How to make an AI video

The most common mistake with AI video is opening a tool, typing one sentence and hoping for the best. The result might look nice, but it has no story, no continuity and no message. A good AI video is made like any other video: planning first, then shooting (in our case, generating), and editing at the end.

I work with Kolbo AI, a platform that brings different image and video models together in one place. The advantage: no separate subscription for every tool, and you can try the same prompt on several models and pick the best result.

The workflow: 5 steps

1. Script: the message, who the video is for, and what the viewer should do at the end.
2. Scene breakdown: a list of short shots, 3 to 8 seconds each.
3. Images: generate an image for each scene first.
4. Video: animate each image into a short clip.
5. Edit: join the clips, add voice, music and captions.

Step 1: a short, sharp script

Before you touch any tool, write the script. For Reels or TikTok, 20 to 40 seconds is enough. The first second matters most: it has to stop the scroll.

• Hook (opening line): a question, a promise, or something surprising.
• Body: 2 to 3 points, no more.
• Ending: a clear call to action, like "send me a message".

Step 2: why images before video?

You can ask a model for video straight from text (Text to Video), but then you have no control over how the character and the place look, and every scene comes out different. The professional way is to generate an image for each scene first (Text to Image), pick the best one, and only then animate it (Image to Video).

• Control: you see exactly how the scene will look before investing in video.
• Continuity: the same character, the same clothes and the same style in every scene.
• Savings: generating an image is far cheaper and faster than generating video.

Step 3: choosing a model

In Kolbo you'll find several models, each strong at something different. Models are updated all the time, so instead of memorizing names, remember how to choose:

• Image models: some excel at realism (people, products), others at design, illustration and text inside the image.
• Video models: some excel at human movement and faces, others at camera moves and landscapes.
• Price and length: advanced models cost more credits. Test on a cheaper model first, and render only the final version on the expensive one.

Step 4: how to write a good prompt

A prompt is the instruction you give the model. The more precise it is, the closer the result is to what you imagined. Write in English, since most models understand it best. A good image prompt answers 6 questions:

1. Who or what: the main subject (a person, a product, a place).
2. What's happening: what they are doing.
3. Where: the setting and background.
4. How it's shot: shot type (close-up, wide), angle, lens.
5. Light and mood: golden hour, studio lighting, neon, dramatic.
6. Style and format: photorealistic, cinematic, illustration; and the aspect ratio, 9:16 for Reels.

Close-up portrait of a confident woman in her 30s, business owner, smiling at the camera, standing in a modern bright café she owns, shallow depth of field, 50mm lens, warm golden-hour light through the window, photorealistic, cinematic color grading, vertical 9:16

A video prompt only describes the movement, because the image already exists: what moves, how the camera moves, and at what pace. One action per clip.

She turns her head slightly toward the camera and smiles warmly. Slow camera push-in. Steam rises gently from the coffee cup on the counter. Natural, subtle movement, cinematic, 5 seconds.

• One action per clip. Three actions in a 5-second clip look strange.
• Camera moves that work well: slow push-in, pan, tracking shot, drone shot.
• Didn't come out right? Change one thing in the prompt and run it again, so you know what helped.
• Save every prompt that worked in one file. Within a month you'll have your own prompt library.

Step 5: editing and finishing

Now join all the clips in an editing app such as CapCut, in the order of the script.

• Voice: your own recorded voice-over, or an AI-generated voice.
• Music: quiet under the voice, with a beat that fits the cuts.
• Captions: a must. Most people watch social video with the sound off.
• Fast cuts: something changes on screen every 2 to 4 seconds.
• Format: 9:16 for Reels, TikTok and Shorts.

Checklist before you publish

• Does the first second stop the scroll?
• Do the character and style look the same in every scene?
• No six-fingered hands, garbled text or distorted faces?
• Are there captions and a clear call to action at the end?

Want to learn this in depth, on your own videos and business? Book a private Zoom lesson with me, or just message me on WhatsApp.