FREE Prompts that Create Viral AI Shorts (for Social Media)

0:00 There are dozens of AI video trends going viral on the internet right [music] now, and you can recreate any one of them with just some simple steps. But that is where most people get stuck. They try [music] to copy the trend, but they can never get their output quite right. So, it just comes out looking

0:12 like a cheap copy. But in this video, I'll break down all seven types of these videos, and I'll give you the exact prompts and workflow so that you can create these for yourself. So, let's start with the style you've definitely seen all over your feed, like these Italian brain rot videos with the meme

0:26 characters running around and causing chaos. Now, the reason most people's version of this style comes out looking bad is that they just describe the character inside the video prompt and hope it stays consistent, but it never really does. So, the character ends up looking a little different in every

0:40 shot, and by the end, it looks like a completely different thing. So, before I animate anything, I'll lock the character down first with something called a character sheet. It's basically one single image that shows the exact same character from multiple sides. And what that does is give the video

0:54 generation model every angle of our character up front. [music] So no matter what happens in the scene, it stays consistent throughout the entire video. Now, for all of this, I'm using Open Art. It's a platform that has all the best AI models in one place, and I'll leave a link in the description if you

1:07 want to follow along. And every prompt I use in this video is free in the description, so you can just copy them straight from there. So, first I'll head over to create in the sidebar, choose image, and set the model to Nano Banana Pro. Then I'll drop a single reference image of our first character into the

1:22 add visual references box. If you're wondering where I got these character photos, you can easily find them on Google. [music] After that, I'll just paste the prompt in and generate. I've now got a character sheet of our video's character, and it has three different angles: front, side, [music] and the

1:35 back part. Then, I'll do the exact same thing for our second character. The only thing that I will change is the reference image I uploaded first, and it came out looking pretty good, too, just like the previous one. Now that both of our characters are locked in, it's time to actually make the video. And with

1:48 these types of videos, prompts are the biggest reason for an AI video to come out, good or bad. So, I want to show you exactly how I build my prompt. I'll head over to video prompt.studio. I'll type in my prompt for the scene and set the target model to sea dance. Then, I'll pick the cinematic category since this

2:03 is a multi-shot narrative. Now, there's a toggle here for Jason or normal, and I always keep it on Jason for generating videos. Here's why that matters. If you give an AI model a long paragraph of prompt, [music] it easily misreads your instructions and scramles the style. But converting that prompt to JSON breaks

2:18 your scene down into clean, separate fields, which is much easier for the AI to understand. [music] Since we are showcasing seven completely different video styles in this video, we need the AI to switch up the rules for every single section. Jason lets the model follow every single prompt perfectly,

2:33 keeping the pacing and visuals perfectly consistent for [music] each individual style we build. So, I'll hit generate and it hands me the full structured JSON prompt. Now, I'll take that into Open Art. I'll jump over to video and switch it to text with reference. Then, I'll set Cance 2.0 O as the model and add

2:47 both of our character sheets in as the references. [music] And I'll leave the audio on for the noise and sound effects. Then I'll set the aspect ratio to 9x6 and the resolution to 1080p. And I'll push the duration up to 15 seconds. [music] Lastly, I'll paste the JSON prompt straight in the box and hit generate.

3:03 [cheering and screaming] As you can see, our video does have that brain rot vibe, but in the scene where our male character hits the stage with his bat, it kind of doesn't make sense. The bars are way up high and he's standing on the ground, so the bat isn't

3:28 really supposed to reach that far. These are just the little things you might have to fix, but honestly, these videos are not supposed to make sense at all, and that is what actually keeps viewers hooked. So, it's up to you what you want to do. As you have seen, because I locked our characters in with the

3:41 character sheet before I ever hit generate, they stayed consistent throughout the whole video. But the next trend is the complete opposite. You don't have to build a character or upload anything at all, which makes it the fastest one on the whole list to pull off. The next trend is also

3:54 something you've definitely watched on the internet. It's actually those oddly satisfying ASMR cutting videos. It's pure texttovideo generation, so you don't have to build a character sheet or upload any reference images. So, as I did before, I'll paste this prompt into video.studio. Studio and build a JSON

4:09 version of it. Then I'll bring it straight into Cedence and paste it into the prompt box. I'll set the aspect ratio to 3x4 and hit generate. Our video turned out incredibly satisfying and I really like the details like how after cutting the glass apple, there's some water left over on the

4:33 table. That's a really nice attention to detail. Now, most of the time when people try to generate videos like this, they can't really get it right. And the reason is the prompt. Firstly, it has to be an extreme close-up and slow motion because that's what makes all the tiny details stand out. Then, it all has to

4:47 happen in one single continuous shot with no cuts because the moment you add cuts in the middle, it just stops being satisfying. And the most important part of the whole video is the sound effects. In the prompt, I explicitly say to use ASMR sounds only, no music and no talking at all. And this is what made

5:03 our results so good. Now, for our next video, I'll follow the exact [music] same steps as before and paste in this prompt. This one turned out pretty satisfying, too. There's this odd little detail where the lava turns into a cream-like texture when it gets scooped out, but honestly, it's negligible. And overall,

5:27 I really like it. And you can make your own versions of these, too. Just take the prompt I used and ask any AI to create variations of it and customize it however you want. But the next trend is in a completely different style. It's a much more cinematic and stylized look and it takes a bit more setup than this

5:42 one did. So the next one is one of those moody animations with the deep [music] voice narrating some eerie story over the top. Now this one actually starts with an image and not a video. So first I'll head into image and set the model to Nano Banana Pro. Now I'll paste in the prompt and hit generate. What I've

5:57 got is a normal image and I'll use this as a starter for our video generation model so it has a better direction when it generates the video. One little detail you need to notice in the prompt [music] is that our character needs to be more distinguishable than everything else in the image. You saw that in the

6:11 one I just generated and the contrast it created is the look we're going for. Now to animate it, I'll take that image over to video.studio and build [music] the JSON prompt. Then I'll drop it into Cance and I'll upload our image as the starting frame. [music] Next, I'll set the aspect ratio to 16 by9 and push the

6:26 duration up to 12 seconds. I'll leave the audio on for the narrator, then hit generate. >> That night, George decided to walk alone through the woods, but nothing could have prepared him for what was waiting for him in the dark. The second prompt gave us a very clean animation of our image. The camera moves

6:46 from the character out over the forest in a drone-like shot. And the overall atmosphere is really giving eerie vibes, and it's very good. And because nothing but the camera glides through the video, it gave our animation a very cinematic look. You might be familiar with this style from a famous YouTube channel

7:01 called Fern. They use these animations to help you visualize the narrative they're trying to tell. And this is honestly a very easy way to generate animations like these. [music] Fern uses actual 3D animators and probably a ton more money to get results like these. But the next one is another different

7:15 but fun animation to make. So this next one is one of those videos where you cut into a completely ordinary object and it turns out to be something totally different on the inside. First I'll head into the image tab and set the model to Nano Banana Pro. Then I'll paste in the prompt and hit generate. It gave me a

7:30 completely ordinary looking brick. Now when you're creating these types of videos, you have to make sure the object looks photorealistic. There shouldn't be any hints that it could be something else. And because our entire angle depends on the surprise, it's a really important step. Now to animate it, I'll

7:43 take my prompt over to videooprompt.studio and build the JSON prompt. Then I'll head into the video tab and set the model to seed dance. After that, I'll upload the image as the starting frame and paste in that JSON prompt. Next, I'll set the aspect ratio to 1x one and keep the duration at 8

7:58 seconds. I'll leave the audio on for the cutting sounds, then hit generate. As you can see, we've got a similarly satisfying video of the brick being cut. But if you notice, when the knife cuts through it, it doesn't make the sound a brick's supposed to make. It sounds more like a cake.

8:22 So, it might give the surprise away a little early, but honestly, it's completely fine. And that's all it takes to create these surprise videos. But the next trend is another weird and funny one. So, let's get into it. And this one is also similar to the last one, but the surprise is a little different. Just

8:36 like the first one, I'll head over to the image tab and set the model to Nano Banana Pro. And I'll paste in this prompt and hit generate. As you can see, I've got a very normal looking bowling ball. So when it does the complete opposite of what it's supposed to, that contrast makes it really satisfying.

8:50 Next, to animate it, I'll upload it to sea dance as the starting frame. Then I'll paste in this JSON prompt, which I converted like I showed you before. But before I generate, I'll set the aspect ratio to 3x4 and the duration to 8 seconds. And I'll keep the audio on. Then I'll hit generate.

9:12 In the video, our solid bowling ball is being squished like a rubber ball. It's really unexpected for someone watching this. And the way I achieved this is by adding one line in the prompt. I literally told the model that even though the ball looks completely rigid, it should squish and [music] compress

9:27 like soft memory foam. And that's the whole trick. But the next trend is a little different, and it's a lot more practical than the ones I've been showing you. And this is something you can actually use on a real product. I'll make the headphones in the image tab using Nano Banana Pro with this prompt.

9:40 It gave me a pretty good-look one. Next, I will upload it as the starting frame as usual and paste in my prompt. And as you can see, I've got a very minimal floating animation of our headphones. and it looks really clean and definitely worthy of using on an actual ad. And honestly, you can use this on pretty

9:55 much any product you'd like. Now, the next trend is really different from what we've been creating so far, and it's easily one of the most fun ones to make. And this is where someone stops a random person and asks them a question, all in a hyperrealistic style. Unlike some of the earlier ones, this is pure text to

10:09 video, so there's no need to put in any image reference at all. So, I'll paste my prompt into Sea Dance and set the aspect ratio to 9x6. I'll bump the duration up to 15 seconds, too, since it's going to be a long conversation. Then I'll hit generate. >> Quick question. What's something you've

10:23 never told anyone? >> Uh, I still water my ex's plants. I have a key I never gave back. >> Wait, does she know? >> No, I go in on Tuesdays. I've also rearranged her furniture twice slightly, just to see if she notices. >> That's okay. Anything else? >> I made a spare key for a new place, too.

10:33 I haven't used >> I think we're done here. >> Tell her I said hi. >> What we've got is a hyperrealistic AI interview with one person asking another a question. And the whole conversation is really weird. And that weirdness is exactly what keeps the viewer hooked through the whole video. So, the

10:48 weirdness is actually a good thing here, but there's a little detail you need to keep in mind. Instead of just describing a street interview and hoping the AI comes up with something good, write out the entire conversation yourself, like I did in my prompt. And you've got to do it line by line. Timestamp every

11:02 question and every answer and add a quick note on what happens at each one. And that's what lets me control the whole dynamic of the video. And you can use this in all kinds of creative ways. From generating videos just for fun to making marketing videos for your product or company. And whatever you want to

11:16 make, it can all be done inside one platform. The platform is Open Art. And if you want access to some of the best AI image and video models, now is a great time to upgrade. Move to advanced or above before July 31st and get up to 40% off participating models for the rest of 2026. So sign up to Open Art

11:32 with my link in the description and make your first trend today. Thanks for watching and I'll see you in the next