How to make AI videos, step by step
From the first idea to a finished video: what to decide, what to write, and how to put the shots together.
Making an AI video comes down to five decisions: where the picture comes from, which model makes it, what you ask for, the shape and length, and how the clips become a finished video. This guide walks through each one.
1. Start from text or from a picture
There are two ways in:
- Text to video. You describe the shot and the model invents everything in it. It is the fastest way from an idea to a moving picture. See text to video AI.
- Image to video. You upload a picture and it becomes the first frame. Use this whenever a face, product, logo or place must look exactly right. See image to video AI.
If you have no picture but need one, make it first with the AI image generator, then animate it.
2. Choose the model for the shot
Every model has a strength, and the right one depends on what is in the shot:
| If the shot needs | Try |
|---|---|
| Natural movement of people and animals | Kling |
| Cinematic framing, consistent characters | Seedance |
| Photoreal people and spoken dialogue | Veo |
| Expressive faces, fights and dance | Hailuo (MiniMax H3) |
| Anime, 3D or clay styles | PixVerse |
| A cheap draft | Wan or Seedance Fast |
When you are unsure, send the same prompt to two models and keep the better take. The models page compares them all with real examples and prices.
3. Write the prompt like a shot list
A video model makes one continuous shot, so describe one moment: the subject and what it does, the setting, the light, and one camera move.
A red fox trots toward the camera along a frosty forest path, early morning light through the trees, telephoto lens, slow motion.
For image to video, describe only the movement, because the picture already shows what is in the frame. There is more detail in writing prompts that video models understand.
4. Set the shape, length and quality
- Shape: 16:9 for YouTube and websites, 9:16 for TikTok, Reels and Shorts, 1:1 for feeds, 21:9 for a cinematic look.
- Length: most clips are 5 to 10 seconds. Short clips stay more coherent, so only go longer when the shot needs it.
- Quality: draft at 720p, then make the final version at 1080p or 4K.
The price in credits updates as you change these, so you always know the cost before you generate.
5. Add sound
Several models make sound with the picture, and Veo can voice dialogue. You can also add a voiceover, music or sound effects afterwards, or lip sync a face to new speech.
6. Turn clips into a finished video
One clip is a shot, not a video. Director turns an idea into a list of shots, renders them, and cuts them together on a timeline where you can trim, add transitions, titles and audio, then export one file. See making a finished video with Director.
Common mistakes
- Too much in one prompt. Four events in one sentence give a muddled clip. Split them into shots.
- Wrong shape. Generating 16:9 and cropping to vertical wastes most of the frame. Choose 9:16 from the start.
- Fixing everything at once. If a result is close, change one thing (the wording, the model or the length) and try again.