I Made AI Videos That Got 6M Views & Started a Trend — Here's the Exact Workflow

0:00These two videos got over 6 million

0:02views and started an entire trend that

0:05flooded my feed with people recreating

0:07them. And today, I'm showing you the

0:08exact workflow I used to make both of

0:11them. Hey, I'm Karima and welcome back

0:13to my creative space. By the end of this

0:15video, you'll have the exact setup to

0:16generate cinematic videos and turn your

0:19own ideas into viral trends. But real

0:21quick, before we jump into the tutorial,

0:23I'm not just going to hand you my

0:25prompts so you can copy-paste [music]

0:26them. The whole point of this channel is

0:28to show you how I actually think,

0:30experiment, and build my videos that

0:32bring me millions of views. So, you can

0:34leave this video feeling confident

0:36enough to make your own art. Okay, step

0:39one. Before we generate anything, we

0:41have to talk about character sheet. If

0:42you've ever tried making an AI video,

0:44you've probably noticed that the

0:46person's face suddenly changes, their

0:48outfit morphs, or they look like a

0:50completely different human by the end of

0:52the clip. It is so much easier than what

0:54it looks like.

0:55>> [music]

0:55>> Here's the workflow for both of my

0:57videos. For a character sheet, first, I

0:59need some inspirations. So, I open up

1:02Pinterest and search for an aesthetic or

1:05outfit I want to wear.

1:06>> [music]

1:06>> In this case, a woman in a night armor.

1:09I save a few pictures that I like.

1:12[music] Next, I just grab my phone and

1:13take a few clear selfies of myself from

1:15different angles. Make sure you have

1:17good lighting so your face features are

1:19clear. Now, we go to Claude or any AI

1:22chatbot. Dropping my selfies and the

1:24Pinterest outfit references, [music] and

1:26this is what I wrote. A tip for you

1:28guys, my base prompt

1:30a character sheet is actually the same

1:32across all of my videos. The only part I

1:35change is the specific outfit I wear.

1:37Okay, Claude replied with this prompt.

1:39Take a look. It described what every

1:41part of my body should look like, the

1:43outfit in detail, and even mentioned the

1:46positive constraints. Positive

1:47constraints in AI [music] prompts are

1:49basically rules that explicitly tell the

1:51AI what to include, do, or follow to

1:54shape a precise output. I used Text to

1:56Video AI for both my videos, by the way.

1:59Head over to the images section. There

2:02are a few image generator models here,

2:04but my absolute [music] go-tos are Nano

2:07Banana Pro or SeaArt 5.0 Pro. Here,

2:10upload your selfies, reference photos,

2:13paste the prompt or write into the

2:14prompt box, specify the aspect ratio as

2:1716:9 because we need a horizontal image,

2:21and hit generate. Don't settle for the

2:23first generation. Batch a few

2:25variations, look through them, and pick

2:27the character sheet that looks the most

2:29like you and captures the vibe you want

2:32best. Oh, and a quick bonus tip before

2:33we move on to the next step. [music] If

2:35your video has a key object your

2:37character holds or interacts with, like

2:39the helmet or the sword I dropped in my

2:41first reel, you have to do the same

2:44thing here, too.

2:45So, I took my character sheet and sent

2:47it back to Claude and asked for a neat,

2:50refined sword and helmet suited to the

2:54girl and her costume.

2:56And it replied with this prompt. By the

2:58way, all of my prompts can be found in

3:00the Notion guide attached in the

3:02description below. Okay, now that we

3:04have our character sheets, we can move

3:05on to the next part. Step [music] two is

3:08creating the locations, and I've got

3:09good news. This step is so much faster

3:12and simpler than creating the character

3:13sheet. You can go to Pinterest and grab

3:15reference photos just like we did for

3:17the outfits, but I sometimes just

3:18describe the idea in my head directly to

3:21Claude and see what it makes. For my

3:22first video, I knew I wanted a cinematic

3:25[music] field. You know those vibrant

3:27red poppy flowers with the black centers

3:29in the middle. So, I just told Claude

3:31the vibe. I want

3:34a location, early misty morning, no

3:38orange tint, poppy field with a foggy

3:41[music] background, trees and hills

3:45fading into mist. The background should

3:47feel deep and layered. And it generated

3:52this prompt. Drop that into Hicks field,

3:54choose 16 by 9 as your aspect ratio, and

3:57[music] hit generate. And boom, you have

4:00your first background environment locked

4:03in. For the second video, I wanted

4:04something more [music] dramatic. I

4:06pictured a giant ancient edge of a

4:09massive cliff overlooking a breathtaking

4:11view. Same exact process. I described it

4:14to Claude and got the prompt. Generated

4:16in Hicks field and here's what we got.

4:18You can experiment with creating your

4:20own locations. Watch till the end to see

4:22what other versions I have created with

4:24this video. Before we jump into the

4:25prompting part, let me clarify why

4:27generating with AI becomes so easy at

4:29this point. It's because our assets are

4:32ready. Think of it like cooking a meal.

4:35You have your ingredients ready first,

4:36right? Our ingredients are our assets

4:39[music] that we use to cook our video.

4:41In our case, we already have our

4:42character sheet, our location, and our

4:44helmet and sword prop sheet ready to go.

4:46So, how do you [music] actually do the

4:48prompting part? I open up Claude and

4:50drop in my assets and [music] first ask

4:52it to title them for me in my upcoming

4:54prompt. And here's the thing, I don't

4:56just describe [music] the video. I

4:58explain it like it's an actual story.

5:01For my first reel, I told Claude [music]

5:02that my knight isn't just walking. I

5:05said she just returned from a very

5:07difficult battle. She's exhausted, she

5:09feels [music] completely alone, and she

5:12walks like she just wants to lay down

5:14and take a rest. Now, here's a massive

5:16project for you. [music] Look at your AI

5:18generations from the standpoint of the

5:20cameraman. Imagine you're physically

5:22shooting the scene with your camera in

5:24your hands. Don't just write [music] two

5:26basic sentences. Now, here's how I

5:28directed the camera for this shot. I

5:30said, "Start low on her feet, so you see

5:32her heavy walking motion and see her

5:34drop the heavy helmet on the ground.

5:36[music] Then, slowly rise on her body as

5:39she trudges through the field. After

5:40that, pan around to a side angle, so she

5:43falls back onto the grass, so the shot

5:45ends up looking at her from the side.

5:47Claude took all of that and replied with

5:49a crazy detailed prompt. It broke down

5:52the scene frame by frame, the exact

5:55action timing, [music]

5:56and look at this, it even gave me a

5:58physics box. This physics box is so

6:01important because it tells AI how

6:03objects should behave so they look

6:05realistic. It specified that the sword

6:07should feel long and heavy. Even when

6:09[music] she falls, she doesn't drop like

6:10a rigid block. She transfers her weight

6:13the way she would in real life. [music]

6:15Claude even added sound effects cues

6:16into the prompt automatically. In

6:18Hicksville, there's a sound effect

6:20toggle that generates audio right inside

6:22the model without you having to do it

6:24later. Okay, now that we have our assets

6:27and prompt ready, we can finally start

6:29generating the video. Now we head over

6:31to Hicksville AI, click on video, upload

6:34your media asset, wait a second for it

6:36to process, and tag your images. Copy

6:39the prompt Claude wrote, paste [music]

6:40it in, and turn on the audio toggle

6:43right here for those sound effects. For

6:46the model, choose Seeden 2.5. [music]

6:48This is literally the most insane model

6:51out here right now for video generation.

6:53[music] It handles multi-asset tags

6:54easily, understands complex camera

6:57movements without messing up your face,

6:59and captures [music] emotion really

7:01well. It generates at 720p right now

7:04with higher quality [music] updates

7:06coming soon. Set your video duration,

7:08pick your aspect ratio, and hit

7:10generate.

7:12Now while we wait for it to finish

7:13generating, [music] here are two huge

7:15pro tips to keep you from burning

7:17through your money [music] and credits.

7:19Pro tip number one, don't batch early.

7:22Batching just means running multiple

7:23generations at once. On your first

7:25prompt, run one generation first. Check

7:28if the camera physics feel right before

7:30you batch a bunch of variations. And pro

7:32tip number two, test in lower quality

7:34first. If you're testing a brand new

7:36prompt, run your test draft at 480p or

7:39lower resolution first. That way if you

7:41need to change something in your prompt,

7:42[music]

7:43you end up saving your credits and

7:44money. Okay, our test generation

7:46finished.

7:47>> [music]

7:47>> Take a look. It looks super cool, but I

7:50don't like the fact that it created this

7:52huge dirty path wide in the middle of

7:55the field.

7:56>> [music]

7:56>> It kind of ruins my shot. This is

7:58exactly why we test. So, I go back to

8:00Claude

8:01>> [music]

8:01>> and ask it to edit the prompt. I told,

8:04"Remove the dirt path. [music]

8:07I want a huge clean field of poppy

8:11flowers.

8:12Make her face look

8:14more lively and [music] add a light wind

8:18blowing her

8:20hair and her cloak." Okay, why wind is

8:23key? If you have a character wearing a

8:25cloak outdoors and it's completely

8:27[music] still, it instantly looks fake.

8:30Adding a little breeze to your hair and

8:32your clothes makes the AI generation

8:34look [music] 10 times more realistic.

8:37Claude gave me the updated prompt, so I

8:39paste it back to Hicksfield. Look at

8:41this. The dirty path is gone. Her face

8:43looks more lively. The cloak sways

8:46naturally in the wind and the heavy

8:48movement is so smooth. This is basically

8:51how you work with AI. It's a two-app

8:54workflow. You write and edit your

8:56prompts [music] in an AI chatbot like

8:57Claude and then you generate them in

9:00Hicksfield. That's literally how I

9:02generate all of my videos. And this is

9:04the exact workflow I used for my second

9:06viral video, the one [music] sitting at

9:08the edge of a cliff with a book. So, I

9:09prep my asset sheets, describe my story

9:12and camera angles to Claude. So, laying

9:16down by the cliff with the book over my

9:18face, I sit up, I take the book off and

9:21I look out over the view and I generate

9:24all of this [music] in

9:27so well on the very first try that I

9:30didn't even need to make any edits. By

9:32the way, here are some other versions I

9:34have created for the sake of

9:35experimenting.

9:36>> [music]

9:36>> Feel free to build with your own idea,

9:38too. And now we have the workflow of

9:40both generations. Over 6 million views

9:43and thousands of recreations and an

9:45entire social media trend

9:47>> [music]

9:47>> made using simple assets, conversation

9:50with Claude and Seed on 2.5. All of my

9:52exact prompts, assets, and the Notion

9:54guide are attached in the description

9:56below, so you can try this out yourself.

9:58If there's another viral video on my

10:00Instagram you want me to break down

10:01step-by-step like this, [music]

10:04let me know in the comments. Drop a like

10:06if this was helpful. Subscribe for more

10:08real breakdowns, and I'll see you in my

10:10next one. Bye.