You want a twenty-second video. To announce an event, showcase your shop, wish someone a happy birthday. Two years ago, you needed a camera, time and a bit of know-how. Today, a few sentences are enough.
In theory. In practice, the first attempts are often disappointing: a face that changes from one shot to the next, a hand that warps, movement that goes off in all directions.
The good news is that most of these failures can be avoided with a simple method. Here it is.
1. Think in shots, not in films
Today's tools produce short sequences: a few seconds, up to around thirty for the most advanced, as we saw with Wan 3.0. So there is no point asking for "a promotional film for my bakery".
Instead, break your idea down into three to five shots. The shopfront at daybreak. Hands shaping the bread. A customer smiling as they leave. Each shot will be a separate request.
2. Write each shot like a director
For each shot, state four things: who, where, what is happening, and how it is filmed. "Close-up of hands kneading dough, warm morning light, the camera moves in slowly."
The more concrete it is, the better the result. Vague descriptions produce vague images.
3. Start from an image
This is the trick that changes everything. Instead of asking directly for a video, start by creating, or choosing, a still image of your character or your setting. Then ask the tool to animate it.
That way your character keeps the same face from one shot to the next, instead of transforming each time.
4. Make several attempts, keep the best
Nobody gets the right shot first time. Generate two or three versions of each sequence, and keep the one that works. If the tool allows it, do your tests at reduced quality, and only switch to high quality once you have found the right movement: that is where you save the most.
5. Edit, and take care of the sound
Assemble your shots in editing software; free options exist for both computer and phone. Add music you have the rights to, some text, a voice. Sound often accounts for half the effect, and it is the part everyone forgets.
Using a real person's face without their consent. It is tempting, to make a friend laugh or imitate a celebrity, but it is prohibited in many cases, and it is precisely what fuels deepfakes. Reputable platforms increasingly refuse it. Stick to invented characters, or to people who have said yes.
Which tool should you choose?
There are two main families.
Model makers: Kling, Seedance at ByteDance, Wan at Alibaba, or Google's models. Each has its strengths, and they all evolve every month.
The platforms that bring them together, such as Higgsfield. They let you compare several models in one place, with simple camera controls. That is handy when you are starting out. Just read carefully what your plan actually covers, particularly so-called "unlimited" offers and when your credits expire: our investigation into Higgsfield explains why. If you want to try it: discover Higgsfield.
Affiliate link: if you subscribe via this link, Alyriaz may receive a commission, at no extra cost to you.
Before you publish
If your video features realistic people or situations, state that it was created with AI. It is honest towards those who watch it, and it is increasingly a legal requirement: Europe now requires artificial content to be labelled.
As for the rest, enjoy yourself. Your first video will not be perfect. The third will already be much better.