Native Consistent Storytelling in AI Video is HERE! | Full Breakdown

That. Welcome back, everybody. I hope you enjoyed that little video in the intro. That video was entirely generated with AI front to back, mostly using cling's brand new Elements feature, which aims to do consistent objects, consistent backgrounds, consistent characters with AI video. And the tool is pretty good so far, so we're going to be talking about that a lot. But I'm also going to go over some of the other AI tools that I use to make the video and we're also going to take a look at what the possible future of AI video creation could look like and the overall race for consistency. Because CLING isn't the only platform out there that's trying to do this right now. Without further ado, let's dive right in. First up, CLING elements. Here is the user interface. As you can see on the left hand side, you can upload up to four separate elements that are going to be included in your final video. But you don't have to do four, you can do just two or one, which I did in some circumstances for the video you saw in the intro. These elements can be people, they can be animals, they can be objects, or they could be scenes and backgrounds. All you have to do is upload those images, those reference elements, and then describe in the prompt their actions and interactions. It's very seamless. It's pretty much exactly how you would expect it to behave. I like the UI and I like the way they went about it. Obviously, one of the main use cases for Elements is going to be character consistency. And as I hinted at earlier, they are not the only company that's trying to do this. Here's a pretty great example of how things can be combined together. We upload this lady. We've also got this crown made out of gems, and then we've got this background with stage lights and it combines them in a realistic way. Now, it's not perfect. Cling, being a Chinese model, has quite a lot of Chinese video data, so I think it changed the ethnicity of the character here just a little bit. But as you saw from my example, I don't think it changed my ethnicity too much at all. I think it was pretty stable in maintaining my facial features. The crown is. Is pretty darn similar, even up to the point where we've got these smaller gems and these larger gems on top. So that's really impressive in my mind. And then obviously the stage lighting in the background colors and all, it animated it as well. Looks pretty stellar. Here's another one. We've got the photo of a cat. We've got the glasses and then the leather jacket. Combine all of them together in elements and you get something pretty awesome. So plenty of examples. Everything's always linked down below. You guys know my videos by now. Let us now talk specifically about the video saw in today's intro. As you can see, I generated quite a few clips to put that all together, but really not as many as I thought I would have to. Cling Elements is far more consistent than I originally anticipated. And while Cling Elements is far from perfect, it is the best implementation I've seen for consistent characters, consistent backgrounds, and overall consistent storytelling. Just because you can combine multiple elements, different types of things together in the same video, rather than just a consistent character or just a consistent object. Alright, so here is the first example that I ever generated with Cling Elements. As you can see, it's a video of me kind of creepily walking around a server farm. Probably going to steal a few GPUs now. I will say, guys, to use Cling Elements in the first place, I did have to pay for a standard plan, which was about 10 bucks renews monthly. You can cancel it whenever, but still, ten bucks to generate what you saw in the intro is pretty darn cheap, honestly. And AI video will only become cheaper and cheaper over time. You can see the reference images. First off was this, you know, somewhat creepy photo of me. Obviously this is one of the photos I will commonly use in my thumbnails. I've got my hand on my chin like this and I'm just staring into the camera, smiling. It was able to capture that very, very well. It definitely looks a lot like me. The eyes are a little soulless and dead and creepy. Sure. The hair is also really accurate down to the very curls. And so is the sweater that I'm wearing. It's definitely a gray sweater. Plus for good measure, it did also put the hand on my chin, but I might not want that in every single scenario. So I was pleasantly surprised when I continued to use certain images of me that it didn't constantly. Just put my hand on my chin, for example, when using that as a reference image, I left everything on default for this first prompt that you see here. Standard mode, which gives you faster creation, takes about one to six minutes to generate in my experience, and it costs fewer. Cling credits can also do different durations from 5 to 10 seconds, and I use variations of all of these settings now. They also have a professional mode which gives you richer details, superior quality, and overall better prompt adherence. But the generations are about twice as Long. In my experience, you're looking at anywhere from six to 12 minutes, I would say. Next up, I tried something else. You can see I used that same photo of me with my hand on my chin. I also used this funny image of a llama eating a GPU inside of a server farm. And then finally I uploaded that same exact server farm background photo. Again, it ignored that photo for some reason, but the combination of the other two elements was surprisingly good. The prompt here was a man looks surprised as he finds a llama in the server farm. So not even prompted in here that the llama is supposed to be eating the gpu, but it sure as hell scarfed it down. And you can see I look pretty surprised. There's a little bit of morphing on my face and a little bit of weirdness for sure. But definitely still combining these elements together in a way that we haven't seen video generation do in the past. Again, this was on standard mode 5 seconds. Next up, I decided to start actually moving into creating the video you saw in the intro. Okay. That was my third generation and this is probably the best generation that I got from the whole stack, in my opinion. First, we're uploading the background image. I think that cling elements might actually be put priority to the first image that you upload in here. So that's something to think about. And there's a reason I chose the setting as the priority. I wanted consistent settings above all else when creating this. And it definitely accomplished that pretty well for the image of myself. As you can see, it's a little bit different. I'm still sitting in my chair at home. I've kind of got a scowl on my face or, you know, I'm serious, I'm focused. Still sitting in the chair though. And you only see like the. Not even the top half of my body. But it was able to create a cinematic, very nice looking, Indiana Jones style shot of me walking into this pyramid area or this excavation site with all these bricks, very dramatic lighting. It slowly sort of pans down. It's awesome. I really like this clip. And you can see while the background might not be perfectly consistent in this shot here, there's an archway, very particularly sized archway with a light coming from one side. And there we see an archway with light coming from one side. It's just shot from a different angle. So it's picking up on the details of this background and applying me in a way that makes sense. It's. It's crazy. Cling elements is totally wild. You see, for the prompt here I did. Cinematic moment. A man slowly walks into an ancient tomb. He has a serious face. So it's still a very simple prompt, but it nailed it overall. Now this one, I will say I did use Professional mode, which I think enhanced the quality level of the clip as a whole and its ability to mesh everything together well. But that doesn't mean standard mode is useless by any means. Professional mode does take, like I said, twice as long to generate and costs an extra 10 credits. Next up, I wanted to create the villain and I wanted to put him, or see if I could put him in the same place as me. And you can see I actually didn't end up using this clip in my final shot, even though it was generated with professional mode and looks pretty cool as well. This is the villain of the story. This is the singular image that I used for that character. Glowing red eyes, this, you know, pale metallic face, but very humanoid looking. You can see this character also has a futuristic background attributed to him, which could cause issues with cling elements. But honestly, it handled it better than I expected. As you can see in this clip, it kind of combines both backgrounds together, which in some cases you might want to see, but not in this particular one. And that's why I use the clip, because obviously it wouldn't make any sense to the story. There's no, you know, blue technology in this tomb, especially not in this clip. You see of me walking around. So if we're gonna have these in the same story, you know, they gotta be consistent. As you can see, things are kind of generated out of order. I made this kind of entirely just experimenting with cling and then put it together. So I'll show you the timeline of the final video and some of the extra things that I did to put it all together at the end. Next up, we've got me wandering through the halls of the ANC tomb with a flashlight. Here is the background image that I uploaded. Obviously much more dimly lit. We've got a box over here on the side. We've got this piece of stone or the stone artifact. And then the image again is just me chilling inside of my room. But I've got my hand up to my face and I'm looking very surprised. I'm a little frightened. And sure enough, in the final video, it definitely combined all this together. I will say that it doesn't exactly look like this reference image that I uploaded. And by the way, if you're wondering where these reference images come from, all of them were generated with ideogram. AI so it didn't get the roam exactly perfect, but it was good enough to end up using in the video, especially since it's so dark and I'm just wandering around with a flashlight. And it also made sure that my hand was up to my face. So that is something that I'm going to warn you guys about. With Cling Elements, if your character is doing an action in the reference image, it might appear in that final video generation, but not always, especially if you prompt it not to do so. Anyways, this was View From Behind. Man slowly explores the inside of a tomb using a flashlight to light the way. Honestly, another thing that was really impressive about what Cling Elements is able to do is the fact that I can change the camera angle. It's not always going to have me facing forward like in this reference image. It's able to take the concept of me. It's able to take the concept of this room and rotate it entirely behind. And now it's a shot following me from behind as I explore. I still have the same hair. I still have the same face. It's really cool. This was also a professional mode generation. There's a lot of warping though, still in some of the textures. It's still AI mushiness a little bit, but that's all going to improve in the future. Getting to the point where we can include these elements and prompt it in a specific way has been a long ride. And once we can nail this elements feature down very well, all that's left to do is to finish up with the quality right, and make sure that textures don't get mushified. Next Generation. This was a flop. I was trying to get the evil villain running behind me in that same exact scene. You can see it did a pretty decent job with the background. But this clearly is picking up elements from video game training. Third person video game training, where we have a smooth camera pan from behind. It doesn't feel very cinematic. I should have prompted it to be cinematic in the field and I didn't do that specifically. Next up, we have a brand new scene for me entering the tomb. Obviously, as I was creating the mini movie here, the little two minute, I was realizing I needed more, you know, establishing shots to establish what this character is doing and how he got into the tomb. So another image of the entrance to the pyramid generated with ideogram and then combined with just a random image of me. I'm doing a silly face in this one, but you can't even see it in the final render because again, in the prompt, I specifically asked for a shot from behind. Slightly nervous looking man. And I think you can even see in the walk. I am kind of walking a little bit nervously, but it did a great job, you know, moving me into the entrance of the pyramid. Also in professional mode for this one. Next up, I retried the shot of him entering that specific area, you know, where the sunlight is coming down. It actually made it look more like sunset, which gave me the idea of putting this shot further in the video, further in time, which worked out pretty well. But it got his shimmering face, face, his glowing red eyes very nicely. And, you know, he's still got that suit on. So it actually did a really great job with him as a consistent character. And of course, the background with the archway. And this is a much wider shot than before. So it honestly felt pretty seamless, I think in the final video. Next up, we have a shot that went very, very well with this one right here. I used the same exact images. So this is the same thing of me covering my face, shocked or surprised. And then I used this same inside of the tomb reference image. And it gave me a very cinematic view. Me holding the flashlight, looking very surprised, finding something in that dark tomb. It looks awesome. Very cinematic. And again, keeps my hair consistent, keeps the shirt consistent. Like, that's small details, but they mean a lot. Because you'll notice that for sure when you're watching something and you're trying to follow along. A story explores the inside of a tomb. Pretty basic prompt. Also professional mode. Next up, I wanted to redo the failure from this generation, so I used a better prompt here. Again, the same two images, but I want you to take a look at how consistent it was able to get the setting in this one. Sure, there's a little bit of glitchy weirdness going on in here, but that's something that is going to be improved in the future. We've got the same exact little artifact right here. The same exact little piece of wood. It's the same exact setting. And of course, the glowing red eyes on this humanoid robot character as he moves through the room. I mean, it actually follows him and animates that particular setting. It's so cool. Consistent elements for sure. I actually prompted it for a still camera shot, but it followed him from behind instead. And it worked out anyways. And this was, by the way, guys, just a standard mode generation. So it was able to pick up that entire scene there. This entire consistent scene just in standard mode. The glitching that we saw earlier here with that movement probably wouldn't be as bad in professional mode. Next up, I was trying to get some establishing shots of our villain near the pyramid, but it was having a real difficult time. I think this prompt was a little too complicated for Kling elements. It's cinematic video begins with a view of the great pyramid, which it did. Then it pans to reveal the AI robot and he is headed to the pyramid. I mean, it was close. We start with the view of the pyramid as we're supposed to. Then the robot comes into view and he's pretty consistent. A little mushy on the face there, but has the red eyes, looks like the same character. He's just walking away from the pyramid instead of towards it. And then it also pans to show another pyramid here. So it was just a little bit confused. The prompt coherency, the level of prompt understanding is not yet good enough to be able to take a lazy prompt like this and and just send it through perfectly. Here's another generation in standard mode, this time with me trying to enter the pyramid. But again, we sort of had the same issue that you just saw where it's got me coming out of the pyramid or walking away from it instead. Because I prompted a shot from the front and it wants to use that as a consistent background. So in order to get that shot from the front that I was trying to achieve in this scenario, I had to just get a basic desert background. As you can see here. Again, just generated ideogram, uploaded the same photo of myself, and basically just did the same exact prompt. So I'm in the desert and I'm stumbling upon something in the desert. This was also done with standard mode. Again, a little bit of face glitchiness. Not perfectly looking like me, but pretty close. Close enough to be at least usable for these little experiments. Definitely the same hair, definitely the same shirt. At the very least, I think it's just the facial features that maybe it's a little bit messed up on. Next up, use the same background, but then this time with the villain. So we could show the villain appearing and ready to head towards the pyramid. And that worked out pretty great. He is a little bit squirrely, a little bit weird, kind of coming out of the ground in this scenario as the evil robot. But he's very consistent. A nice grin on his face as the camera zooms in on the character. It was absolutely good enough. Next up, try generating some more videos of me exploring the tomb. This one messed up the flashlight and put it on my shoulder for some reason. But, hey, it was able to pick up on the fact that my shirt here actually has a little logo. And it tried to replicate that, which was cool. Again, this one, for example, was on professional mode, but it put that flashlight on my shoulder like I didn't want it to. So the difference between professional mode and standard mode isn't all that great, in my opinion. Especially not worth the double generation time. Next up, we got this shot again with me with the flashlight, and it was able again to replicate this exact scene with the box with the little artifact with the blue light in the back background. Very consistent. This turned out pretty great. The beginning obviously got some weirdness going on in my face. This was only on standard mode, and it nailed it yet again. Next up, we get to the doorway. So I'm opening the giant golden gates and pushing them forward. This was actually a professional mode shot that I didn't end up using in the video. But this one was also really good. It's 10 seconds long. It's a nice shot from below, and I'm like, ripping open these doors. And there's obviously an issue, though, with this, and it's not perfectly consistent. It has decided to rip the sleeves off of my sweater. But other than that, it was so close to being perfectly consistent as I opened those giant golden doors. So super cool. And obviously, here is the giant golden doors originally generated by our friend Ideogram, and they're very, very consistent. It's very close. I don't know what kind of magic they're doing in the background with elements to be able to put me in there with the photo and have it so close and consistent. But it's. It's crazy. Um, it also messed up sometimes. Like, since I was sitting in a chair, the AI was able to pick that up. It was able to pick up the fact that I was sitting in this chair at a desk. So sometimes it would have issues, such as this one where I'm trying to open the golden door, but I'm sitting in a chair. So you'll see more of that later. Finally, this is the winner that I ended up using. You can see there's some big morphing going on in the beginning of the clip, but I just cut that part out and then continued from this end, which was all really good idea. Get up. I've got the full sweater on, and I swing the doors open confidently. And it's a nice little cinematic clip. You can see it was literally just view from behind. Standing up. Slowly pushes the giant gold doors open in the tomb Just the standard mode generation. Here is some more crazy weirdness that didn't work out quite well. Was able to capture like the soft glow of the light and a lot of the details of the interior of the tomb. But it still has me sitting down in a gaming chair and then glitching and flipping around and then going to find a chest which I then open. And I still use the end of that clip as well with the, the opening of the chest because I thought it was quite nice. But yeah, cinematic. Man kneels down to open the ancient treasure inside of the tomb. He looks excited. So the, the issue is that it was trying to put me in the chair and then open the chest all in the same clip and it just totally screwed up. We've got a better clip though, where I'm actually leaning down and you know, feeling around the chest and then finally opening it up, which was pretty cool. I like the cinematic shot where it starts up at my face and then pans down to reveal the object I'm working with. It's very cinematic. That's very nice. And again, same two reference images. This one was also standard mode. Finally we've also got those same reference images. Different prompt this time though, to get a close up of me kneeling down with my face illuminated by yellow light, looking surprised at what I found inside of the chest. And obviously that ended up being the RTX -4080 or whatever ancient GPU from the BC times that's inside of the the tomb. I didn't end up using this clip. I was trying to get me reaching out to pull out gpu, but I was starting to run out of credits and cling. Just did not want to do it. As you can see, we've got the same two reference images, but this time I added the one of the GPU that was generated by Ideogram. This was also a standard mode generation here. I mean it still got all the elements like I'm still in there. It's just a little bit mushy, it's a little bit weird. I'm like conjuring dust out of the GPU for some reason. And then we've also got this one again sitting in the chair. This time I pick up what looks to be a golden power supply or something like that from a chest. So it's just not super consistent. I wasn't able to use this. It was a rerun essentially in professional mode of this generation. But finally I was able to get something somewhat decent where I finally pick up an object out of the chest. That at first appears to have three fans, so it was consistent enough for me to include. Once you go beyond two elements, again, it starts to become a little bit more difficult to combine them all together, especially with a more complex prompt like this, where I'm in an ancient tomb, I'm illuminated by yellow light. I'm looking surprised. I'm picking up the GPU out of the chest. It's. It's starting to, you know, push the model to its limits. Here is another failed generation in professional mode. Again, it still got me in that chair and in the gaming chair lifting something out of a chest. Finally, though, I was able to get this shot, which I thought was pretty decent. The lighting wasn't exactly what I wanted. It was a little bit more blue, but he bursts out of the ground like he does earlier, and I thought that his face was very consistent, you know, that metallic pale skin with the glowing red eyes. Pretty good. And like I said, I was running out of credits, so I pretty much had to use it. I also got this generation of me looking really scared, you know, seeing this evil monstrosity, the villain, for the first time. And I thought that I looked pretty frightened in this particular clip, so it was very useful. And then finally, I wanted to try to get everything in the same shot. And it seems like the more elements you add, the more clink struggles, the more complex the prompt as well. So this was kind of everything at once, and it was able to do it. Okay job. By the way, this was only in standard mode, so we've got the consistent background here. Again, that same ancient tomb with the glowing yellow lights, it's also got me in there again. It put me in a gaming chair. It sees that gaming chair. So it had me in there. But the GPU actually looked pretty good. I had it in my hand. It has the three fans there. Shockingly decent. And then finally, it did get the humanoid robot in here, but he looks quite a bit different than the reference image. He's got the glowing red eyes and the pale face, but his body looks a little bit different. So I would like to have rerun this one. But like I said, I ran out of credits to accommodate for this one. In the final video, I was actually able to just zoom in and cut the chair out entirely, and it looked somewhat decent. Good enough to include as a. As a little part of this experiment. So, yeah, overall, I'd say the success rate was maybe about 50 to 70%, at least for what I was trying to achieve with this. Obviously, it's all far from perfect. But this is the closest to consistent scenes, consistent characters, consistent objects, consistent backgrounds that I've ever seen from an AI video platform. And I think the end result from this whole video demonstrated that pretty well. Alright, so moving on from cling elements, I'm going to show you the editing and how I put everything together. It's going to look a little bit more complicated and we'll talk about storytelling with AI video and the creative process. I'm also going to mention some of the other tools that I use to bring this all together. As I mentioned in the beginning of the video, it's back to back AI generated. So let's take a look. For those of you who are content creators, here is some video editing software you've probably never seen before. This is Vegas Pro. I still use this to edit videos. A little bit off topic, I'm not going to get into my creative process and how I put my videos together in this particular video, if that's something you'd be interested in, I guess let me know. But yeah, Vegas Pro is the editing software that I use. So that's what I put this all together with. These two bars right up top are just the cinema bars, if you will. So that's these black bars at the top and the bottom. I wanted it to feel kind of like a movie, so I tossed those in there. Obviously these are all the AI generated video clips that make up the entire thing. And then down here we've got all of the different audio. And the audio is very, very messy. So a lot of these different channels for the audio could be theoretically combined together. But I just didn't really bother because I whipped this up pretty quickly. In fact, in total, it took me about three hours to produce this little two minute short. Now most of that was just me sitting around waiting for cling to generate the videos. So the main bottleneck actually for the AI video creation process in my mind right now is waiting for the videos to generate quite simply. So the faster those generations come out at a higher quality, the faster you're gonna be able to put things like this together. And also just wait until AI video editing software comes out and can put all of this together for you just have to upload the files or prompt an AI. Now shots such as this one that you can see this establishing imagery to just show that we are in the desert, kind of setting the viewer up for what they can expect or where they can expect the story to take place. These were all generated with Runway ML. Honestly, I could have used any AI video generator to do stuff like this. It was all just image to video. I use Runway because I have a ton of Runway credits and it generates pretty quickly. So this was also generated in Runway. This was just supposed to be, you know, an establishing shot of me. I wanted to have the Nike sneakers in there because I thought it made me look like a casual YouTuber person, which I am. So I thought it would just go well. Obviously we had sound effects, stepping sound effects throughout all of this. All of these sound effects were generated by elevenlabs sound effects. And I've also got a ton ton of 11 labs credits. So it's got decent settings as well for creativity and length. So I was able to do things pretty precisely. So anytime you hear like, you know, the footsteps, That's all done through 11 labs sound effects. Now for the music in the background, that was through pseudo AI. I could have used, used Yudio or any other music generation platform really. But Suna was able to give me a decent sounding cinematic adventure opening in like two tries. So. So here's where I kind of want to talk about the creative process. Editing is extremely powerful. It's just as, as powerful in terms of visual storytelling as the shots themselves. So you have to keep in mind that I am the one who edited all of this together and I did so with intention. That's something that AI still has yet to do with the exception of invideo. And video is pretty good at it. But take these insinuate. Since I didn't have a good clip generated by the AI video of someone stopping in their tracks. It insinuates that I've stopped, which means I found whatever I was looking for in the desert. So there are, there's definitely a lot of influence, right? Some creative influence from me as the director, if you will, or the editor of this, that helps bring the story together in a way that feels more natural, more cinematic. And again, the cold reveal of me, you know, stumbling upon whatever I was looking for I think adds to that cinematic effect. Then you know, we've got the cut over just revealing the pyramid. Obviously that's where I'm headed. No mysteries there in the storytelling. Then we've got a fade out. The music kicks up a notch as I enter the pyramid. There's a lot going on here as well that I should probably explain. So in this shot where I'm walking into the pyramid generated very nicely by elements, you can't see my feet moving, you can still see my steps sometimes, you know, when you do these AI sound effects that aren't directly tied to the video. They don't match up well, but sometimes they match up surprisingly well. Like in this scenario. It's pretty believable to your brain that the sounds you're hearing of walking on sand are my footsteps in the video. It's pretty believable. But obviously, once we change to this shot, which is showing me walking much more slowly than the shot prior, I had to use different sound effects. And I also wanted some reverb, some stone sound as I entered the tomb. I'm not really on the sand anymore. I'm inside of the pyramid, so a little bit more of a thud there. I prompted 11 labs a little bit differently to achieve that. And also, since we're entering this darker area, this area that has the potential for danger and mystery, I did get a pretty cool sound effect from 11 labs that was like this, like, sound. I'll bump it way up so you guys can hear it more clearly. You know, it's. It revokes that feeling of. Of mystery. Right. Anyways, then I combined a bunch of drilling and rock sound effects, you know, in order to properly reveal our villain. And also, since we had such a long hold on the. The villain reveal, the reveal of his face, we also added some robo effects in there. Of course, show like, this guy is a ro.