pretty good, actually.

I would play it in the background.

Yeah, that's like a nice screensaver.

That is pretty cute, honestly.

Yeah. This has come a long way, no question. Right? You send that to somebody that's a monkey on roller skates, you're not going to question what you just saw.

No, but it's pretty cute, actually.

And then. All right, here's our T Rex wearing mittens, making brownies on Mars. So one thing to notice, and this is something I always look for when I'm looking at new video or image models, is when you stuff a bunch of things into the prompt. Did it get all the things right? And we have a T. Rex. We got some mittens. They're kind of weird mittens, but we got some mittens and they're definitely on Mars.

It's a T. Rex, Matt. It's going to wear weird.

That is true. What kind of mittens would a T. Rex wear? I mean, it's decent, right? Because it got all of the elements. And that's one of the things that a lot of these models struggled with early on, where if you give it like three or four elements that you want them all in the same image, it would really struggle to put all those. You might get one or two, but the other couple things you added to the prompt, it would miss. These models have gotten so, so, so much better at making sure everything you plugged into the prompt actually makes it into the video.

Yeah, that's pretty cool. Honestly, Runway have come so far.

All right, so for the next test, let's go to ChatGPT here and have ChatGPT generate a prompt. And the reason I like doing this is I tend to come up with fairly vague prompts where ChatGPT will sort of flesh it out for us. Right. It'll add more details. It'll add scene details and color details and lighting and maybe even, like, what camera it was. Sh. ChatGPT helps with a lot of that kind of stuff. To give you a more, like, fleshed out prompt where you could just start from, like, a general concept. What's another concept?

Let's give it a role. Like, let's sell it like photographer, like something. And like, tell it like you are this. And I need to create an ad for this company, for example.

All right, you're a world class cinematographer. And dp, generate a video prompt for me. Include scene details, camera, camera angle, and any other details necessary to get a good shot. Yeah, create the prompt about, let's see, what's a scene. We can try.

Even though I read fantasy, I don't know why the scenes are not. Let's do a dragon, though.

Okay.

Create a prompt about a dragon in a colorful fantasy world. Well, let's try that one next. And we'll test to see if it can do, like, actual ip. All right, so it's giving us a pretty detailed prompt because look at this. It's actually giving us, like, broken out bullets of camera and movement, lighting and atmosphere. World aesthetic. Okay, so this was it, sort of breaking down the scene. And then this is the actual prompt.

And then there's a prompt.

I thought it was gonna have me paste this whole thing in, which it probably would have worked, but I'd prefer to just give it a prompt.

But Runway would break down. Runway would cry.

Runway would run away.

All right, should we make it like more than five seconds?

Well, it'll let me generate up to eight. It gives me five or eight. So I'll generate eight this time. And let's get that one going.

Yeah.

The other thing I want to test is, I want to test starting from an image. So let's go to mid journey.

Mid journey? Yeah. Oh, is that you?

Yeah, that's supposed to be me holding a fire extinguisher and putting out a fire. But it's an extinguisher that's creating a fire instead of putting out a fire.

Setting you on fire.

Yeah, setting me on fire. All right, so let's start with an image prompt. Any ideas come to you for an image that we can start with?

Let's do Hansel and Gretel.

Yeah, I should do it in 16,9. Actually use this prompt again. Let's go 16,9. And I'm going to generate it again. Oh, these are all very, like Cartoony style.

And by the way, you can search in midjourney in the images, like, above of someone that already did it, and it could give you better results.

Oh, yeah, that's a good idea.

Tab above it.

Although some of these are pretty good, so their video generator doesn't give you the option to start with an image if you're using gen 4.5. So we only have text as the starting, it looks like on this model.

Yeah.

Okay, well, that's good to know. So, yeah, as of right now, it doesn't look like we can start from an image. All right, if you want the full breakdown of all these AI video tools, we put together a complete guide between Runway, Cling Veo, and Sora. It includes universal prompts that work across all these platforms, real test results with the same prompts, pricing breakdowns, and a framework on when to use which tool. Get it right now. Click the link in the description. Now let's get back to the show. All right, so here's our cinematic shot of an iridescent dragon.

Let's see. Ooh.

I mean, that's pretty good. I'm not seeing a whole lot wrong with it.

It is pretty good.

It's very consistent. I'm not seeing, like, the dragon change shape. I'm not seeing any of this stuff in the background disappear. I think it's pretty impressive.

That was pretty good.

I'm a little bummed we can't do an image prompt as a start. So that's Runway.

Honestly, it's not that bad. The more that you give it, obviously, the details, as we said, the more it gives you, the better results. The dragon thing is absolutely amazing. I think the more that you write into it and you give it the context of what it is and stuff, like, imagine doing it, like, Ghibli style or anime style, and it just comes out of the style of these animes that you watch and stuff. So, so good. Honestly, so good. They've come so far.

Yeah.

10 times better than VO or SORO. Honestly, when it comes to Nano Banana, though, Nano Banana is still up here. But, yeah, yeah.

My opinion on this. 4.5. Right now, I think it's fairly on par with VO 3.1. I'm not seeing anything about this video model that makes me, like, super, super blown away. Like, it's a huge improvement. Like, it's definitely a huge improvement over what we got a year ago, two years ago. But over the sort of, like, current crop of available models, I'm not really seeing, like, a huge leap between VO 3.1 and this. And, you know, Sora, too, is good as well. But Sora is, like, pretty good at generating people and, like, sort of realistic scenarios.

Yeah, we should give that to Sora. Honestly, I keep bashing Sora every single time I go on this. And I don't apologize, by the way. I don't know if, like, I'm not going to say sorry about that, because it still sometimes sucks.

Sora is probably the most realistic people stuff people think.

Right?

Like, if you're generating videos of people doing something, almost anything else outside of people doing stuff, I don't really feel like it's that good.

And to be fair, you test a lot of these video tools. You know, like, I write a newsletter, and, like, sometimes I don't have the time to do everything. But you're, like, on the Internet 24 7, so, like, you're gonna do it.

Just call me a huge nerd. It's okay. I get it.

Last episode, I did that. But I feel like you'd see the comparison. Like, for someone that doesn't really test them, I think it's pretty good for me. But, like, I think you've seen better. You know, to be fair.

Yeah. I would say it's, like, probably fairly on par with VO 3.1. To me, I'm not, like, blown away, as in, like, this is a huge leap over what we already had. And one of the things that I always like to think about, especially when I'm making videos on my main YouTube channel and is, like, what can we do now that we couldn't do yesterday? And if, like, the news I'm talking about doesn't really have any sort of implication in that world, like, if there's something new that got released and I try it and I go, okay, but I've already been able to do that. This doesn't make it easier. It doesn't make it faster. It doesn't make it better. So what is the point of this announcement? And if I feel that way about something, I usually just don't even talk about it in my videos because it's solving a problem that's been solved by another product already that I've probably already talked about.

Someone did it before. Yeah.

And so with Runway 4.5, I think my sort of initial gut reaction to it is, like, it's pretty much doing the same thing we've already been getting from something like VO 3.1 or something like, the newest models of cling. It feels pretty on par with those. So I don't really Feel like this is any sort of huge leap saying that VO 3.1 is very, very expensive to use. I think Runway is more economical. I don't have the pricing in front of me, but I think it's a little bit less expensive to use.

Yeah.

To do a lot of generations with VO.

How much is VO?

I think on like their $20 a month plan, you can generate something like 10 images a month or something like that. On their Ultra plan, I don't know if there's a limit of how much you can generate, but it costs $250 a month.

Oh my God.

It costs a car payment to actually have an Ultra plan from Google. It's pretty crazy. So most people who use VEO don't actually use it straight through Google. They'll go to a tool like Leonardo or crea. I think Higs Field has it in it. There's all these other platforms that use the API and baked it into their product and you can actually use it a lot cheaper if you don't use it directly through Google for whatever reason.

Yeah.

But anyway, I feel like this model is pretty on par with what we've gotten already, but it might be a more economical way to. To do what we've already gotten.

Yeah.

If something is just as good as what we've already got, but you can do it faster and cheaper, that's still a plus for people. And I feel like this might do it slightly faster and slightly cheaper than vo. What would I say it's a giant leap over vo? Not really. That's kind of my take.

I'm pretty sure we're going to see everything flooded on TikTok right now because TikTok is like the, the. I think the age range is a bit on the, you know, Gen Z side and like a bit of Millennials and a lot of Gen Alpha. So I don't know if like all of them have the amount of money that people can buy their memberships into veo. I think they're gonna go with like Runway for now just to be able to get these high end videos because they blow me away. And this generation is so creative. Extremely creative. I always joke about this and a lot of people on the Internet joke about we are definitely gonna get scammed eventually by something because you wouldn't know the difference. But it's pretty good. It's pretty good. These generators. I think we're gonna see a lot of it flooded on TikTok.

Oh yeah.

But I want to see on like cling. Do you think like we can do a comparison of what Runway can do. Like what Cling can do.

Yeah. So a couple weeks ago, Cling did like five days of announcements where every day for five days they rolled out like a new Cling model. One of the models was called cling01. Cling01. They're claiming it's like the nano banana of video. Right. So like, you can give it an initial video and tell it to edit anything about the video. And it will edit that stuff. Right. You can reskin it with a different design, take a real life video, turn it into a Studio Ghibli style. It'll sort of re skin the video in Studio Ghibli. Or you can have, you know, somebody walking through the shot and then say, hey, remove the people in the background and it'll remove those in the video. Or add Voldemort in the background and it'll add him walking up to you or whatever. Right? So it's like they're claiming it's like the Nano Banana video and that you can give it a video image or text like any of these things, and it will try to combine them and edit them together and create a video out of it. But there was also Kling image model, which was like identical to nanobanana. It created images, but you can edit images with it. Pretty much anything you can do with Nanobanana you can do with the cling01 model. But it was an image model, not a video model. Then they released Kling Video 2.6. Kling Video 2.6 is kind of like their version of Veo. It will generate audio. The one we were just looking at, Runway ML, was not generating audio on the video. This new cling 2.6 generates video or audio into the video, just like VO and Sora do. They also released an avatar model, which I've never played with. Creating AI avatars isn't really something that I have a ton of interest in, so I haven't touched that one yet. But I have played around a little bit with video 2.6 and cling01, the nano banana of video. So those are the ones I have played with. What I can do is I can start by showing some of the stuff I already generated while we're waiting for some of our new generations to happen.

Yeah, and let's also run what GPT wrote for like the dragon thing and like put it here just so it compares to see what comes up with.

Yeah. So this is the video 2.6. This is like their version of Runway 4.5 VO 3.1 Sora 2. Like, this is their top of the line video generator here. So I'm gonna grab our prompt from Runway, paste it back here. So this is the same dragon shot that I'm submitting. This one can do five or 10 seconds. I'll do 10 seconds on this one and we'll go ahead and generate our dragon. We can throw in our T. Rex wearing mittens, making brownies on Mars.

I played around today because, like, I was really curious about, like, what it can do and stuff. And I was like creating content for like my page. And I don't know if you've seen that movie. Death becomes her Mara Streep. And like.

Oh, yeah, yeah, it was a long time ago. That's like an older movie, right?

There's an amulet that basically gives you immortality and stuff. And like it has some kind of potion. And so I took that picture and I put it in cling and I said the amulet is sort of like glowing or like, kind of like radiating. And the results were pretty cute. Like, I'll send it to you.

Oh, cool. Actually, let's throw in our image to video too, because we did generate an image inside of Mid Journey. We didn't get to play with that one. So I'll use that as our starting frame. We'll pull in our Hansel and Gretel image here. All right, let's generate that one. Okay, so we've got those generating. While those are generating, I can kind of show you some of the other stuff that I did.

Yes, please.

So the first thing I tested was I wanted to test like the lip syncing. So I made this video of a wrapper wrapping into the camera because this one does come with audio light, lights, flash, systems crash moving in in a digital hot dash. So, you know, the audio wasn't great. He was supposed to say, lights, flash, systems crash, moving in a digital dash. That's what he was supposed to say. But he kind of goes, lights, lights, flash, flash. Right? And he's just kind of like repeating himself. So the words I asked him to say didn't quite come out correctly. But I mean, what are your thoughts on the video?

I mean, the hand stuff are still like a bit iffy, but like, yeah, it feels like he broke seven phalanges and stuff.

But yeah, yeah, his fingers look totally jacked right there. Or.

Or he has like, he's double jointed, like, you know, like. Like me because I'm double jointed as well. I could be that. But yeah, I mean, it looks Good. I think if, like, you gave it more details. Yeah. The speech thing was not on point, but it's pretty good.

Right.

I don't think others could do the same.

Yeah. I mean, vo, I think does that better, honestly. Vo 3.1. You're gonna get speech.

Yeah, but, like, who has 200?

Yeah. And okay. I mean, Sora does it pretty well too. Like, Sora's lip syncing.

I have a beef with Sora. I don't want to talk about her today.

So the next thing I wanted to test here was sort of like, sound generation for, like, does it generate somebody walking through leaves? So I basically said, have them walk on leaves. Have them walk on wood. Have them walk on concrete.

The obsession of men.

And I wanted to see if the sound sort of changed based on what they were.

Is it obsession? Men in general feel like this obsession to basically pick up ice or something that's frozen and I throw it away, as well as walk on dead leaves. I've seen it all over the world. It's universal. There's videos online of men basically picking up the crunchiest sleeve and just stepping. It's a men thing.

That's funny. I've never heard of that trend. I didn't know it was a thing.

It's a thing.

But here's what this video came out like. I mean, it's okay. To me. It doesn't really sound like crunching leaves. It just kind of sounds like. Like, I don't know.

Yeah. Someone opening up, like a chip bag or something. It's fine.

Yeah, it's. It's okay. Like, I'm not, like, blown away, but it's, you know, it's decent. So then I wanted to do one to see if, like, the audio would change drastically where, like, they're in a loud nightclub and then they walk outside, and then it's, like, quiet outside. And this is what I got with that.

Oh, interesting. I mean, yeah. But then it sounds like a horror movie when you come out. Like, you know.

Yeah, it's definitely got some issues. Like, first of all, what kind of nightclub just has, like, a gate to the outside when it's snowing outside, Right? Like, it's not a door. It's like a gate that you can see through, I think.

Yeah. I haven't been to Berlin, but I heard this is how they do their nightclubs, so I'm not gonna assume. I think this is how the nightclubs in Berlin are.

But then, you know, when you get outside, the noise stops. But if you were just behind a Gate. Would the noise actually stop? Cause it's just a gate. Not a gate door. Anyway, that's probably a little too much to ask of the AI Also, why.

Are they jumping so much? Like, what kind of a party is that?

Yeah, yeah, it's just a giant cult gathering. That's all it is.

I didn't want to say it. Oh, are we fighting? What is this?

So this is, like, I was going to test dialogue. I wanted to see if the guy starts talking. If the woman can interrupt. This is. Let's just watch it.

I really think we should consider the Stop.

I don't want to.

She's about to stop him.

She looks like she's about to slap him. The funny thing is, though, like, when she yells stop, he also yells stop at the same time. Watch his face when she yells stop. So, like, they both yell stop together for some reason, which is a little.

Weird imitating her or something, you know?

Like, it's like, I know you're about to yell stop, so I'm just gonna say it before you can. And then I also say, like, the prompt. She's supposed to raise her hand and go, stop, but she yells stop. And then raises her head like she's about to slap him.

She's about to mark his face with, like, five fingers.

And then this one was meant to be like, a little kind of like, ASMR test with a marble rolling down a plastic track. And then it was supposed to hit the dominoes and then smack the balloon and pop the balloon. That's what it was supposed to do. Keyword, supposed to do.

What contraption is that?

I was just wanting to see if it would make the good sound of the ball rolling and the dominoes falling and stuff didn't quite come through.

Oh, what the. What's that take coming out of the balloon?

I don't know. It looks like shaving cream coming out of the balloon or something. I don't know.

I think because it picks up, like, the asmr. That happens, like, on social media, where, like, people fill up balloons with, like, shaving cream. I don't know if you've seen them, but.

Oh, is that a thing people do? I never see those.

It is. That's probably why it's picking up that kind of algorithm and, like, kind of getting, you know, inspired by it, probably.

Yeah. Yeah. But I mean, the sound isn't bad, but the visuals didn't match my prompt at all. Right. The dominoes were supposed to be at the end of the thing, so the ball was supposed to Hit the dominoes. Knock the dominoes down. The dominoes were supposed to bump the balloon. Right. It was supposed to be this chain of events, but it put the dominoes next to the, like, slide that the ball came down and only knocks it over one domino. Then the ball itself hits the balloon, and instead of popping, the balloon just kind of gets a little hole in it, and then shaving cream starts coming out.

That's why. Yes.

So if you're listening to the audio of this podcast and you're like, what's the big deal?

What is happening? Watch it.

You need to come watch this.

Yeah. Just for you to understand what we're seeing right now, you know?

Yeah, exactly. All right, so that was all video 2.6. Let's see if our generations of video 2.6 are done. So we've got our dragon shot that we're comparing against runways.

The Dragon looks better than Runway, honestly.

Let's take a look at the Runway just so we can compare them back to back. So here's what Runway generated. No audio on Runway. So if you're listening on audio, we're watching the dragon video that Runway generated, and it's pretty decent.

Pretty good. It's pretty good.

I like the colors of it. I think the colors are nice.

They're vibrant, which is. You don't see it a lot. Yeah.

All right, so now I'm going to jump back over here to cling 2.6. This is the exact same prompt. I just copied and pasted it.

Wow.

Much better physics. Much better.

Oh, yeah.

There is something wonky with the legs right here in the beginning. I get really nitpicky with these. Most people probably don't care, but if you notice right here, he's got three legs on one side and one leg on the other side.

Maybe he has a deformity, like, we shouldn't judge.

Oh, yeah, I guess. Yeah. Sorry, I'm dragon shaming now. Okay. But I actually think cling did it better.

Look at this face. It's really good. Look at the scales. Yeah, I like it.

Overall. Cling did it better, I think. I mean, Kling has audio where Runway didn't, but just, like, the physics and, like, the speed that the dragon moves and the speed the wings are moving, it just feels more like what you'd see in a movie than the Runway version.

It looks majestic. Honestly, that's how you should see a dragon.

Yeah. Yep. All right, let's try our dinosaur. There's a dinosaur wearing mittens. There we go.

Yeah. Looks good.

All right, so A T. Rex wearing mittens making brownies on Mars. Here's what we got. This one's going to have audio. Almost done with the brownies. It was not expecting that.

Me neither.

I was not expecting the dinosaur to start talking to us.

Different barefoot contests or something like, what is happening BBC food right now.

I mean, it's pretty good. There's not a whole lot of motion to the video. It's just the dragon sort of stirring. And then he talks to the camera saying, almost done with the brownies.

And the mittens look better.

The mittens do look better. They look like mittens. They look like what you'd expect mittens to look like. Yeah, I think they gave him too long of arms, though. I don't think T. Rex is too far.

Yeah, T. Rex should be, like, up here with the arms.

All right. Monkey on roller skates having a blast on these skates. It's funny because the monkey's just riding one roller skate like it's a monkey on a roller skate.

Why is he hunched over like this? What are you doing?

Yeah, that one, I think I give it to Runway runways was a little more cartoony. This one went a little bit more for realism.

You could see the fur. Yeah, you can see the fur on the. On the monkey.

I think it's funny how, like, literal cling makes these props where it makes the character turn to the camera and say what they're doing.

Yeah. Almost done with the brownies. That's the funniest thing I've ever seen. Whenever you need to test out, like, other. We should probably run with the T. Rex making brownies on Mars.

Yeah. Yeah. That's a nice, like, new benchmark to work off of, I think. All right, so create an ad for a meal prepping business. I wanted to test again of just, like, vague prompt. Let's see what it gives us. All right, so we have like, a, you know, a fitness looking woman on camera. Let's see what she does and says.

Get your week's meals ready in minutes.

I mean, there's a lot less going on than the other one.

I mean, she's not freaking out. She's not spitting stuff from her mouth. But I also feel like if you give it, like, more details, like, I think she cut the container in half, which is weird.

Yeah, right there at the very end. Her tongs are almost going through the container also.

It becomes a knife. Is that right?

Yeah, it does. You're right. It goes from tongs to knife.

That's weird. But I think visually, way Better than the freak out.

Yeah. She's not like vomiting food back out. She looks very realistic, right? Maybe a little like too sort of polished and realistic. Like you get this thing with AI images and videos where the people almost start to look like plasticky because they're like almost too perfect skin looking and stuff, you know?

Yeah.

So you have a little bit of that, but it's, you know, more realistic looking of a person than Runway. Like, if we look at Runway here, the motion is fast, but the people, there's just some like weird uncanniness about like all their movements.

I still don't understand what happened to that woman. Like, someone asked her. So she's spitting us where this one.

Like, it definitely looks more like a real person. You don't have some of the weird like artifacts you get with the movement, but there's also a lot less movement going on.

Yeah, this is better.

All right, let's see what our image to video did. So here's our Hansel and Gretel. My prompt was Hansel and Gretel skip towards the cabin as the camera slowly rotates around them.

Cute. Cute.

And that actually came out pretty decent. Yeah.

Yeah. I was just scrolling on X because I generally I'm not on X because I feel like it's too much for me. So I saw this thing, I was like, really curious of like how people are using Kling. And like I saw this person kind of take an image from midjourney, put it on Kling, and it turned out to be cinematic. It was like cinema. It was really good.

Yeah. No, I tend to find that when you're using video models, you almost always get a better result if you start from an image and generate a video off the image. Saying that. I much prefer to test text to video because what it should be doing is you give it a text prompt, it always seems to generate videos better. If you start from an image, generate it into a video. Why isn't it? When I give it a text prompt, it generates an image that's a decent image and then use that as the starting image to generate the rest of the video. So you'd think you'd be getting fairly equal results. But for whatever reason, starting from an image almost always, always creates a better result. And I don't quite understand it because I do feel like a text to video should have a very similar result. Just start with an image. Just do that part behind the scenes so it feels like we're starting with text, you know.

Yeah. But also, I think the image that comes out. It is formed in a way that looks pro, rather than you generating something from scratch. You know, like, this has the details. It doesn't need more details.

Right. So a camera angle prompts or lighting.

Exactly.

Or color prompts. All of that kind of stuff already exists. You don't necessarily have to put it to the text prompt because it's there. They have the, you know, the example already.

Yeah.

All right. So the other model that they came out with was this new cling01 model. And this one, you can actually feed it images, video and text, Any combination of image, video, and texts. So some of the prompts that I gave this one, just to kind of give you an example of what it was capable of. See how the prompt has, like, references to the images that I uploaded. So I gave it an image of, like, this Bugatti supercar here. And I gave it this image of, like, a close up of a pizza slice.

Now I'm hungry, bro.

And then I said, animate image 1, which was the car driving fast on a racetrack. The car is made entirely of the material from image 2, which is the cheese pizza. As the car turns corners, the body should stretch and melt slightly due to heat and centrifugal force, dripping cheese onto the asphalt.

It's very technical, by the way. Yeah.

This is what it generated. And this one does not have audio, but you could see it took the cheese from the pizza, made that, like, the paint job of the car. And I was able to edit those two things together and then create a video from the sort of two blended things. Now, the cheese was supposed to shift around, like, as it went around corners, but the video sort of stops right before it goes around the corner. So I don't know how well it did the physics portion of it, but it definitely took the texture of the pizza and added it to the car.

Yeah, it looks good, though.

The second prompt was I gave it this image of, like, the internals of a watch, and then I gave it a video of, like, a drone flying through a canyon. But I said, bring this image, which was the internals of the clock, to life using the camera movement and speed of. And this was the drone flying through a canyon. We're flying deep inside the watch mechanism. The gears must rotate in sync with the speed of the camera movement. And so the idea was it was supposed to look like a drone flying through the sort of, like, inner workings of this watch. Didn't quite work out super well. In fact, I'm going to unmute this one because it's Kind of funny. It took the audio from the drone, but none of the video from the drone. So if you're just listening to this in audio, what that was was just a picture of, like a watch, but with drone sounds over it. The other thing that I find super interesting, you probably can't see this, but down in the corner it says, veo, I'm in CLING right now. But the reason it did that was because my video of a drone flying through the canyon was generated with veo. I generated that with Void, brought it over to here, and then tried to make that drone flying through the clock. And it kept the VIO watermark in the CLING video.

Oh, my God.

So that's what happened there. This was actually one of my favorite ones. I gave it a video of me sitting in my office reading a book here. And then I gave it an image that I generated, I don't know, in mid journey or something, of like, a futuristic transparent tablet. And I said, replace the book in this video with the tablet. From this image, adjust the hand grips to hold the thin tablet naturally. The light from the tablet must illuminate the face and palms. Right? And the original video was me holding up a book. And this is what it generated. You can see I'm sitting here looking at this tablet now. Now it's not lighting up my face like it's supposed to. And when I'm done, it kind of.

Folds up like a book, which is so bad, though.

And then at the end of the video, I folded the book and set it down, and I wanted to see what it would do with the tablet. And the tablet sort of folded. But I mean, the first five seconds of this before the folding kind of looks wonky. I think it looks really good.

I'm telling you. I think, like, you can, like, sell this, put it out, like, on the Internet and stuff and, like, freak people out and, like, make an ad.

Be like, hey, buy the new clear tablet.

Yeah. Apple came out with this new clear tablet. And, like, I'm giving it away for people for, like, you know, that would be so, so cool. Do it on, like, April fool or something.

It's a good idea. I should try that. See if people buy it.

Well, I just said it on the Internet now it's just out.

Yeah, yeah, yeah. I'll do something similar. So if I'm promoting some sort of weird tablet or gadget on April 1st, you'll know what's going on.

All right.

So I gave it a video of a man on a treadmill. This is Just a stock video I found on storyblocks. And I said change the aspect ratio to horizontal because this Original video is 9:16 vertical view. Replace the gym environment with the surface of Mars. The person is now walking in a spacesuit. The treadmill is gone. They're walking on a dusty red terrain. In the expanded background show the futuristic colony in the distance. So first thing to notice, it did not change my aspect ratio. It kept it at 9 by 16. Didn't switch it to 16 by 9.

It doesn't listen.

Yeah.

When it comes to that, by the way, I've tried it on several occasions. Not just with this one.

I just don't think they're capable of it yet for some reason.

I don't care. They should be advanced.

I know. It feels like all the image models are capable of outpainting, right? They're all capable of me giving it a 9 by 16 vertical image and then outpainting that image to a 16 by 9. So making it a 16 by 9 and using AI to figure out what goes in the sides of the image, right?

Yeah.

If image models are capable of that, why aren't video models? Because theoretically a video is moving pictures. It's a whole bunch of pictures moving at a rapid frame rate. I don't understand why we can't do it on video. If we could do it on image. And if we can do this stuff starting from an image and we have outpainting for images. It doesn't compute why they haven't figured out how to get this to work.

On video if it can generate the most complex prompt in the history of everything. Because I'm pretty sure someone out there has neurodivergence like me and wants everything to be so compacted and given the most detailed prompt ever. So that the end result would be. If it can do that and generate that video, it can definitely change the aspect ratio. Don't piss me off.

Well, here's the video that it generated. So the original was somebody running on a treadmill. And this video, it's the same motion of them running on the treadmill, but the person got changed. You can't really tell. I know it was small on my screen, but the original person was a black guy and this person's a white guy. So wait, just totally changed the person in the video.

Cling. Should we have a talk? Why not? That's not good. That's not good, honey. We should fix that.

But. So it changed the face. It did put him on Mars and in a spacesuit. But it definitely changed the person in the spacesuit. It didn't change my aspect ratio.

You can change a race, but you can change an aspect ratio. Really? Fix it.

Yeah, fix it. So anyway, that's what it did. But what I was testing here was, you know, giving it a starting video and seeing if it can sort of reskin the starting video. And it literally reskinned.

Bad, bad cling.

Bad cling. So my last one, this one was actually kind of hilarious. So I gave it an image of myself just sort of standing there with my arms out, right? And then I said performing a tap dance in the style of. And I gave it the old, like, rubber hose style, like, animation screenshot. The character must completely adopt the rubber hose body proportions, noodle limbs, black and white aesthetic. But the face must remain recognizable. The face must remain recognizable. Key ingredient.

Who is that?

The original image was an image of me.

What is that?

And this is the recognizable face that it generated.

Could be you in another life.

The funny thing is it kept my shirt perfectly. The shirt that this cartoon character is wearing is the exact shirt that I'm wearing in my original image. It just ignored the part about making.

It look like, why don't you have pants? Is my question. Where are your pants?

Because the old noodle cartoons, I guess they didn't wear pants. Maybe, I don't know.

It took off your pants, which is.

Take off the pants.

Pretty sus. But, like, my question is, like, where did you take that picture? Like, why do you have, like, photo shoots and stuff? I want to have photo shoots as well.

This was for, like, a podcast photo shoot I did like 10 years ago. This image is like a 10 year old image.

10. You look the same. It's like, drop your skincare mat, Wolf, right now.

So, yeah, those were the tests that I did on cling01.

Did the public normies like me have access to to qu.

Yes. In fact, I think you can generate a certain amount for free per day. Oh, let me actually just check. All right, so here we go. BASIC is free forever.

Oh.

So I'm on this plan right here, which is a little under $9 a month.

That's pretty good. That's not bad for the stuff that we've done. It's pretty good.

Yeah. So it says you can generate about 33 videos a month. And I mean, I've actually never hit my limits because I'm not sitting here like, generating tons and tons and tons of videos. I'm doing like 10 tests, and then I'm moving on.

I'm Pretty sure you do more, but you don't talk about it, but probably go to other stuff and like, generate more. You wouldn't even know your limits because I know for a fact you test them all.

I do test them all, but some of them, I do have like, special access, you know, like, they hooked me up.

Oh, yes. We don't. And like, you do. Yes. Just flaunt that around, why don't you?

So, yeah, those are the new video models that have come out.

They're pretty good.

There was also the McDonald's ad. We wanted to talk about that really quickly because it's, you know, somewhat related.

I wanted for us to like, you know, like how we did the Runway thing with the meal prepping thing and like, basically they just pulled it down when it comes to the, the ads that McDonald did and everyone, everyone hated it. Everyone. It's not as wholesome as the Coca Cola one. The Coca Cola one was fine, but the McDonald's one was so bad, it was nightmarish. You're using the McDonald's cup without a straw. It's disgusting on normal days. Are you not gonna use a straw? No.

I just think it's funny that that was the thing that you picked up.

On and not like everything else is like. Because I read a lot of weird things and everything is very thrillery and dark stuff. I didn't have an issue with that. But this. I had an issue.

That's so hilarious. It's like you've got AI people getting pulled by trolleys and FL out windows. And the one thing you pick up on is they didn't use a straw.

You couldn't translate that. Anyway, that's my take on the McDonald's thing.

Yeah.

Yeah.

I think, if I'm being honest, I don't think the ad looks that bad. But saying that, I feel like people have been so inundated with like, AI video and images on social media lately. Like, you can't go on Facebook and scroll. Like, there is so much AI slop just in the Facebook feed. Right? Like there's AI generated videos. And my wife gets confused all the time. Like, I remember during Halloween there was all sorts of videos. She's like, look at this awesome Halloween costume. And I'm like, that's AI honey. Right?

Your wife gets confused. My 59 year old mother is always sending me things, Matt, asking me if I could find a way if it's real or not. What do you mean, bro?

Like, yeah, I think people are just like saturated with that stuff. People have gotten so Much of that in all of their feeds. And they're just so over this, like AI slop that like everybody's putting out now. And they're so over seeing stuff in their feeds and going, is this real or AI? I don't even know anymore. And now you have like companies like McDonald's and Coca Cola going and doing the same thing. And people are just like, oh my God, we're already exhausted from normal people doing this. Why do we have to see this from the big freaking corporations now too? Right? And I feel like that's more of the blowback than anything is like we're already inundated and frustrated with how much slop we're seeing. Why are you guys spending money on slop two? You guys are mega corporations with billions and billions and billions of dollars. Hire VFX artists, hire actors. Like, why are you giving us slop two? You know, that I think is like what a lot of people have a problem with.

Not even McDonald's is going to get broke if like they hired someone to make their ads. Honestly, I didn't like the idea that they're using AI ads. And I'm a person that advocates for AI every single day. That's me.

So I think AI needs to be used as like an assistive tool, not as like the main tool. And I feel like things like this ad, they just went all in and made everything AI and made it obviously AI. Where I think if they were to do like let's say 80% of the video is real people, real actors, real VFX, you know, CGI, not actual AI generated stuff. But then maybe there's like one or two scenes in there that they can't figure out how to get with real humans or it would be cost prohibitive to do it with CGI. So you do AI to finish that little like 10%. I don't think anybody would have a problem with that. People are like, the big movie studios are already doing that, right? Big movie studios are already making movies still the old fashioned way. But there's little shots here and there where AI makes it a lot faster, but nobody really notices. It's not like AI slop within a bigger movie. It's like, okay, we needed to de age this person. So we used AI to make their face look a little younger for that one scene patching up. I don't think people have a problem with that. People have a problem with, all right, you just went and gave them 20 prompts and stitched a video together and said, look, here's our Christmas ad this year. People have an issue with that.

Yeah. With no straws also.

But anyway, that was a nice look at the world of AI video. It started real fun talking about all of the latest models that we can play with.

And then we spoke about McDonald's. Yeah, yeah.

But you know, in some scenarios, AI.

Kind of sucks in these areas. Yes, honestly. But yeah, I had so much fun on this episode, honestly.

Awesome. Yeah, I mean, it's been a blast making these last handful of episodes, just sort of recapping and playing around with what's going on in the AI world. I'd love feedback from the audience. If you're listening on Spotify or watching on YouTube, let us know in the comments. Spotify actually has comments now, so you can comment there as well. But let us know what you think of this. Do you want us to, like, do more podcasts like this where we go deeper on individual tools and test them out and see what they can do? We always love hearing your feedback and we try to steer these episodes towards what you want to hear and learn from us. So always appreciate it and thank you so much for tuning in. If you like this episode, give it a thumbs up and consider subscribing wherever you listen to your podcasts. And thanks again for tuning in this week.

Bye.