# How the Top 1% Make Cinematic AI Videos (FREE System)

Channel: Parker Prompts
Video: https://www.youtube.com/watch?v=LIK6ZayNndU
Duration: 15 min
Language: English
Words: 3014
Transcript page: https://viewrankai.com/tools/youtube-transcript/LIK6ZayNndU

---

[0:00] If your AI generated videos come out looking fine but never cinematic, you're at the right place because it only takes three steps to go from nothing to a finished cinematic short film. And you don't even have to write the prompts yourself.

[0:19] You're still here. I thought I really thought you'd have gone with them. They crossed the valley at first light. I heard them calling. I just didn't follow. Why not? because nobody down there talks to me. Go and catch them up. Go on. I'll be here tomorrow, same as always. So, in this video, I'll walk you through the whole process from start to finish in four completely different styles. And every one of those styles comes out of the same two tools linked together with an MCP connector. The first one is

[0:56] Hicksfield, the sponsor of this video. And it's an all-in-one platform with the best AI image and video models. It generates everything you're about to see. Clot is the other one, and that's where I do all of the work. It writes all the prompts for the scenes and shows them to me first. And after I approve them, it sends the whole thing straight to Hicksfield, so I never have to go back and forth between two tabs, copying and pasting. And setting up the connector is the one step in the whole

[1:21] workflow that happens outside the chat. I'll open both tools side by side. And on Hicksfield, I'll click MCP in the top navigation bar. There are a lot of options here for different tools, but I'll pick clot. That gives me the link I need and the steps that go with it. Next, I'll switch over to clot and click customize. Then connectors. This is

[1:38] where every MCP and plug-in gets added. The hickfield connector is not there by default, so I'll click add. Type the name in and paste the MCP link I just copied. Then it's a few more buttons to confirm and sign in. And that's it. Now it's connected. Your credits come out of whichever Hicksfield account you sign in with. So pick the one that has your plan on it. If you want to follow along and make this yourself, I've left a link to Hickfield in the description. So now I've got access to every Hicksfield model inside cloth. And from here I can use it like a normal chatbot. And now we

[2:09] are ready to create our first video inside a new chat in Clot. I'll switch over to Fable 5.1 which is the best one entropy has right now. And Hicksfield has a lot of tools and models. So, our first job is to let Clot know exactly which ones to use for what and which settings to run them on. I'll start by telling it I'm making four cinematic short films, each one in a completely different style. Then, right underneath that, I'll lock a few things in for the whole workflow, so I never have to type them out one by one. Again, for the images, it has to use GBT image 2.5

[2:39] Sunburst, and I want those coming out at 4K max quality in a 16x9 aspect ratio because the higher the image quality, the better the final result. And the videos run on CDN 2.5 at 1080p since that's the highest it goes for now. For the rest of the settings, the audio stays on and it's [music] in a 29 by9 aspect ratio, which is that wide cinematic ratio you're used to seeing in movies. And every video comes out at a duration of 30 seconds because that's the longest CDN can do in one go. And a whole short film needs every second of it. Then I'll add a bit more context

[3:11] about how I'm going to build each one and hit enter. Claude reads every setting back to me, so they stay the same across all four films. Next, I'll paste in two reference sheet templates, and Claude will use this to write the image prompts from here on. A reference sheet is one image that shows a character from three angles. And there are two versions because one is for humans, and the other is for animals and creatures. The human template uses one panel for the body and clothes, one for the back, and a close-up that locks the face. The animal one switches those to a side profile, a rear threequarter view,

[3:43] and a tight shot of the head so Clo has the right angles for a different body shape. If you want this for yourself, I'll leave every single prompt down in the description. Then, right before I click enter, I'll tell it to make the characters one at a time instead of both at once. And with that, [music] Cloth has everything it needs. First up is the photoreal film, and I'll describe the story in as much detail as I can. It's about two sisters in a stone room cut into a cliff where the younger one is leaving in the morning and the elder one doesn't want her to go. Claude writes

[4:12] the prompts for the elder sister and this is where you can read it through and change anything you don't like. Mine looks good so I'll tell it to run. And what I get back is a reference sheet for the elder sister with her built and her outfit from the front and the back in the first two panels and a close-up with all her facial details in the [music] third. Then I'll ask for the younger sister and give her the same brows, the same gray green [music] eyes, and the same auburn hair as the elder one, just softer and rounder in the face. [music] It's the human template again, so I'll

[4:39] check the prompt and run it just like before. And you can tell straight away that they are related, which is exactly what I wanted. The reason I make this one at a time is that Clo uses the finished elder sister sheet as a reference for the younger one. So both characters look like they belong in the same video without me having to describe the entire visual style again. Now, for the actual film, I'll ask for seven shots with the camera moving on every single one because if it never moves, everything feels flat. Even when the picture looks great, I also want the dialogue slow and quiet with no

[5:09] narration. Otherwise, Cedance might add a narrator that was never in your prompt. It's still one video generation, but Claude has divided the full 30 seconds up with timestamps. [music] And every shot has its own framing, field of view, camera movement, dialogue, and [music] timing. Field of view is how wide the shot is, and the bigger the number, the wider it gets. So, it's 25 degrees on the younger sister's hands, packing her back, and [music] 90 looking down the whole cliff. The camera move changes from shot to shot two, with a slow push in on the elder sister, a fast pan to follow her to the shutters, and a

[5:41] pull out through the window when she throws them open. The whole prompt finishes with separate rules for the audio, character continuity, and visual style, so the entire film gets made in one go. Next, I'll tell Claude to run it.

[5:59] I'm taking the lift down in the morning. [music] The early one. You've never even been down there. I mean, why would you leave for a place you've never seen? [snorts] That's That's kind of the whole point. Look at it. Just look. You'd leave all of this? I've looked at it every single night of my life. I want to see something I haven't. Okay, then you

[6:22] write to me every week. It comes out looking very cinematic with a really nice story running through the whole thing. I especially like the camera cuts and angles on the reveal of the weird city at the end, and the quality and realism are topnotch, and both sisters in it are the same two characters from earlier, which is why we

[6:40] made those sheets in the first place. They let us lock a specific person, animal, or object into a scene before we ever hit generate because otherwise it's really hard to control who or what shows up in the videos. Doing it up front is a lot cheaper, too, since we're fixing the character on an image instead of a 30-second video. So, in just a few prompts, I've got my first cinematic short film, and it looks incredible. And the next three films also come out of this exact same conversation without me having to start over. The settings and templates are already in there from before. So all I have to give Claude now

[7:12] is the art style, the story, and the characters. So for the one I'm creating now, I'll go with anime. And I'll add a quick note on the look, which is [music] that handdrawn cell style with flat blocked shadows and hard edges. The reason I describe it that way is because anime by itself doesn't really mean that much to an AI model, and Clot turns that short note into a full style description in every prompt. So, the model knows exactly what I'm after. And the story I'm going for is about a girl who rides her dirt bike up a hill full of wind turbines every evening to talk to a

[7:44] young stack. And today, his heart has crossed valley without him because he stayed behind for our main character. The first prompt comes in. And this time it's for the stack. And since he's not human, this is where Claude uses the animal template I gave it earlier, Alaska to run. And the stack [music] is drawn very nicely with that classic anime look, vibrant colors, and sketch

[8:05] details. Next up is the main character. So, I'll give Claude a brief description of what I'm going for. So, 17 years old with a dark brown chinlane pop and a little more detail about how her outfit looks, and then I'll ask it to match the exact [music] style it used for the stack. The prompt looks good so far, so let's see how she comes out. Her look has this rough feel to her, as if she's a forester going through the woods. And the art style is a mix of Gibli and modern sharp animation all the way through. I really like the little details in her outfit like the creases

[8:36] and the shadows on her jacket and the way the whole thing looks. Now for the video prompt, I'll ask for something the first film didn't have, which is a specific transition. I'm doing it because I don't want Claude to come up with everything itself, and I want to try a few things of my own, too. My idea is the camera passing behind one of the turban towers. So the tower wipes across the frame and the video picks up on the other side. And in the prompt, Claude writes it straight [music] into the second shot. And everything else looks good, too. But there is another thing I

[9:04] added here, though. I spelled out that the whole dialogue has to be in English. And it can switch to any other language like Chinese or anything similar because otherwise Cedance 2.5 has a tendency to drift into Mandarin [music] if you let it. So next, I'll tell it to run the video prompt.

[9:24] You're still here? I thought I really thought you'd have gone with them. They crossed the valley at first light. I heard them calling. I just didn't follow. Why not? Because nobody down there talks to me. Go and catch them up. Go on. I'll be here tomorrow. Same as always. The film starts off by showing it dirt bikes wheel spinning and it follows up with cinematic cuts and the transition [music] I asked for which works really well. I wouldn't say it looks 100% like anime but it's not far off. The stag and the girl are animated very nicely even though in my opinion the environment doesn't really blend in with them. But

[10:08] even with that I think it's a big improvement in what video models can do with 2D. So, just by changing the style and [music] story, I've got a second short film that looks completely different from the first one. And for the third film, I'm moving away from traditional Japanese anime altogether because this time I want that Family Guy style of American animation instead. The best part is that workflow stays the exact same as before. So, just like with the anime, [music] I'll type in that I want a cartoon. After that, I'll briefly explain what happens in the story myself. It's about a janitor who finds an armadillo asleep in the bull rack of

[10:41] an empty bowling alley and all [music] it wants is to be bowled down the lane. He says no at first, but then he can't resist and gives it a go and it takes [music] out all 10 pins. The one thing I make really clear is that the armadillo should not speak. Only the janitor talks and he talks to it like it understands everything he's saying. Then I'll describe the armadillo, which is small and round with a bandit shell, a long pink snout, [music] and big black eyes, and ask for its sheet first. Claude builds it on the animal template and saves no talking rule for the video

[11:12] since a single image doesn't need it. And you'll notice that one word, cartoon, has turned into a full style description in the prompt with bold outlines, flat, bright colors, and characters that squash and [music] stretch when they move. It all looks good, so I'll tell it to run. Next up is the janitor who's short and bold with a big gray walrus mustache and work coveralls. And just like the last two films, he gets built off the finished armadillo sheet. So it feels like both of them have the same art style and belong in the same video. That Family Guy look comes through with a modern touch to it. And I really like the

[11:44] colors, especially the teal on the janitor's uniform. Now in the video [music] prompt that no talking crew gets its own block right near the top, so it covers the whole film instead of just one shot. Instead of leaving it as a vague line, Clites out exactly how the armadillo gets [music] its point across, like pointing its snout, nodding, curling into a ball, [music] and dropping when it's disappointed. That way, Cedance has something for it to do whenever it's on screen without a single word of dialogue. Then I'll simply run

[12:11] the whole thing. Out. Come on. Out of there. You're in the ball rack again, and this is the last lane I have left to clean. No, absolutely not. Do you have any idea what they say to me if they find shell marks on the wood? One. One roll and then you go home. That is the best score anyone has put up in here all week. Don't let it go to

[12:42] your head. And the video is impressive with smooth animations and a fun story all the way through. It does get a little odd right at the end, but I still like what it gave me. So, from one single chat, I've got three films that look nothing alike, and I haven't even written a style prompt for any of them myself. And that leaves us with the last one, which is the hardest style for a video model to get right. It's a noir cartoon, which is all about strong contrast with deep black shadows right up against bright highlights. The problem is that AI models tend to soften that look until it

[13:13] feels washed out because most of what they learn from is photos with gradual shading everywhere. And the story is about a magician hiding his rabbit in a top hat on a night drain from a stranger who keeps walking past your cabin but never spots it. The character sheets look like high quality illustrations made by real people. And I really like the angular art style with the sharp contrast of black and white with no other colors. Next, [music] I'll speed through getting the video since it's the

[13:39] same process as before.

[13:46] That's him again. Same guy. Third time now. Okay, just put me under the hat. You told me you were done with the hat. Yeah, well, I'd rather be in it than out here when he opens that door.

[14:09] Has he gone? He's gone. The characters stay sharp and almost pure black and white all the way through. And there are no vibrant colors popping out anywhere on screen, which is exactly what we want. And the whole video has that New Year look I was going for, even with some gray in the backgrounds. Honestly, I think it could be better. But considering it's a short film from a single prompt and about something a models usually struggle at, it's really good. So, whatever style you want, try [music] next. The steps stay exactly the same, and the only thing that changes is the [music] story and the look you ask for. And honestly, the

[14:41] reason I could make all of this just by typing out simple prompts is that Claude did the actual work, and Hicksfield [music] gives you access to every model through a single MCP connector. So, go and sign up to Hicksfield using the link in the description and start making your own short films today. Thanks for watching and I'll see you in the next

---

About this transcript
Read from YouTube's own caption track and laid out by ViewRank AI (https://viewrankai.com).
ViewRank AI finds the videos already beating a creator's own average on Instagram, TikTok and YouTube Shorts, transcribes them from the audio itself in more than 60 languages, and turns what worked into new ideas and scripts.
Free transcript tools, no account needed: https://viewrankai.com/tools
How to read any video this way: https://viewrankai.com/llms.txt