How to Create AI Films Better Than 99% of People

If you only creating AI videos by writing prompts, you're only using about 16% of what AI video can actually do. I've spent more than 5 years working with AI video both as a researcher and a creator, and I can tell you that the difference between generic AI slop and cinematic AI films comes down to these six core AI video features. And once you learn those six, you'll never create AI video the same way again. Starting with the first feature, image to video. This is the most simple feature where we actually upload a reference image into the AI video generator and use that as the first frame. You can also enter a prompt that directs the scene inside the AI video. While image to video is the most basic feature, all the other features are going to build off of this one. By using a reference image, it lets you control the composition of the scene and the exact visual style. For example, if you want a brightly lit scene or if you want something that's darker and more desaturated. If you don't use image references, you're giving up a lot of control and basically asking for random results. The control you get from image to video doesn't just end with the exact visual style. It also extends to controlling how your characters look. It lets you use a design for your characters, like this character sheet that I have right here, and combine them with specific environments that you want the characters to appear inside of. So, I can upload my character and environmental references into an AI image generator and then write a prompt that combines these two elements together. And then using all these different images you've generated for your consistent characters, it's straightforward to use image to video to then animate these scenes. Consistency across different shots is what makes your AI scenes actually feel like a film instead of a montage of random clips. The second core AI video feature is multi-shot generation. Instead of just generating one scene at a time, what multi-shot generation allows us to do is to generate multiple shots at the same time inside the same video clip. This lets you start generating sequences of shots instead of just individual clips. It lets you plan out different camera angles in your scenes and also control the pacing by defining how long each individual shot inside the sequence should be. The key to generating multi-shot videos is your prompts. Inside your prompts you'll need to define the specific different shots that are appearing and what's happening inside of each one. And it can also help to add in timestamps for when you want each individual shot to occur. Late again. Always late when it matters. The train? No, the place we're going. It moves when it shouldn't, stays when it shouldn't. >> Why go there at all? Because once you're invited, you don't get a choice. If you want to level up your multi-shot prompting, use storyboards as image references just like this one that I generated. Generated a 12-panel storyboard using the uploaded images about a woman who boards a train in the forest. There is a train conductor on the train and a few fantasy creatures on the train as well in the visual style of the attached image bright saturated colors. Storyboards allow you to have more control over the exact shots that you want to appear inside your scene. With your generated storyboards, you can simply combine them with your character reference sheets all together inside these AI video generators and give it a prompt that defines what happens inside each storyboard shot. And the AI should do a pretty good job of following the directions that you give it. When it comes to generating AI images, some of the models I'd recommend would be the GPT image 2 model and also Nano Banana Pro, which has been good for a long time. When it comes to generating AI videos, the best models include C-Dance 2.0 and the Kling 3.0 models. To access all these different models, I'm using a platform called OpenArt. I'll put a link in the description if you want to go and use all these different AI models and tools yourself. The third core feature is start-to-end frame animation. This is where we're using two image references, one for the first frame of the AI video, and one for the last frame that we want the AI video to end with. This lets us direct the motion instead of hoping that the AI video is going to generate something that works for us. It lets us create very specific-looking transformations, especially ones on top of our characters. It also lets us control camera movements and motions from one scene to the next, where the camera may be sliding or panning to turn to another character. But beyond that, we can create really advanced camera movements as well. For example, starting inside a train cabin and having the camera fly all the way outside of the train. The map indicates we have arrived at the valley. Then we shall finally see the legendary tree. The only thing this requires is for us to upload a frame for the first shot of the video, and then also upload a frame for the exact ending of what the video is going to look like, adding a prompt that chains those two images together, and it lets us key frame between the two different shots. The fourth core AI video feature is motion transfer. In this case, we're borrowing movements from real footage and transferring them onto an AI video. We can have a fight scene where we have an actual real footage and map those motions onto AI-generated characters. All we have to do is upload that driving real video footage and the AI characters into the C Dance 2.0 AI video model and then write a prompt that tells AI to map the movements from the real footage onto our AI characters. This feature has actually existed for a while, but in recent times it's gotten way better and can be used on even more dynamic action scenes like this one. This is also a great way if you want to act out the scenes yourself and we can transfer that onto a new character in a new environment and also give it a different sounding voice. Okay. Planned is kind of a big word. I just saw a wheat field around. That was when he lied on guard, so I thought I'd take a little for myself. The fifth core AI video feature is dialogue creation, making actual talking AI characters. The simplest way to generate this is to just directly ask in your prompts for the AI characters to talk. You can define the specific words that you need each character to say. This is an especially powerful tool when you combine it with multi-shot generation, where you can have multiple scenes of characters. So, now you can generate a full sequence of shots of a conversation. It's a quiet place. People go there when they're between things. Between what? Where they've been and where they're meant to go next. And they just stay there? Only as long as they need to. You can also use AI lip sync, where you directly generate what the AI dialogue sounds like using a separate tool from the AI video generator. Eleven Labs is a pretty popular one that generates AI dialogue and then you can combine that AI dialogue that you generated using a separate tool with an image reference and use a dialogue to animate the character talking inside the image. There's quite a few different tools for this. For shorter dialogue that isn't that long. C dance to win out seems to do a pretty good job. None of this should exist. The train, the passengers, any of it. But somehow it all feels strangely normal now. You'd be surprised how quickly impossible things become ordinary once you stop resisting them. The sixth core AI video feature is video editing where we're taking a already existing video clip and then using AI to modify it. This can be used in a variety of different ways. We can change the background in the scene so that it's a snow-covered forest landscape with a blizzard outside. We can also change the lighting in the scene so that it's dark at night time with moonlight shining through the window. People go there when they're between things. Between what? Where they've been and where they're meant to go next. >> To use AI video editing like this, all you have to do is to directly upload your already existing video clip inside the AI video model and then in the prompt just directly tell it how you want the video to be modified. It's a quiet place. People go there when they're between things. Between what? Where they've been and where they're meant to go next. >> Another cool way you can use AI video is to directly extend an already existing video clip. What you'll need to tell the AI is to continue the motion from the last frame of the video that you're trying to extend and the AI video generator will generate a new video clip that extends your reference video. AI video editing is where you stop just generating clips and start using AI video as a powerful visual effects tool. We're already running late. If you want 10 more tips on how to create AI films that are extremely realistic, go watch this tutorial right here. You weren't supposed to find this train. Not yet. Where is it going? Somewhere that doesn't stay [music] in one place. That doesn't make sense. It will when you arrive.